CN110555753A - recommendation-based ranking control method and device, computer equipment and storage medium - Google Patents

Abstract

The invention discloses a ranking control method, device, computer equipment and storage medium based on recommendation. The method includes: acquiring product features, wherein the product features include a feature vector of the product obtained by performing feature learning on the knowledge map of the product; embedding the product features into a neural network model preset by the recommendation system for training Generate a sorting control model for intelligent recommendation, wherein the sorting control model is used to generate product sorting rules matching the user based on user behavior data; sort the products according to the product sorting rules generated by the sorting control model A corresponding sorted list of products is generated and recommended to the user. The invention introduces the embedding features learned from the knowledge graph to train the sorting control model, and solves the cold start problem of product recommendation and the sparsity problem of user data.

Description

Recommendation-based ranking control method, device, computer equipment and storage medium

technical field

The present invention relates to the technical field of intelligent recommendation, in particular, the present invention relates to a ranking control method, device, computer equipment and storage medium based on recommendation.

Background technique

In recent years, the rapid development of cloud computing, Internet of Things, mobile Internet, artificial intelligence and other technologies has brought a lot of convenience to people's work and life. Users can easily search for the information they want through the Internet. However, when faced with the explosive growth of network information, it is difficult for users to make efficient choices. The emergence of recommender systems provides an effective way to solve information overload. The recommendation system is a subset of the information filtering system, which aims to predict the user's preference for the product based on the user's preferences, habits, individual needs and product characteristics, recommend the most suitable product for the user, and help the user make a decision quickly .

Traditional recommendation systems are mostly models based on the Attention mechanism of user historical behavior sequences, including content-based recommendation systems, collaborative filtering-based recommendation systems, knowledge-based recommendation systems, demographic-based recommendation systems, and hybrid recommendation systems. system and so on. These recommendation systems use the traditional Embedding&MLP structure, which has problems such as insufficient model generalization ability and insufficient novelty of recommendation results. However, for new users, new products or new systems, due to the lack of historical interaction information, accurate modeling and recommendation cannot be performed, and product recommendation cold start problems exist. Moreover, the features of user portraits are high-dimensional and sparse, the learning effect is poor, and the memory ability of the model is low, so it is difficult for the system to achieve the recommendation effect required by the user.

Contents of the invention

The purpose of the present invention is to solve at least one of the above technical defects, especially insufficient model generalization ability, insufficient novelty of recommendation results, and technical defects such as product cold start in recommendation.

In order to solve the above-mentioned technical problems, the present invention provides a sorting control method based on recommendation, comprising the following steps:

Obtaining product features, wherein the product features include a feature vector of the product obtained by performing feature learning on the knowledge map of the product;

Embed the product features into the neural network model preset by the recommendation system for training to generate a ranking control model for intelligent recommendation, wherein the ranking control model is used to generate products that match the user based on the user's behavior data collation;

The products are sorted according to the product sorting rules generated by the sorting control model to generate a corresponding product sorting list and recommended to the user.

Optionally, the acquiring product features, wherein the product features include the step of feature vectors of the product obtained by performing feature learning on the knowledge graph of the product, further includes:

Obtain the knowledge map of the product;

extracting entity attribute parameters of the product from the knowledge graph;

The entity attribute parameters of the product are converted into feature vectors in the form of embedding.

Optionally, after the step of converting the entity attribute parameters of the product into feature vectors in the form of embeddings, it also includes:

The feature vectors of the products stored in the recommendation system are summarized to generate product features, wherein the product features are embodied as an item embedding matrix.

Optionally, embedding the product features into the neural network model preset by the recommendation system for training to generate a ranking control model for intelligent recommendation, wherein the ranking control model is used to generate and match the user’s behavior data with the user’s behavior data. After the steps described above to match the user to the product ordering rules, also include:

Obtain user behavior data;

Analyze the behavior sequence characteristics of the user according to the behavior data;

Finding a first feature vector matching the user's behavior sequence features from the item embedding matrix of the ranking control model;

Weighting is set on the first feature vector according to the user's behavior sequence feature and a preset weighting rule, so as to generate a product ranking rule matching the user.

