CN116467472B

CN116467472B - Content recommendation method, training method and device of content recommendation model

Info

Publication number: CN116467472B
Application number: CN202210046489.0A
Authority: CN
Inventors: 田世奇
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2022-01-12
Filing date: 2022-01-12
Publication date: 2025-11-07
Anticipated expiration: 2042-01-12
Also published as: CN116467472A

Abstract

This disclosure relates to a content recommendation method, a training method for a content recommendation model, and an apparatus. The method includes: acquiring candidate content associated with triggering content, and object information of a first type of object that has interacted with the triggering content; obtaining first content feature information of the candidate content and first object feature information of the first type of object in the first display mode using a first interaction prediction model corresponding to a first display mode; determining second interaction prediction information of the object to be recommended to the candidate content in a second display mode based on the content feature information of the triggering content, the first content feature information, and the first object feature information; and determining recommended content from the candidate content based on the second interaction prediction information, and pushing the recommended content to the object to be recommended in the second display mode. This method can ensure the diversity and interest of the content viewing format for users.

Description

Content recommendation method, training method and device of content recommendation model

Technical Field

The present disclosure relates to the field of computer technology, and in particular, to a content recommendation method, apparatus, electronic device, storage medium, and computer program product.

Background

With the development of new media, watching content such as short videos on a platform of the new media is a major way for people to enjoy leisure and entertainment in daily life. Currently, on a new media platform, when a user actively selects a viewing content, some content related to the viewing content is recommended continuously for viewing by the user.

In the current content recommendation method, content recommendation is usually performed in a double-row active selection display mode or in a sliding passive display mode, and the application mode of the recommendation method is single, however, the two display modes have advantages and disadvantages, so how to better combine the two display modes to perform content recommendation to a user is a key for improving the browsing experience of the user.

Disclosure of Invention

The disclosure provides a content recommendation method, a content recommendation device, an electronic device and a storage medium, so as to at least solve the problem of single application mode of the content recommendation method in the related art. The technical scheme of the present disclosure is as follows:

according to a first aspect of an embodiment of the present disclosure, there is provided a content recommendation method, including:

Acquiring candidate content associated with trigger content and object information of a first type of object interacted with the trigger content, wherein the first type of object is an object performing interaction in a first display mode, and the first display mode is a content display mode before triggering the trigger content;

Obtaining first content characteristic information of the candidate content in the first display mode and first object characteristic information of the first type object in the first display mode through a first interaction prediction model corresponding to the first display mode, wherein the first interaction prediction model is used for determining first interaction prediction information of the content to be recommended in the first display mode relative to the first type object;

Determining second interaction prediction information of the object to be recommended on the candidate content in a second display mode based on the content characteristic information, the first content characteristic information and the first object characteristic information of the trigger content, wherein the second display mode is a content display mode which is entered after the trigger content is triggered;

And determining recommended content from the candidate content based on the second interaction prediction information, and pushing the recommended content to the object to be recommended in the second display mode.

In an exemplary embodiment, the determining, based on the content feature information of the trigger content, the first content feature information, and the first object feature information, second interaction prediction information of the candidate content by the object to be recommended in the second display mode includes:

Acquiring at least one of historical interaction information of the object to be recommended for the content in a second display mode, content characteristic information of the candidate content and object characteristic information of a second class object interacted with the candidate content, wherein the second class object is an object performing interaction in the second display mode;

And determining second interaction prediction information of the object to be recommended on the candidate content in a second display mode based on the at least one piece of information, the content characteristic information of the trigger content, the first content characteristic information and the first object characteristic information.

In an exemplary embodiment, the determining, based on the at least one piece of information, and the content feature information of the trigger content, the first content feature information, and the first object feature information, second interaction prediction information of the object to be recommended on the candidate content in the second display mode includes:

when the at least one piece of information comprises the historical interaction information and the content characteristic information of the candidate content, the historical interaction information and the content characteristic information of the candidate content are subjected to characteristic extraction processing through an attention unit, so that extracted characteristic information is obtained;

And obtaining second interaction prediction information of the object to be recommended on the candidate content based on the extracted characteristic information, the content characteristic information of the candidate content, the content characteristic information of the trigger content, the first content characteristic information and the object characteristic information of the first class object.

In an exemplary embodiment, the determining recommended content from the candidate content based on the second interaction prediction information includes:

According to the matching information of the candidate content and the trigger content on a plurality of preset attribute information, carrying out queue division on the candidate content to obtain a plurality of candidate content sequences;

According to the second interaction prediction information, sequencing the candidate contents in each candidate content sequence respectively to obtain a plurality of sequenced candidate content sequences;

And determining recommended content from at least one of the sequenced candidate content sequences according to the priority order of the candidate content sequences.

In an exemplary embodiment, the performing queue division on the candidate content according to the matching information of the candidate content and the trigger content on the preset plurality of attribute information to obtain a plurality of candidate content sequences includes:

the method comprises the steps of obtaining a plurality of preset empty queues, wherein the empty queues are determined based on the priority order of the attribute information and the matching number on the attribute information;

and dividing each candidate content into each empty queue according to the matching information of the candidate content and the triggering content and the priority order of each attribute information to obtain a plurality of candidate content sequences.

In an exemplary embodiment, the determining recommended content from at least one of the ranked candidate content sequences according to the priority order of each of the candidate content sequences includes:

Acquiring the target number of the required recommended content;

and selecting candidate contents corresponding to the target number from at least one of the sequenced candidate content sequences as recommended contents according to the order of the priority of each candidate content sequence from high to low.

And carrying out information prediction processing on the content characteristic information of the trigger content, the first content characteristic information and the first object characteristic information through a second interaction prediction model corresponding to the second display mode to obtain second interaction prediction information of the object to be recommended on the candidate content in the second display mode.

According to a second aspect of the embodiments of the present disclosure, there is provided a training method of a content recommendation model, including:

The method comprises the steps of acquiring sample interaction data, wherein the sample interaction data comprises sample content characteristic information of sample trigger content, first sample content characteristic information of sample candidate content associated with the sample trigger content in a first display mode, second interaction information of a second type sample object on the sample candidate content, and first sample object characteristic information of a first type sample object interacted with the sample trigger content in the first display mode, wherein the first sample content characteristic information and the first sample object characteristic information are predicted by a first interaction prediction model corresponding to the first display mode;

Performing information prediction processing on the sample content characteristic information, the first sample content characteristic information and the first sample object characteristic information through a second interaction prediction model to be trained to obtain second interaction prediction information of a second type sample object on the sample candidate content in a second display mode;

training the second interaction prediction model to be trained based on the second interaction prediction information and the loss value between the second interaction information to obtain a trained second interaction prediction model serving as a content recommendation model;

The first display mode is a content display mode before the sample triggering content is triggered, the second display mode is a content display mode which is entered after the sample triggering content is triggered, the first type of sample object is an object which performs interaction in the first display mode, the second type of sample object is an object which performs interaction in the second display mode, and the first interaction prediction model is used for determining first interaction prediction information of the content to be recommended in the first display mode relative to the first type of object.

According to a third aspect of the embodiments of the present disclosure, there is provided a content recommendation apparatus including:

the system comprises an acquisition unit, a display unit and a display unit, wherein the acquisition unit is configured to acquire candidate content associated with trigger content and object information of a first type of object interacted with the trigger content, the first type of object is an object performing interaction in a first display mode, and the first display mode is a content display mode before the trigger content is triggered;

The determining unit is configured to execute a first interaction prediction model corresponding to the first display mode to obtain first content characteristic information of the candidate content in the first display mode and first object characteristic information of the first type object in the first display mode, wherein the first interaction prediction model is used for determining first interaction prediction information of the content to be recommended in the first display mode relative to the first type object;

The prediction unit is configured to execute second interaction prediction information of the candidate content by the object to be recommended in a second display mode based on the content characteristic information, the first content characteristic information and the first object characteristic information of the trigger content, wherein the second display mode is a content display mode which is entered after the trigger content is triggered;

And the recommending unit is configured to execute the step of determining recommended content from the candidate content based on the second interaction prediction information and pushing the recommended content to the object to be recommended in the second display mode.