Optionally, the user's behavior sequence feature is a single category feature of the user, at least embodied as one of the following: interest feature, consumption feature.

Optionally, when the user's behavior sequence feature is an interest feature, the first feature vector is weighted and set according to the user's behavior sequence feature and a preset weighting rule, so as to generate a user-matched The steps for product sorting rules also include:

Obtain operation information during the process of browsing the product by the user;

Analyzing the operation information to obtain the user's interest characteristics;

Find the second feature vector corresponding to the feature of interest from the item embedding matrix in the ranking control model according to the feature of interest of the user, so as to weight the second feature vector to generate Matching product collation.

Optionally, when the user's behavior sequence feature is a consumption feature, the first feature vector is weighted and set according to the user's behavior sequence feature and a preset weighting rule, so as to generate a user-matched The steps for product sorting rules also include:

Obtain the consumption records of the user within the preset time period;

Analyzing the consumption records to obtain the consumption characteristics of the user;

According to the consumption characteristics of the user, the third eigenvector corresponding to the consumption characteristics is found from the item embedding matrix in the ranking control model, so as to weight the third eigenvectors to generate Matching product collation.

In order to solve the above technical problems, the present invention also provides a sorting control device based on recommendation, including:

An acquisition module, configured to acquire product features, wherein the product features include a feature vector of the product obtained by performing feature learning on the knowledge map of the product;

The processing module is used to embed the product features into the neural network model preset by the recommendation system for training to generate a ranking control model for intelligent recommendation, wherein the ranking control model is used to generate and match the user's behavior data according to the user's behavior data. User matching product sorting rules;

An execution module, configured to sort the products according to the product sorting rules generated by the sorting control model to generate a corresponding product sorting list and recommend them to the user.

Optionally, the recommendation-based ranking control device further includes:

The first acquisition sub-module is used to acquire the knowledge map of the product;

The first processing submodule is used to extract the entity attribute parameters of the product from the knowledge map;

The first execution sub-module is used to transform the entity attribute parameters of the product into feature vectors in the form of embedding.

Optionally, the recommendation-based ranking control device further includes:

The second execution sub-module is used for summarizing the feature vectors of the products stored in the recommendation system to generate product features, wherein the product features are embodied as an item embedding matrix.

Optionally, the recommendation-based ranking control device further includes:

The second acquisition sub-module is used to acquire user behavior data;

The second processing sub-module is used to analyze the user's behavior sequence characteristics according to the behavior data;

The third processing submodule is used to find the first feature vector matching the user's behavior sequence features from the item embedding matrix of the ranking control model;

The third execution submodule is configured to set the weight of the first feature vector according to the user's behavior sequence feature and preset weighting rules, so as to generate a product ranking rule matching the user.

Optionally, the user's behavior sequence feature is a single category feature of the user, at least embodied as one of the following: interest feature, consumption feature.

Optionally, when the user's behavior sequence feature is an interest feature, the recommendation-based ranking control device further includes:

The third acquisition sub-module is used to acquire the operation information in the process of the user browsing the product;

The fourth processing sub-module is used to analyze the operation information to obtain the user's interest characteristics;

The fourth execution submodule is used to find a second feature vector corresponding to the feature of interest from the item embedding matrix in the ranking control model according to the feature of interest of the user, so as to compare the second feature vector Perform weighting settings to generate product sorting rules that match the user.

Optionally, when the user's behavior sequence feature is a consumption feature, the recommendation-based ranking control device further includes:

The fourth obtaining sub-module is used to obtain the consumption records of the user within the preset time period;

The fifth processing sub-module is used to analyze the consumption record to obtain the consumption characteristics of the user;

The fifth execution submodule is used to find a third feature vector corresponding to the consumption feature from the item embedding matrix in the ranking control model according to the consumption feature of the user, so as to calculate the third feature vector Perform weighting settings to generate product sorting rules that match the user.

In order to solve the above-mentioned technical problems, the present invention also provides a computer device, including a memory and a processor, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the processing The controller executes the steps of the above recommendation-based ranking control method.

In order to solve the above-mentioned technical problems, the present invention also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, one or more processors execute the above recommendation-based Sequence the steps of the control method.

The beneficial effects of the present invention are:

The present invention acquires the product features corresponding to the product through the knowledge map of the product, and then embeds the product features as an intermediate parameter into the preset neural network model for training and generates a sorting control model for intelligent recommendation, and then acquires The user's behavior data and the product ranking rules matching the user are obtained from the ranking control model based on the user's behavior data, and finally the products are sorted according to the product ranking rules generated by the ranking control model to generate the corresponding product ranking List and recommend to users. The above method solves the cold start problem of product recommendation by introducing the knowledge map of the product to embed the product features obtained from the knowledge map into the neural network model for model training. The entity attribute class embedding vector feature searches and analyzes the user's behavior sequence characteristics, and generates a product ranking rule that matches the user based on the behavior sequence characteristics, which overcomes the high-dimensional and sparse user portrait features that make it difficult for the system to meet the user's needs. The question of effect.

Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description, or may be learned by practice of the invention.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

FIG. 1 is a schematic flowchart of a basic method of a recommendation-based ranking control method provided by an embodiment of the present invention;

FIG. 2 is a schematic flowchart of a method for acquiring product features in a recommendation-based ranking control method provided by an embodiment of the present invention;

3 is a schematic flowchart of a method for generating product ranking rules based on user behavior data by the ranking control model in the recommendation-based ranking control method provided by an embodiment of the present invention;

FIG. 4 is a schematic flowchart of a method for generating product ranking rules based on user interest characteristics in the recommendation-based ranking control method provided by an embodiment of the present invention;

5 is a schematic flowchart of a method for generating product ranking rules based on user consumption characteristics in the recommendation-based ranking control method provided by an embodiment of the present invention;

FIG. 6 is a basic structural block diagram of a recommendation-based ranking control device provided by an embodiment of the present invention;

FIG. 7 is a block diagram of a basic structure of a computer device provided by an embodiment of the present invention.

Detailed ways

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

In some processes described in the specification and claims of the present invention and the above-mentioned drawings, a plurality of operations appearing in a specific order are contained, but it should be clearly understood that these operations may not be performed in the order in which they appear herein Execution or parallel execution, and the sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. Additionally, these processes can include more or fewer operations, and these operations can be performed sequentially or in parallel. It should be noted that the descriptions of "first" and "second" in this article are used to distinguish different messages, devices, modules, etc. are different types.

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

Example

Those skilled in the art can understand that the "terminal" and "terminal equipment" used here not only include wireless signal receiver equipment, which only has wireless signal receiver equipment without transmission capabilities, but also include receiving and transmitting hardware. A device having receiving and transmitting hardware capable of performing bi-directional communication over a bi-directional communication link. Such equipment may include: cellular or other communication equipment, which has a single-line display or a multi-line display or a cellular or other communication equipment without a multi-line display; PCS (Personal Communications Service, personal communication system), which can combine voice, data Processing, facsimile and/or data communication capabilities; PDA (Personal Digital Assistant, Personal Digital Assistant), which may include radio frequency receiver, pager, Internet/Intranet access, web browser, notepad, calendar and/or GPS (Global Positioning System (Global Positioning System) receiver; a conventional laptop and/or palmtop computer or other device having and/or including a radio frequency receiver. As used herein, a "terminal", "terminal device" may be portable, transportable, installed in a vehicle (air, sea, and/or land), or adapted and/or configured to operate locally, and/or In distributed form, the operation operates at any other location on Earth and/or in space. The "terminal" and "terminal equipment" used here can also be communication terminals, Internet terminals, music/video playback terminals, such as PDAs, MIDs (Mobile Internet Devices, mobile Internet devices) and/or with music/video playback terminals. Functional mobile phones, smart TVs, set-top boxes and other devices.

The user terminal mentioned in this embodiment is the above-mentioned terminal.

Please refer to FIG. 1 . FIG. 1 is a schematic flowchart of a basic method of a ranking control method based on recommendation provided by an embodiment of the present invention.

As shown in Figure 1, the described ranking control method based on recommendation includes the following steps:

S100: Obtain product features, wherein the product features include a feature vector of the product obtained by performing feature learning on a product knowledge graph.