In an exemplary embodiment, the prediction unit is further configured to perform obtaining at least one of historical interaction information of the object to be recommended for the content in the second display mode, content feature information of the candidate content, and object feature information of a second class of objects interacted with the candidate content, where the second class of objects are objects performing interaction in the second display mode, and determine second interaction prediction information of the object to be recommended for the candidate content in the second display mode based on the at least one information, the content feature information of the trigger content, the first content feature information, and the first object feature information.

In an exemplary embodiment, the prediction unit is further configured to perform, when the at least one piece of information includes the history interaction information and the content feature information of the candidate content, feature extraction processing is performed on the history interaction information and the content feature information of the candidate content through the attention unit to obtain extracted feature information, and based on the extracted feature information, the content feature information of the candidate content, and the content feature information of the trigger content, the first content feature information, and the object feature information of the first type object, second interaction prediction information of the object to be recommended on the candidate content is obtained.

In an exemplary embodiment, the recommending unit is further configured to perform queue division on the candidate content according to matching information of the candidate content and the trigger content on a plurality of preset attribute information to obtain a plurality of candidate content sequences, respectively sort the candidate content in each candidate content sequence according to the second interaction prediction information to obtain a plurality of sorted candidate content sequences, and determine recommended content from at least one sorted candidate content sequence according to a priority order of each candidate content sequence.

In an exemplary embodiment, the recommending unit is further configured to perform obtaining a plurality of preset empty queues, wherein the empty queues are determined based on the priority order of the attribute information and the matching number on the attribute information, and the candidate content is divided into the empty queues according to the matching information corresponding to the candidate content and the trigger content and the priority order of the attribute information, so as to obtain a plurality of candidate content sequences.

In an exemplary embodiment, the recommending unit is further configured to perform obtaining a target number of required recommended contents, and select, from at least one of the ranked candidate content sequences, candidate contents corresponding to the target number as recommended contents in order of high priority of each of the candidate content sequences.

In an exemplary embodiment, the prediction unit is further configured to execute a second interaction prediction model corresponding to the second presentation mode, and perform information prediction processing on the content feature information of the trigger content, the first content feature information, and the first object feature information, so as to obtain second interaction prediction information of the object to be recommended on the candidate content in the second presentation mode.

According to a fourth aspect of embodiments of the present disclosure, there is provided a training apparatus of a content recommendation model, including:

The system comprises an acquisition unit, a display unit and a display unit, wherein the acquisition unit is configured to execute acquisition of sample interaction data, the sample interaction data comprises sample content characteristic information of sample trigger content, first sample content characteristic information of sample candidate content associated with the sample trigger content in a first display mode, second interaction information of a second type of sample object on the sample candidate content, and first sample object characteristic information of a first type of sample object interacted with the sample trigger content in the first display mode, wherein the first sample content characteristic information and the first sample object characteristic information are predicted by a first interaction prediction model corresponding to the first display mode;

The prediction unit is configured to execute information prediction processing on the sample content characteristic information, the first sample content characteristic information and the first sample object characteristic information through a second interaction prediction model to be trained, so as to obtain second interaction prediction information of a second type sample object on the sample candidate content in a second display mode;

The training unit is configured to perform training on the second interaction prediction model to be trained based on the second interaction prediction information and the loss value between the second interaction information, and obtain a trained second interaction prediction model as a content recommendation model;

According to a fifth aspect of embodiments of the present disclosure, there is provided an electronic device, comprising:

A processor;

A memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the method of any of the above.

According to a sixth aspect of embodiments of the present disclosure, there is provided a computer readable storage medium, which when executed by a processor of an electronic device, causes the electronic device to perform the method as set forth in any one of the preceding claims.

According to a seventh aspect of embodiments of the present disclosure, there is provided a computer program product comprising instructions therein, which when executed by a processor of an electronic device, enable the electronic device to perform the method of any one of the above.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:

After the candidate content related to the trigger content and the object information of the first type of object interacted with the trigger content are obtained, the first content characteristic information of the candidate content in the first display mode and the first object characteristic information of the first type of object in the first display mode are obtained through a first interaction prediction model corresponding to the first display mode, the second interaction prediction information of the object to be recommended to the candidate content in the second display mode is further determined based on the content characteristic information, the first content characteristic information and the first object characteristic information of the trigger content, the recommended content is determined from the candidate content based on the second interaction prediction information, and the recommended content is pushed to the object to be recommended in the second display mode. According to the method, a second display mode is added in a detail page of trigger content selected by an object to be recommended, second interaction prediction information of the object to be recommended to the candidate content in the second display mode is determined based on first content feature information and first object feature information of the candidate content and the first type object in the first display mode, the selection of the recommended content is performed based on the first interaction prediction information, the linkage relation between the trigger content and the second display mode is fully considered, smooth transition between the two display modes is realized, diversity and interestingness of a user watching content form can be ensured, the right of active selection is reserved, passive selection is provided, and certain correlation exists between the two information, so that consumption experience and consumption duration of the user can be improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.

Fig. 1 is a flow chart illustrating a content recommendation method according to an exemplary embodiment.

FIG. 2 is a schematic diagram of a model structure of an interactive predictive model, according to an example embodiment.

FIG. 3 is a flowchart illustrating a recommended content determination step according to an exemplary embodiment.

FIG. 4 is a flow chart illustrating a method of training a content recommendation model, according to an exemplary embodiment.

Fig. 5 is a block diagram illustrating a structure of a content recommendation device according to an exemplary embodiment.

FIG. 6 is a block diagram illustrating a training apparatus of a content recommendation model, according to an exemplary embodiment.

Fig. 7 is a block diagram of an electronic device, according to an example embodiment.

Detailed Description

In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims. It should be further noted that, the user information (including, but not limited to, user equipment information, user personal information, etc.) and the data (including, but not limited to, data for presentation, analyzed data, etc.) related to the present disclosure are information and data authorized by the user or sufficiently authorized by each party.

It should be noted that, in the present disclosure, the first display mode is a content display mode before triggering the triggering content, and the second display mode is a content display mode entered after triggering the triggering content. Specifically, in one embodiment, the first exhibition mode may be a multi-column actively selected exhibition mode, the second exhibition mode may be a downslide passive exhibition mode, for example, the first exhibition mode is a dual-fall stream exhibition mode, the second exhibition mode is a single-fall stream exhibition mode, or the first exhibition mode is a three-column waterfall stream exhibition mode, the second exhibition mode is a single-fall stream exhibition mode, and so on. In another embodiment, the first display mode and the second display mode may be multiple columns of actively selected display modes, but the columns of the first display mode and the second display mode are different. For example, the first demonstration mode is a three-column waterfall flow demonstration mode, the second demonstration mode is a double-waterfall flow demonstration mode, or the first demonstration mode is a double-waterfall flow demonstration mode, and the second demonstration mode is a three-column waterfall flow demonstration mode. In practical applications, those skilled in the art may set the specific forms of the first display mode and the second display mode according to the needs, which is not particularly limited in this disclosure. For convenience of explanation, embodiments of the present disclosure will be described below by taking a double-row active selection display mode as a first display mode and a single-row sliding passive display mode as a second display mode.

In an exemplary embodiment, as shown in fig. 1, a content recommendation method is provided, where the method is applied to a terminal to illustrate the method, it is understood that the method may also be applied to a server, and may also be applied to a system including the terminal and the server, and implemented through interaction between the terminal and the server. The terminal can be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things equipment and portable wearable equipment, and the internet of things equipment can be smart speakers, smart televisions, smart air conditioners, smart vehicle-mounted equipment and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.