Facing the explosive growth of network information, the present invention provides a ranking control method based on recommendation, which can be applied in recommendation systems. Because all products have their corresponding product attribute parameters, including but not limited to application scenarios, categories, prices, etc. of the products. In this method, feature learning is performed on all products stored in the recommendation system to obtain feature vectors corresponding to each product in the recommendation system.

In some embodiments, please also refer to FIG. 2 . FIG. 2 is a schematic flowchart of a method for acquiring product features in the recommendation-based ranking control method provided by an embodiment of the present invention.

As shown in Figure 2, the step S100 includes the following steps: S110: Acquire the knowledge graph of the product; S120: Extract the entity attribute parameters of the product from the knowledge graph; S130: Convert the entity attribute parameters of the product is a feature vector in the form of embedding.

In this embodiment, when learning the features of the product to obtain the feature vector of the product, the pre-saved knowledge map of the product is first called, and the knowledge map records data about product-related attribute parameters and the relationship between each attribute parameter. Then extract various relevant attribute parameters of the product from the knowledge graph, and use the TransH/TransE algorithm to transform the various relevant attribute parameters extracted from the knowledge graph of the product, thereby forming a feature vector in the form of embedding. Wherein, embedding is a vector matrix, and in the vector matrix, the parameters representing the attributes of the product are expressed in the form of vectors. Further, product features are generated by summarizing embedding vectors corresponding to all products in the recommendation system, wherein the product features are embodied as an item embedding matrix (Item Embedding Matrix, M for short).

S200: Embedding the product features into a preset neural network model for training to generate a recommendation-based ranking control model, wherein the ranking control model is used to generate a ranking that matches the user based on the user's behavior data. Product sorting rules.

The item embedding matrix (Item Embedding Matrix, referred to as M) is originally an intermediate parameter that the deep neural network model that performs product recommendation in the recommendation system needs to learn during the training process. In this embodiment, feature vectors in the form of embedding vectors representing product attributes are obtained by calling the pre-saved product knowledge map, and then performing feature learning on the knowledge map, and then the item embedding matrix is formed by summarizing these feature vectors. After the item embedding matrix is obtained, the item embedding matrix is embedded into the neural network model for training to generate a ranking control model that can realize intelligent recommendation. Wherein, the ranking control model has the function of generating a product ranking rule that matches the user based on the user's behavior data, so that the recommendation system can make a list of all products in the recommendation system according to the product ranking rule before performing intelligent product recommendation. or partial products to sort. Specifically, the product sorting rule can predict and analyze the user's behavior based on the user's behavior data, predict the probability that the user will choose each product in the recommendation system, and then sort the products according to the probability.

The neural network model in the recommendation system itself needs to learn various relevant attribute parameters of the product. In this embodiment, the embedding vectors learned from the product knowledge map are summarized to form an item embedding matrix and embedded into the neural network model to integrate The embedding matrix of this project is used as an intermediate parameter for training the neural network model, and at the same time freezes the original embedding learning of product-related attribute parameters in the neural network model, so that the model does not need to learn the embedding vector by itself when training the neural network model. It avoids the situation that the product clustering effect learned by the model itself is not ideal. It can be understood that, in the item embedding matrix, the weight of each feature vector can be set correspondingly according to the degree of influence or importance of each feature vector on the product, so that the model training is more accurate and the product clustering effect is better. ideal.

S300: Sort the products according to the product sorting rule generated by the sorting control model to generate a corresponding product sorting list and recommend it to the user.

In this embodiment, after the ranking control model is trained and generated, when the recommendation system needs to perform intelligent recommendation operations, the user's behavior data is input into the ranking control model to combine the input behavior data with the ranking In the control model, the item embedding matrix formed by the product features is compared, so as to predict and analyze the user's behavior, predict the probability of the user choosing each product in the recommendation system, and then generate the corresponding value according to the predicted probability of the user choosing each product. The product sorting rules that match the user described above. Then sort the products stored in the recommendation system according to the product ranking rules, generate a product ranking list that meets the user's interests and preferences, and recommend the generated product ranking list to the user to complete the intelligent recommendation of the recommendation system operate.