In this embodiment, the method includes the steps of:

in step S110, candidate content associated with the trigger content and object information of a first type of object interacted with the trigger content are obtained, wherein the first type of object is an object performing interaction in a first display mode, and the first display mode is a content display mode before triggering the trigger content.

The triggering content can be understood as content that an object to be recommended actively selects to browse, and the triggering content can be in the form of video, text or image.

The object information may be understood as information characterizing the identity, interest, etc. of the object, for example, the object information may be information of age, occupation, interest, etc.

The first display mode may be a double-row active selection display mode.

In a specific implementation, the trigger content can be displayed on a main page in an application program, the content on the main page is displayed in a first display mode, a user can select any one of the contents in the main page to browse and watch through sliding operation, the content selected and triggered on the main page by an object to be recommended is the trigger content, in practical application, a plurality of contents associated with each content on the main page can be preconfigured to serve as candidate contents, when the trigger operation of the object to be recommended for any one of the contents on the main page is received, the triggered contents serve as trigger contents, and the candidate contents associated with the trigger contents are obtained. Meanwhile, the object information of the first type object interacted with the trigger content can be obtained, and specifically, the object information of the first type object performing at least one interaction behavior such as praise, comment, collection, watching and forwarding on the trigger content can be obtained.

For example, in a short video viewing platform, on a video main page displayed in a double-row active selection display mode, a user can select any one video in the video main page to view through sliding operation, the video triggered in the main page is the trigger video, candidate videos associated with each trigger video can be preconfigured so as to facilitate the subsequent selection of recommended videos from the candidate videos, and the recommended videos are recommended to the user, so that after the user selects the trigger video, the terminal can display the recommended videos associated with the trigger video for viewing by the user.

In step S120, first content feature information of candidate content in a first display mode and first object feature information of a first type object in the first display mode are obtained through a first interaction prediction model corresponding to the first display mode, where the first interaction prediction model is used for determining first interaction prediction information of content to be recommended in the first display mode relative to the first type object.

The first interaction prediction model can be used for acquiring characteristics of any content or any object in a first display mode, and obtaining first interaction prediction information of any object and any content based on the characteristics.

In a specific implementation, sample content and a first type of sample object in a first display mode can be obtained in advance, and a first interaction prediction model to be trained is trained through the sample content and the first type of sample object, so that the trained first interaction prediction model is obtained. More specifically, referring to a model under a dash-dot line in fig. 2, the model structure diagram of the first interaction prediction model includes two multi-layer perceptrons, which are respectively used for extracting features of sample content and a first type of sample object in a first display mode to obtain first sample content feature information of the sample content and first sample object feature information of the first type of sample object, and further performing dot multiplication processing on the first sample content feature information and the first sample object feature information to obtain a click rate, and training the click rate to obtain a first interaction prediction model corresponding to the first display mode. The method comprises the steps of obtaining first content characteristic information of candidate content in a first display mode by carrying out characteristic extraction processing on the content characteristic information of the candidate content related to trigger content through the first interaction prediction model, and obtaining first object characteristic information of a first type object in the first display mode by carrying out characteristic extraction processing on the object information of the first type object through the first interaction prediction model.

In step S130, second interaction prediction information of the candidate content by the object to be recommended in a second display mode is determined based on the content feature information, the first content feature information, and the first object feature information of the trigger content, where the second display mode is a content display mode entered after the trigger content is triggered.

The content characteristic information is information representing content category and label, and can be understood as descriptive information of the content.

The second display mode may be a sliding passive display mode.

The second interaction prediction information may be viewing duration, or may be other information representing interaction conditions.

In a specific implementation, after obtaining the content feature information of the trigger content, the first content feature information of the candidate content in the first display mode, and the first object feature information of the first class object in the first display mode, the information prediction processing is performed on the content feature information of the trigger content, the first content feature information, and the first object feature information through a second interaction prediction model corresponding to the second display mode, so as to obtain second interaction prediction information of the candidate content by the object to be recommended in the second display mode. More specifically, feature extraction processing may be performed on the content feature information of the trigger content by the feature extraction unit to obtain extracted feature information of the trigger content, and second interaction prediction information of the object to be recommended on the candidate content in the second display mode may be determined based on the extracted feature information, the first content feature information, and the first object feature information.

In step S140, based on the second interaction prediction information, recommended content is determined from the candidate content, and the recommended content is pushed to the object to be recommended in the second display mode.

In a specific implementation, at least one recommended content can be determined from the candidate contents according to the order of the interest degree represented by the second interaction prediction information from large to small, and the determined at least one recommended content is pushed to the content to be recommended in a second display mode.

For example, taking the second interaction prediction information as the viewing duration, the longer the viewing duration, the higher the interest degree is indicated, so that the candidate contents can be ordered according to the sequence from the long viewing duration to the short viewing duration, the recommended content with the viewing duration of TopN (N is more than or equal to 1) is determined from the ordered candidate contents, and the recommended content is pushed to the object to be recommended.

According to the content recommendation method, after candidate content related to the trigger content and object information of a first type of object interacted with the trigger content are obtained, first content characteristic information of the candidate content in a first display mode and first object characteristic information of the first type of object in the first display mode are obtained through a first interaction prediction model corresponding to the first display mode, second interaction prediction information of an object to be recommended to the candidate content in a second display mode is determined further based on the content characteristic information, the first content characteristic information and the first object characteristic information of the trigger content, recommended content is determined from the candidate content based on the second interaction prediction information, and the recommended content is pushed to the object to be recommended in the second display mode. According to the method, a second display mode is added in a detail page of trigger content selected by an object to be recommended, second interaction prediction information of the object to be recommended to the candidate content in the second display mode is determined based on first content feature information and first object feature information of the candidate content and the first type object in the first display mode, the selection of the recommended content is performed based on the first interaction prediction information, the linkage relation between the trigger content and the second display mode is fully considered, smooth transition between the two display modes is realized, diversity and interestingness of a user watching content form can be ensured, the right of active selection is reserved, passive selection is provided, and certain correlation exists between the two information, so that consumption experience and consumption duration of the user can be improved.

In order to further improve the accuracy of the second interaction prediction information of the object to be recommended to the candidate content in the determined second display mode, the present disclosure further provides another exemplary embodiment, where in this embodiment, the step S130 may be specifically implemented by the following steps:

Step S130A, obtaining at least one of historical interaction information of an object to be recommended for the content in a second display mode, content characteristic information of candidate content and object characteristic information of a second class object interacted with the candidate content;

Step S130B, determining second interaction prediction information of the candidate content by the object to be recommended in the second display mode based on at least one piece of information and the content characteristic information, the first content characteristic information and the first object characteristic information of the trigger content.

In a specific implementation, referring to the upper model in fig. 2, which is a model structure schematic diagram of the second interaction prediction model, as shown in the drawing, when determining the second interaction prediction information of the object to be recommended to the candidate content in the second display mode, the input condition may include at least one of the historical interaction information of the object to be recommended for the content in the second display mode, the content feature information of the candidate content, and the object feature information of the second object interacted with the candidate content, in addition to the content feature information of the trigger content, the first content feature information of the candidate content in the first display mode, and the first object feature information of the first object in the first display mode. And carrying out information prediction processing on the at least one piece of information, the content characteristic information of the trigger content, the first content characteristic information and the first object characteristic information through a second interaction prediction model to obtain second interaction prediction information of the candidate content of the object to be recommended under a second display mode.

In this embodiment, on the basis of the content feature information, the first content feature information and the first object feature information of the trigger content, at least one of the history interaction information of the object to be recommended for the content in the second display mode, the content feature information of the candidate content, and the object feature information of the second object interacted with the candidate content is combined, so as to implement the prediction of the second interaction prediction information from multiple dimensions, and improve the accuracy of the determined second interaction prediction information.

Further, in an exemplary embodiment, the step S103B may be implemented by:

Step S130B-1, when at least one piece of information comprises the history interaction information and the content characteristic information of the candidate content, carrying out characteristic extraction processing on the history interaction information and the content characteristic information of the candidate content through an attention unit to obtain extracted characteristic information;

step S130B-2, based on the extracted characteristic information, the content characteristic information of the candidate content, and the content characteristic information, the first content characteristic information and the object characteristic information of the first class object of the trigger content, obtaining second interaction prediction information of the candidate content by the object to be recommended.

In a specific implementation, referring to the model structure schematic diagram of the second interaction prediction model shown in fig. 2, when the at least one piece of information includes the historical interaction information and the content feature information of the candidate content, the Attention unit (Attention mechanism) may further perform feature extraction processing on the historical interaction information and the content feature information of the candidate content to obtain the extracted feature information, and combine the extracted feature information, the content feature information of the candidate content, the content feature information of the trigger content, the first content feature information and the object feature information of the first object to obtain the second interaction prediction information of the candidate content by the object to be recommended.

More specifically, the attention unit is used for carrying out feature extraction processing on the history interaction information and the content feature information of the candidate content, so that a weight distribution of the history interaction information is learned by the attention unit and is applied to the content feature information of the candidate content, thereby selecting important feature information of the candidate content and inhibiting irrelevant feature information.

In this embodiment, the attention unit performs feature extraction processing on the historical interaction information and the content feature information of the candidate content, so that the interest weight distribution of the object to be recommended can be learned based on the historical interaction information, thereby extracting the content feature information of the candidate content more conforming to the object to be recommended, determining the subsequent second interaction prediction information, and improving the adaptation degree of the determined second interaction prediction information and the object to be recommended.

In an exemplary embodiment, as shown in fig. 3, the step S140 may be specifically implemented by the following steps:

step S310, according to the matching information of the candidate content and the triggering content on the preset attribute information, the candidate content is subjected to queue division to obtain a plurality of candidate content sequences.

Wherein the attribute information represents an expression attribute of the content itself, and the attribute information may include identification information (which may also be referred to as ID information), classification information, tag information, and the like.

The matching information indicates matching results of the candidate content and the trigger content on each attribute information, for example, the matching information may be that the candidate content matches with the identification information of the trigger content, that the classification information does not match, and that the tag information matches.

In a specific implementation, after determining candidate content associated with the trigger content, identification information, classification information and tag information of the candidate content and the trigger content can be respectively obtained, each attribute information of the candidate content and the trigger content is respectively and correspondingly matched to obtain matching information of the candidate content and the trigger content on the attribute information such as the identification information, the classification information and the tag information, and the candidate content is further divided into a plurality of preset empty queues according to the matching information to obtain a plurality of candidate content sequences.

For example, if the identification information of the trigger content is a, the classification information is B, the tag information is C, the identification information of the candidate content is a ^*, the classification information is B ^*, and the tag information is C ^*, the identification information a is matched with the identification information a ^*, the classification information B is matched with the classification information B ^*, the tag information C is matched with the tag information C ^*, and the identification information matching result, the classification information matching result, and the tag information matching result of the candidate content and the trigger content are obtained, and the identification information matching result, the classification information matching result, and the tag information matching result are used as the matching information of the candidate content and the trigger content.

Step S320, sorting the candidate contents in each candidate content sequence according to the second interaction prediction information, so as to obtain a plurality of sorted candidate content sequences.

In a specific implementation, the candidate contents in each candidate content sequence can be respectively sequenced according to the sequence from the big interest degree to the small interest degree of the second interaction prediction information representation, so as to obtain a plurality of sequenced candidate content sequences.

Step S330, according to the priority order of each candidate content sequence, the recommended content is determined from at least one ordered content sequence.

The priority order of each candidate content sequence may be determined based on the priority order of each attribute information and the number of matches on each attribute information, and the higher the number of matched attribute information, the higher the priority of matched attribute information. The priority order of the attribute information can be determined based on the granularity of the expression of the attribute information, and the finer the expression granularity is, the higher the priority is. For example, the priority order of the respective attribute information may be identification information > classification information > tag information.

In a specific implementation, the selection of the recommended content may be performed from the ranked candidate content sequence with the highest priority according to the priority order of each candidate content sequence, and when the candidate content in the ranked candidate content sequence with the highest priority is selected, the selection of the recommended content is performed from the ranked candidate content sequence with the highest priority, and so on. And when the recommended content is selected in each ordered content sequence, selecting the recommended content according to the order of the interest degree represented by the second interaction prediction information from large to small.

In this embodiment, after obtaining candidate contents associated with the trigger content, firstly, according to matching information of the candidate contents and the trigger content on a plurality of preset attribute information, the candidate contents are subjected to queue division to obtain a plurality of candidate content sequences, so that the ranking of the candidate contents on the attribute information dimension is realized, further, according to second interaction prediction information of the candidate contents by the object to be recommended, the candidate contents in each candidate content sequence are respectively ranked, so that the interior of each candidate content sequence is also ordered, and finally, according to the priority order of each candidate content sequence, recommended contents are selected from at least one ranked candidate content sequence, so that the more relevant content to the trigger content is more easily reserved as recommended contents, thereby improving the correlation degree of the recommended contents and the trigger content, and improving the adaptation degree of the recommended contents and the object to be recommended.

In an exemplary embodiment, in step S310, according to matching information of the candidate content and the trigger content on a plurality of preset attribute information, the candidate content is subjected to queue division to obtain a plurality of candidate content sequences, which may be implemented by the following steps:

Step S310A, acquiring a plurality of preset empty queues, wherein the empty queues are determined based on the priority order of each attribute information and the matching number on the attribute information;

Step S310B, dividing each candidate content into each empty queue according to the matching information corresponding to the candidate content and the triggering content and the priority order of each attribute information, and obtaining a plurality of candidate content sequences.

In a specific implementation, since the attribute information has a plurality of pieces, the matching number of the candidate content and the trigger content on the attribute information is also a plurality of pieces, so that a plurality of empty queues can be preset according to the matching number on the attribute information, and under the same matching number, further queue subdivision is performed according to the priority order of each piece of attribute information.

For example, 3 attribute information such as identification information, classification information and label information are arranged, the priority order is that identification information > classification information > label information, three empty queues such as 3, 2 and 1 can be divided according to the matching number on the attribute information, further, when the matching number is 2 and 1, the queues can be further subdivided according to the priority order of the attribute information, for example, when the matching number is 2, three queues can be further subdivided, namely, the identification information and the classification information are matched, the identification information and the label information are matched, and when the matching number is 1, the identification information, the classification information and the label information are matched, and when the matching number is 1, three queues can be further subdivided, namely, the identification information, the classification information and the label information are matched, and accordingly, when the attribute information is three, 1+3+3=7 empty queues can be preset.

And in the obtained preset empty queues, dividing the candidate content into the empty queues according to the matching information corresponding to the candidate content and the triggering content and the priority order of the attribute information to obtain a plurality of candidate content sequences.

For example, if the candidate content and the trigger content are matched on the identification information and the tag information, and the corresponding matching number is 2, the candidate content may be divided into queues where the identification information and the tag information are matched.

In this embodiment, on the one hand, the queues are initially empty according to the matching number on the attribute information, and on the other hand, when the matching number is the same, the queues are further subdivided according to the priority order of the attribute information, so that the queue division on two dimensions of the matching number and the attribute information is realized, and the accuracy of the candidate content division result can be ensured.

In an exemplary embodiment, in the step S330, the determining the recommended content from the at least one sorted candidate content sequence according to the priority order of the candidate content sequences may be implemented by the following steps:

step S330A, obtaining the target number of the required recommended content;

Step S330B, selecting the candidate content corresponding to the target number from at least one of the ordered candidate content sequences as recommended content according to the order of the priority of each candidate content sequence from high to low.

In the specific implementation, when the total number of candidate contents in the candidate content sequence with the highest priority is greater than or equal to the target number, selecting the candidate contents corresponding to the target number from the candidate content sequence with the highest priority according to the sequence from the large interest degree to the small interest degree represented by the second interaction prediction information as recommended contents, and when the total number of candidate contents in the candidate content sequence with the highest priority is less than the target number, sequentially selecting the candidate contents from the candidate content sequence with the highest priority to the candidate content sequence with the lowest priority until the candidate contents meeting the target number are obtained as recommended contents.

It will be appreciated that in another embodiment, the truncated selection may also be performed on each of the ranked candidate content sequences, that is, topM candidate contents are selected from each of the ranked candidate content sequences according to the target number and used as recommended contents. The target number can be equally graded into a plurality of sub-numbers, and the corresponding number of candidate contents are selected from each ordered candidate content sequence to be used as recommended contents. The selection manner of the specific recommended content can be determined according to actual requirements, and the disclosure is not limited in particular.

In this embodiment, when selecting recommended content from each ordered candidate content sequence based on the target number of required recommended content, on one hand, the candidate content sequence to be selected is sequentially determined according to the priority order of each attribute information, and on the other hand, the recommended content is selected according to the second interaction prediction information in the sequence, so that the obtained recommended content can give consideration to the priority of the attribute information and the priority of the second interaction prediction information, and the satisfaction degree of the user is improved.

In an exemplary embodiment, in the step S130, the determining, based on the content feature information of the trigger content, the first content feature information, and the first object feature information, the second interaction prediction information of the candidate content by the object to be recommended in the second display mode may be implemented by performing information prediction processing on the content feature information of the trigger content, the first content feature information, and the first object feature information through a second interaction prediction model corresponding to the second display mode, so as to obtain the second interaction prediction information of the candidate content by the object to be recommended in the second display mode.

In a specific implementation, sample content characteristic information of sample triggering content, first sample content characteristic information of sample candidate content related to the sample triggering content in a first display mode and first sample object characteristic information of a first type sample object interacted with the sample triggering content in the first display mode can be obtained in advance, the information is taken as input, second interaction prediction information is taken as output, a second interaction prediction model to be trained is trained, a second interaction prediction model after training is completed is obtained, and further after obtaining content characteristic information, first content expression information and first object characteristic information of the triggering content, the content characteristic information, the first content expression information and the first object characteristic information of the triggering content can be subjected to information prediction processing through the second interaction prediction model after training is completed, so that second interaction prediction information of the candidate content of the object to be recommended in the second display mode is obtained.

Referring to the schematic model structure diagram of the second interaction prediction model shown in fig. 2, when the second interaction prediction model is used for triggering content feature information, first content feature information and first object feature information of the content, feature extraction processing can be performed on the content feature information of the triggered content through a multi-layer perceptron in the second interaction prediction model to obtain extracted feature information of the triggered content, merging processing is performed on the extracted feature information, the first content feature information and the first object feature information to obtain merged information, and the merged information is input into the multi-layer perceptron to obtain second interaction prediction information of the candidate content of the object to be recommended.

In this embodiment, when the second interaction prediction information of the candidate content by the object to be recommended is predicted by the second interaction prediction model, the linkage relationship between the trigger content and the second display mode is considered, so that the accuracy of the predicted second interaction prediction information of the object to be recommended can be ensured.

In an exemplary embodiment, as shown in fig. 4, a training method of a content recommendation model is provided, and this embodiment is applied to a terminal for illustration by using the method, it is understood that the method may also be applied to a server, and may also be applied to a system including the terminal and the server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the steps of:

In step S410, sample interaction data is acquired, wherein the sample interaction data comprises sample content characteristic information of sample trigger content, first sample content characteristic information of sample candidate content associated with the sample trigger content in a first display mode, second interaction information of a second type sample object on the sample candidate content, and first sample object characteristic information of a first type sample object interacted with the sample trigger content in the first display mode;

In step S420, performing information prediction processing on the sample content feature information, the first sample content feature information and the first sample object feature information through a second interactive prediction model to be trained, so as to obtain second interactive prediction information of the second type sample object on the sample candidate content in a second display mode;

In step S430, training the second interaction prediction model to be trained based on the second interaction prediction information and the loss value between the second interaction prediction information, to obtain a trained second interaction prediction model as a content recommendation model;

The first display mode is a content display mode before the sample triggering content is triggered, the second display mode is a content display mode which is entered after the sample triggering content is triggered, the first type of sample object is an object which performs interaction behavior in the first display mode, the second type of sample object is an object which performs interaction behavior in the second display mode, and the first interaction prediction model is used for determining first interaction prediction information of content to be recommended in the first display mode relative to the first type of object.

In a specific implementation, sample candidate content associated with sample trigger content and object information of a first type of sample object interacted with the sample trigger content can be acquired first. And processing object information of the sample candidate content and the first type object through a first interaction prediction model corresponding to the first display mode to obtain first sample content characteristic information of the sample candidate content in the first display mode and first object characteristic information of the first type sample object in the first display mode. And acquiring second interaction information of a second type of sample object on sample candidate contents and sample content characteristic information of sample trigger contents, and forming sample data by the second interaction information, the sample content characteristic information, the first sample content characteristic information and the first sample object characteristic information.

Referring to the model structure schematic diagram of the second interaction prediction model shown in fig. 2, sample content feature information, first sample content feature information and first sample object feature information in sample data are taken as model input, second interaction prediction information of the sample candidate content by the predicted second type sample object is taken as output, regression training is performed through a loss value between the second interaction prediction information and actual second interaction information in the sample data until the preset training times or the loss value reaches the preset precision, training is ended, and the trained second interaction prediction model is obtained and is taken as a content recommendation model.

In this embodiment, in addition to introducing the sample content feature information of the sample trigger content in the input information of the content recommendation model, the feature information of the sample candidate content and the first type sample object in the first display mode is combined, and the linkage relation between the trigger video selected in the first display mode and the second display mode is fully considered, so that the content recommendation model can capture the implicit correlation relation between the sample candidate content and the sample trigger content, and through such model design, the data migration and the correlation consideration in the relevant service scene are realized, the estimated result of the obtained content recommendation model is more accurate, and the combination of different scenes can be realized.

In an exemplary embodiment, in order to facilitate understanding of the embodiments of the present disclosure by those skilled in the art, a description will be given below of a content recommendation method provided in the present disclosure taking an application scenario of a short video as an example.

At present, the consumption scene of the short video generally has an active consumption form of a double-waterfall stream and a downward-sliding passive consumption of a single-waterfall stream, the active consumption form can provide the user with the option, but the user has high operation complexity and weak immersion, and the passive consumption can provide the user with smooth look and feel and immersion experience. Therefore, how to better combine these two consumption forms is a key to enhance the user's browsing experience.

In order to be compatible with two consumption preferences, the scheme considers that a sliding passive consumption function is added in a video detail page after active selection of a user, but in order to enable smooth transition between the two consumption, the method and the device fully consider the linkage relation between a trigger entry selected by the user and the sliding consumption in technical implementation of a recommendation system, ensure diversity and interestingness of a consumption form of the user, reserve the right of active selection and provide passive selection, and have certain relativity between the two, so that consumption experience and consumption duration of the user are improved.

The embodiment is applied to linkage of the homepage waterfall stream recommendation and the video detail page recommendation stream, and realizes recommendation mechanisms of different streams inside and outside. By modifying the consumption form and constructing the double recommendation system, and in the detail page recommendation funnel, the triggering video actively selected by the user is fully considered, and the recommendation system construction of relevance and diversity is completed. The embodiment comprises the following steps:

(1) And acquiring the candidate video associated with the trigger video, and acquiring the estimated watching time length of the candidate video by the user.

Specifically, the estimated viewing time length can be determined through an interactive prediction model shown in fig. 2, the interactive prediction model is in a double-tower structure, a model at the lower side of a dotted line in the figure can be recorded as a first interactive prediction model, a model at the upper side of the dotted line in the figure can be recorded as a second interactive prediction model, the main objective of the whole model is to estimate the time length distribution (WTD: WATCH TIME distribution) of a user likely to consume on a current candidate video, the calculation method of the objective is to count the overall distribution of different viewing time lengths by using 7 days of data triggering the consumption of a video detail page, the viewing time length is divided into barrels by using an equal-frequency strategy, and the quantile of the distribution of each estimated sample is determined, so that the estimation of the viewing time length is performed to supplement a short plate of a traditional video recommendation method.

It can be seen from the figure that the model further applies the data of the external video stream (i.e. the video stream in the first display mode) to train, the model comprises two parts, the input of the part below the dotted line in the figure is the data of the external video stream, the data is used for training out the video expression information of the trigger video in the external video stream, and the video expression information obtained by training is input into the model at the upper side, so that the internal and external information intercommunication is realized, and more knowledge is provided for the model. The upper part of the dotted line in the figure is used for estimating the estimated watching time of a user on the candidate video, the input of the network structure of the upper part is sample data of an internal video stream (namely, a video stream in a second display mode), in addition to the historical interaction data of the user in and out, the video characteristic information of the candidate video and the user information of the user in and out, video expression information of a trigger video is added, namely, the consumption behavior of the current internal stream is generated by the video of which external stream, so that a model captures the estimated implicit correlation relationship between the candidate video and the trigger video, and based on the user behavior and the video information, an attention model structure is added, a pre-estimated value is output through a fully connected network layer, the watching time distribution of the candidate video is estimated, the pre-estimated value is limited to be 0-1, the model is trained by a regression mode, and a regression loss value (Huber loss) is calculated by a reverse gradient method. Through the model design, data migration and correlation consideration in related service scenes can be realized, so that model estimation is more accurate, and combination of different scenes is realized.

(2) And dividing each candidate video into a plurality of preset empty queues according to the matching condition of the candidate video and the trigger video on the identification information, the classification information and the label information respectively, so as to obtain a plurality of candidate video sequences.

(3) And sequencing each candidate video in the sequence from long to short according to the estimated watching duration of each candidate video in each candidate video sequence to obtain a plurality of sequenced candidate video sequences.

(4) Selecting recommended videos, namely selecting the candidate videos in the ordered candidate video sequences with the highest priority, selecting the candidate videos in the ordered candidate video sequences with the highest priority after all the candidate videos in the ordered candidate video sequences with the highest priority are selected, and finally selecting the candidate videos in the ordered candidate video sequences with the lowest priority. And when selecting among the sorted candidate video sequences, selecting according to the order from long to short of the estimated viewing duration.

(5) And forming the recommended videos into a recommended video sequence according to the selected sequence of the recommended videos, and recommending the recommended videos to the user selecting the trigger video.

Through experiments, the method introduces video watching time length to determine recommended videos, so that the video playing time length of a detail page recommendation system is increased by 1.2%, and the detail page recommendation is greatly improved on related linkage through improving the correlation between the trigger video and the candidate video, so that the similarity score and the high score ratio are increased by 0.3%. By adopting the waterfall stream-added detail page downslide mode, the service time of an application program can be increased by 1++ from the aspect of long-term indexes, the daily active user quantity (DAU) index also has a remarkable forward effect, the mode of internal and external linkage can be described, better consumption, novel browsing modes and various interaction experiences are brought to users, and therefore statistical data are improved.

The video recommending method provided by the embodiment considers the correlation with the current video in the recommending system of the video detail page, and considers the correlation in each funnel stage of each recommendation. First, in the recommended recall phase, a separate trigger video-related video trigger source is provided to ensure the related provisioning of the bottom layer. And secondly, in the fine-ranking order modeling stage, the trigger video features and the current video features are subjected to certain fusion modeling, so that the order model is ensured to capture complete system knowledge. And in the final multi-target sorting stage, the sorting stage is carried out by fully utilizing the correlation, so that the recommended linkage of the inner and outer double flows is realized. On one hand, the method adds the characteristics of the outflow video (obtained by estimating the click rate) into the ordering model of the inner stream and estimates the duration of the inner stream, and on the other hand, in the breaking stage, the matching relationship between the type of the video of the inner stream and the triggering of the outflow click is judged to be broken, and the related consideration is properly weakened in the continuous sliding process of the user, so that the user is prevented from producing fatigue feeling, different experiences are provided, and finally the browsing duration is improved.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

It should be understood that the same/similar parts of the embodiments of the method described above in this specification may be referred to each other, and each embodiment focuses on differences from other embodiments, and references to descriptions of other method embodiments are only needed.

Based on the same inventive concept, the embodiments of the present disclosure also provide a content recommendation apparatus for implementing the above-mentioned related content recommendation method, and a training apparatus for implementing the above-mentioned related content recommendation model training method.

Fig. 5 is a block diagram illustrating a structure of a content recommendation device according to an exemplary embodiment. Referring to fig. 5, the apparatus includes an acquisition unit 510, a determination unit 520, a prediction unit 530, and a recommendation unit 540, wherein:

The acquisition unit 510 is configured to perform acquisition of candidate content associated with the trigger content and object information of a first type of object interacted with the trigger content, wherein the first type of object is an object performing interaction in a first display mode, and the first display mode is a content display mode before triggering of the trigger content;

The determining unit 520 is configured to execute a first interaction prediction model corresponding to the first display mode to obtain first content feature information of the candidate content in the first display mode and first object feature information of the first type object in the first display mode;

the prediction unit 530 is configured to execute second interactive prediction information of the candidate content by the object to be recommended in a second display mode based on the content characteristic information, the first content characteristic information and the first object characteristic information of the trigger content;

and a recommending unit 540 configured to determine recommended content from the candidate content based on the second interaction prediction information, and push the recommended content to the object to be recommended in the second presentation mode.

In an exemplary embodiment, the prediction unit 530 is further configured to perform obtaining at least one of historical interaction information of the object to be recommended for the content in the second display mode, content feature information of the candidate content, and object feature information of a second class of objects interacted with the candidate content, where the second class of objects are objects performing interaction in the second display mode, and determine second interaction prediction information of the object to be recommended for the candidate content in the second display mode based on the at least one information, the content feature information, the first content feature information, and the first object feature information of the trigger content.

In an exemplary embodiment, the prediction unit 530 is further configured to perform, when at least one piece of information includes the historical interaction information and the content feature information of the candidate content, feature extraction processing is performed on the historical interaction information and the content feature information of the candidate content through the attention unit to obtain extracted feature information, and based on the extracted feature information, the content feature information of the candidate content, and the content feature information of the trigger content, the first content feature information, and the object feature information of the first type object, second interaction prediction information of the candidate content by the object to be recommended is obtained.

In an exemplary embodiment, the recommending unit 540 is further configured to perform queue division on the candidate content according to matching information of the candidate content and the trigger content on a plurality of preset attribute information to obtain a plurality of candidate content sequences, respectively sort the candidate content in each candidate content sequence according to the second interaction prediction information to obtain a plurality of sorted candidate content sequences, and determine the recommended content from at least one sorted candidate content sequence according to the priority order of each candidate content sequence.

In an exemplary embodiment, the recommending unit 540 is further configured to perform acquiring a plurality of preset empty queues, wherein the empty queues are determined based on the priority order of each attribute information and the matching number on the attribute information, and each candidate content is divided into each empty queue according to the matching information corresponding to the candidate content and the trigger content and the priority order of each attribute information, so as to obtain a plurality of candidate content sequences.

In an exemplary embodiment, the recommending unit 540 is further configured to perform obtaining the target number of required recommended contents, and select, as recommended contents, candidate contents corresponding to the target number from at least one of the ranked candidate contents sequences in order of higher priority of each candidate contents sequence.

In an exemplary embodiment, the prediction unit 530 is further configured to execute a second interaction prediction model corresponding to the second presentation mode, and perform information prediction processing on the content feature information, the first content feature information, and the first object feature information of the trigger content, so as to obtain second interaction prediction information of the candidate content by the object to be recommended in the second presentation mode.

Fig. 6 is a block diagram illustrating a structure of a content recommendation device according to an exemplary embodiment. Referring to fig. 6, the apparatus includes an acquisition unit 610, a prediction unit 620, and a training unit 630, wherein:

The acquisition unit 610 is configured to perform acquisition of sample interaction data, wherein the sample interaction data comprises sample content characteristic information of sample trigger content, first sample content characteristic information of sample candidate content associated with the sample trigger content in a first display mode, second interaction information of a second type sample object on the sample candidate content, and first sample object characteristic information of a first type sample object interacted with the sample trigger content in the first display mode;

The prediction unit 620 is configured to perform information prediction processing on the sample content feature information, the first sample content feature information and the first sample object feature information through a second interactive prediction model to be trained, so as to obtain second interactive prediction information of the second type sample object on the sample candidate content in the second display mode;

The training unit 630 is configured to perform training on the second interaction prediction model to be trained based on the second interaction prediction information and the loss value between the second interaction prediction information, so as to obtain a trained second interaction prediction model as a content recommendation model;

The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.

The respective modules in the content recommendation apparatus described above may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

Fig. 7 is a block diagram illustrating an electronic device 700 for implementing a content recommendation method, according to an example embodiment. For example, the electronic device 700 may be a mobile phone, computer, digital broadcast terminal, messaging device, game console, tablet device, medical device, exercise device, personal digital assistant, or the like.

Referring to FIG. 7, an electronic device 700 can include one or more of a processing component 702, a memory 704, a power component 706, a multimedia component 708, an audio component 710, an input/output (I/O) interface 712, a sensor component 714, and a communication component 716.

The processing component 702 generally controls overall operation of the electronic device 700, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing component 702 may include one or more processors 720 to execute instructions to perform all or part of the steps of the methods described above. Further, the processing component 702 can include one or more modules that facilitate interaction between the processing component 702 and other components. For example, the processing component 702 may include a multimedia module to facilitate interaction between the multimedia component 708 and the processing component 702.

The memory 704 is configured to store various types of data to support operations at the electronic device 700. Examples of such data include instructions for any application or method operating on the electronic device 700, contact data, phonebook data, messages, pictures, video, and so forth. The memory 704 may be implemented by any type of volatile or non-volatile memory device or combination thereof, such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, optical disk, or graphene memory.

The power supply component 706 provides power to the various components of the electronic device 700. Power supply components 706 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for electronic device 700.

The multimedia component 708 includes a screen between the electronic device 700 and the user that provides an output interface. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensor may sense not only the boundary of a touch or slide action, but also the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 708 includes a front-facing camera and/or a rear-facing camera. When the electronic device 700 is in an operational mode, such as a shooting mode or a video mode, the front camera and/or the rear camera may receive external multimedia data. Each front and rear camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

The audio component 710 is configured to output and/or input audio signals. For example, the audio component 710 includes a Microphone (MIC) configured to receive external audio signals when the electronic device 700 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may be further stored in the memory 704 or transmitted via the communication component 716. In some embodiments, the audio component 710 further includes a speaker for outputting audio signals.

The I/O interface 712 provides an interface between the processing component 702 and peripheral interface modules, which may be a keyboard, click wheel, buttons, etc. These buttons may include, but are not limited to, a home button, a volume button, an activate button, and a lock button.

The sensor assembly 714 includes one or more sensors for providing status assessment of various aspects of the electronic device 700. For example, the sensor assembly 714 may detect an on/off state of the electronic device 700, a relative positioning of the components, such as a display and keypad of the electronic device 700, the sensor assembly 714 may also detect a change in position of the electronic device 700 or a component of the electronic device 700, the presence or absence of a user's contact with the electronic device 700, an orientation or acceleration/deceleration of the device 700, and a change in temperature of the electronic device 700. The sensor assembly 714 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 714 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 714 may also include an acceleration sensor, a gyroscopic sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 716 is configured to facilitate communication between the electronic device 700 and other devices, either wired or wireless. The electronic device 700 may access a wireless network based on a communication standard, such as WiFi, an operator network (e.g., 2G, 3G, 4G, or 5G), or a combination thereof. In one exemplary embodiment, the communication component 716 receives broadcast signals or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 716 further includes a Near Field Communication (NFC) module to facilitate short range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, ultra Wideband (UWB) technology, bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the electronic device 700 may be implemented by one or more Application Specific Integrated Circuits (ASICs), digital Signal Processors (DSPs), digital Signal Processing Devices (DSPDs), programmable Logic Devices (PLDs), field Programmable Gate Arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic elements for executing the methods described above.

In an exemplary embodiment, a computer readable storage medium is also provided, such as memory 704 including instructions executable by processor 720 of electronic device 700 to perform the above-described method. For example, the computer readable storage medium may be ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

In an exemplary embodiment, a computer program product is also provided, comprising instructions executable by the processor 720 of the electronic device 700 to perform the above-described method.

It should be noted that the descriptions of the foregoing apparatus, the electronic device, the computer readable storage medium, the computer program product, and the like according to the method embodiments may further include other implementations, and the specific implementation may refer to the descriptions of the related method embodiments and are not described herein in detail.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any adaptations, uses, or adaptations of the disclosure following the general principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A content recommendation method, characterized in that it includes:

Obtain candidate content associated with the triggered content, as well as object information of a first type of object that has interacted with the triggered content; the first type of object is an object that performs interactive behavior in a first display mode, and the first display mode is the content display mode before the triggered content is triggered;

The first content feature information of the candidate content in the first display mode and the first object feature information of the first type of object in the first display mode are obtained through the first interaction prediction model corresponding to the first display mode; the first interaction prediction model is used to determine the first interaction prediction information of the content to be recommended in the first display mode relative to the first type of object.

Based on the content feature information of the triggered content, the first content feature information, and the first object feature information, the second interaction prediction information of the object to be recommended to the candidate content in the second display mode is determined; the second display mode is the content display mode entered after the triggered content is triggered.

Based on the second interactive prediction information, recommended content is determined from the candidate content, and the recommended content is pushed to the object to be recommended in the second display mode.

2. The method according to claim 1, characterized in that, determining the second interaction prediction information of the object to be recommended to the candidate content in the second display mode based on the content feature information of the triggering content, the first content feature information, and the first object feature information includes:

The system acquires at least one of the following: historical interaction information of the object to be recommended in relation to content in the second display mode; content feature information of the candidate content; and object feature information of a second type of object that has interacted with the candidate content; the second type of object refers to the object that has performed interactive behavior in the second display mode.

Based on the at least one piece of information, as well as the content feature information of the triggering content, the first content feature information, and the first object feature information, the second interaction prediction information of the object to be recommended to the candidate content is determined in the second display mode.

3. The method according to claim 2, characterized in that, determining the second interaction prediction information of the object to be recommended to the candidate content in the second display mode based on the at least one piece of information, the content feature information of the triggering content, the first content feature information, and the first object feature information, includes:

When the at least one piece of information includes the historical interaction information and the content feature information of the candidate content, the attention unit performs feature extraction processing on the historical interaction information and the content feature information of the candidate content to obtain the extracted feature information.

Based on the extracted feature information, the content feature information of the candidate content, the content feature information of the triggering content, the first content feature information, and the object feature information of the first type of object, the second interaction prediction information of the object to be recommended on the candidate content is obtained.

4. The method according to claim 1, wherein determining the recommended content from the candidate content based on the second interactive prediction information includes:

Based on the matching information of the candidate content and the triggering content on multiple preset attribute information, the candidate content is divided into queues to obtain multiple candidate content sequences;

Based on the second interactive prediction information, the candidate content in each candidate content sequence is sorted to obtain multiple sorted candidate content sequences.

Recommended content is determined from at least one sorted candidate content sequence according to the priority order of each candidate content sequence.

5. The method according to claim 4, characterized in that, the step of dividing the candidate content into queues based on the matching information of the candidate content and the triggering content on a preset plurality of attribute information to obtain a plurality of candidate content sequences includes:

Obtain multiple preset empty queues; the empty queues are determined based on the priority order of each attribute information and the number of matches on the attribute information;

Based on the matching information between the candidate content and the trigger content, and the priority order of each attribute information, each candidate content is divided into each empty queue to obtain multiple candidate content sequences.

6. The method according to claim 5, characterized in that, determining recommended content from at least one sorted candidate content sequence according to the priority order of each candidate content sequence includes:

Obtain the target number of recommended content;

Candidate content corresponding to the target number is selected from at least one sorted candidate content sequence in descending order of priority and used as recommended content.

7. The method according to claim 1, wherein determining the second interaction prediction information of the object to be recommended to the candidate content in the second display mode based on the content feature information of the triggering content, the first content feature information, and the first object feature information includes:

By using the second interaction prediction model corresponding to the second display mode, information prediction processing is performed on the content feature information of the triggered content, the first content feature information, and the first object feature information to obtain the second interaction prediction information of the object to be recommended on the candidate content in the second display mode.

8. A method for training a content recommendation model, characterized in that it includes:

Acquire sample data; the sample data includes sample content feature information of sample trigger content, first sample content feature information of sample candidate content associated with the sample trigger content in the first display mode, second interaction information of the second type of sample object with the sample candidate content, and first sample object feature information of the first type of sample object that has interacted with the sample trigger content in the first display mode; the first sample content feature information and the first sample object feature information are predicted by the first interaction prediction model corresponding to the first display mode;

The second interactive prediction model to be trained performs information prediction processing on the sample content feature information, the first sample content feature information and the first sample object feature information to obtain the second interactive prediction information of the second type of sample object on the sample candidate content in the second display mode.

Based on the loss value between the second interaction prediction information and the second interaction information, the second interaction prediction model to be trained is trained to obtain the trained second interaction prediction model, which serves as the content recommendation model.

Wherein, the first display mode is the content display mode before the sample-triggered content is triggered, the second display mode is the content display mode after the sample-triggered content is triggered, the first type of sample object is the object that performs interactive behavior in the first display mode, the second type of sample object is the object that performs interactive behavior in the second display mode, and the first interaction prediction model is used to determine the first interaction prediction information of the content to be recommended in the first display mode relative to the first type of sample object.

9. A content recommendation device, characterized in that it comprises:

The acquisition unit is configured to acquire candidate content associated with the triggering content, as well as object information of a first type of object that has interacted with the triggering content; the first type of object is an object that performs interactive behavior in a first display mode, and the first display mode is the content display mode before the triggering content is triggered;

The determining unit is configured to execute a first interaction prediction model corresponding to the first display mode to obtain first content feature information of the candidate content in the first display mode, and first object feature information of the first type of object in the first display mode; the first interaction prediction model is used to determine the first interaction prediction information of the content to be recommended in the first display mode relative to the first type of object.

The prediction unit is configured to perform a second interaction prediction based on the content feature information of the triggered content, the first content feature information, and the first object feature information to determine the second interaction prediction information of the object to be recommended to the candidate content in the second display mode; the second display mode is the content display mode entered after the triggered content is triggered.

The recommendation unit is configured to determine recommended content from the candidate content based on the second interaction prediction information, and push the recommended content to the object to be recommended in the second display mode.

10. The apparatus according to claim 9, wherein the prediction unit is further configured to perform the following operations: acquiring at least one of the following information: historical interaction information of the object to be recommended in relation to content in the second display mode, content feature information of the candidate content, and object feature information of a second type of object that has interacted with the candidate content; the second type of object being an object that performs interactive behavior in the second display mode; and determining second interaction prediction information of the object to be recommended in relation to the candidate content in the second display mode based on the at least one information, the content feature information of the triggering content, the first content feature information, and the first object feature information.

11. The apparatus according to claim 10, wherein the prediction unit is further configured to perform feature extraction processing on the historical interaction information and the content feature information of the candidate content through an attention unit when the at least one piece of information includes the historical interaction information and the content feature information of the candidate content, to obtain extracted feature information; and based on the extracted feature information, the content feature information of the candidate content, the content feature information of the triggering content, the first content feature information, and the object feature information of the first type of object, to obtain second interaction prediction information of the object to be recommended on the candidate content.

12. The apparatus according to claim 9, wherein the recommendation unit is further configured to perform the following: dividing the candidate content into queues based on matching information of the candidate content and the triggering content on a preset plurality of attribute information to obtain a plurality of candidate content sequences; sorting the candidate content in each of the candidate content sequences according to the second interaction prediction information to obtain a plurality of sorted candidate content sequences; and determining recommended content from at least one of the sorted candidate content sequences according to the priority order of each of the candidate content sequences.

13. The apparatus according to claim 12, wherein the recommendation unit is further configured to perform the acquisition of a preset plurality of empty queues; the empty queues are determined based on the priority order of each attribute information and the number of matches on the attribute information; according to the matching information corresponding to the candidate content and the trigger content, and the priority order of each attribute information, each candidate content is divided into each of the empty queues to obtain a plurality of candidate content sequences.

14. The apparatus according to claim 13, wherein the recommendation unit is further configured to perform the following: obtain a target number of recommended content; select candidate content corresponding to the target number from at least one sorted candidate content sequence in descending order of priority of each candidate content sequence, and use it as recommended content.

15. The apparatus according to claim 9, wherein the prediction unit is further configured to execute an information prediction processing on the content feature information of the triggering content, the first content feature information, and the first object feature information through a second interaction prediction model corresponding to the second display mode, to obtain second interaction prediction information of the object to be recommended on the candidate content in the second display mode.

16. A training device for a content recommendation model, characterized in that it comprises:

The acquisition unit is configured to acquire sample interaction data; the sample interaction data includes sample content feature information of sample trigger content, first sample content feature information of sample candidate content associated with the sample trigger content in a first display mode, second interaction information of a second type of sample object with the sample candidate content, and first sample object feature information of a first type of sample object that has interacted with the sample trigger content in the first display mode; the first sample content feature information and the first sample object feature information are predicted by a first interaction prediction model corresponding to the first display mode;

The prediction unit is configured to perform information prediction processing on the sample content feature information, the first sample content feature information, and the first sample object feature information through a second interactive prediction model to be trained, so as to obtain the second interactive prediction information of the second type of sample object on the sample candidate content in the second display mode.

The training unit is configured to train the second interaction prediction model to be trained based on the loss value between the second interaction prediction information and the second interaction information, so as to obtain the trained second interaction prediction model as a content recommendation model.

17. An electronic device, characterized in that it comprises:

processor;

Memory used to store the processor's executable instructions;

The processor is configured to execute the instructions to implement the content recommendation method as described in any one of claims 1 to 7 or the training method of the content recommendation model as described in claim 8.

18. A computer-readable storage medium, characterized in that, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the content recommendation method as described in any one of claims 1 to 7 or the training method of the content recommendation model as described in claim 8.

19. A computer program product comprising instructions, wherein when the instructions are executed by a processor of an electronic device, the electronic device is able to perform the content recommendation method as described in any one of claims 1 to 7 or the training method of the content recommendation model as described in claim 8.