In some specific embodiments, please refer to FIG. 3 . FIG. 3 is a schematic flowchart of a method for generating product ranking rules based on user behavior data by the ranking control model in the recommendation-based ranking control method provided by an embodiment of the present invention.

As shown in FIG. 3 , step S400 to step S700 are further included after step S200 . Among them, S400: Obtain the user's behavior data; S500: Analyze the user's behavior sequence characteristics according to the behavior data; S600: Find out from the item embedding matrix of the ranking control model that matches the user's behavior sequence characteristics. The first feature vector; S700: Set the weight of the first feature vector according to the user's behavior sequence feature and preset weighting rules, so as to generate a product ranking rule matching the user.

In this embodiment, when the recommendation system performs an intelligent recommendation operation on a user, the user's behavior data is obtained, and the user's behavior data includes the user's previous behavior data within a preset time period and/or the user's current time Behavior data executed by the node. After the user's behavior data is obtained, analyze the user's behavior sequence characteristics according to the user's behavior data, for example, analyze a series of behavior data of the user over time, including analyzing the mutual characteristic relationship between each behavior. Take the time-sensitive product purchased by a user as an example. For a certain type of service product, the product purchased by a user for the first time is time-sensitive for one month. After the product purchased for the first time expires, the product purchased for the second time The timeliness of the product is 3 months. After the product purchased for the second time expires, the timeliness of the product purchased for the third time is half a year. After the product purchased for the third time expires, the timeliness of the product purchased for the fourth time is 1 year. Based on the above behavior data, it can be analyzed that the user's behavior sequence is characterized by successive extension of timeliness. Furthermore, a first feature vector matching the behavior sequence features of the user is found from the item embedding matrix of the ranking control model. Specifically, as in the above-mentioned example where a user purchases a time-sensitive product, by using a method such as a lookup function to search from the item embedding matrix (ItemEmbedding Matrix, referred to as M), the user's behavior sequence features are obtained. The first feature vector that matches is the "expiry date" feature. Based on the "validity period" feature, combined with the user's behavior sequence characteristics, the "validity period" feature is weighted according to the preset weighting rules. For example, in the above-mentioned example where the user purchases a time-sensitive product, the user's corresponding behavior sequence feature is The timeliness is gradually extended, and the timeliness of the last purchased product is 1 year. Based on this information, it can be inferred that the user prefers to buy products with a slightly longer timeliness than 1 year. Furthermore, if the relevant The timeliness of the product is roughly divided into 1 month, 3 months, half a year, 1 year, 2 years, 5 years, 7 years, 10 years and other categories. Then, at this time, it is estimated that the result is slightly longer than 1 year The time limit is 2 years, that is, with 2 years as the center, each time limit category is weighted according to the preset weighting rules. The preset weighting rules can be obtained through model training or through custom settings, for example, through self-defined When defining the settings, the weight ratio of the center can be set to 1, and the farther away from the center, the weight ratio decreases step by step for weighting. Specifically in this embodiment, as in the example of the above-mentioned user purchasing a time-sensitive product, which is gradually reduced by 0.1, then for 1 month, 3 months, half a year, 1 year, 2 years, 5 years, 7 years, For categories such as 10 years, the corresponding timeliness weight ratio can be set to 0.6, 0.7, 0.8, 0.9, 1, 0.9, 0.8, 0.7 in sequence. Furthermore, the product sorting rule is to sort the products according to the weight ratio from high to low. Therefore, according to the user's preference, a product sorting rule matching the user can be formulated, which accurately reflects the user's degree of preference for the product. Further, the user's behavior sequence feature is a single category feature of the user, which can be selected from a limited number of categories, at least embodied as one of the following: interest feature, consumption feature.

In some specific embodiments, please refer to FIG. 4 . FIG. 4 is a schematic flowchart of a method for generating product ranking rules based on user interest characteristics in the recommendation-based ranking control method provided by an embodiment of the present invention.

As shown in FIG. 4 , when the user's behavior sequence feature is an interest feature, the step S700 may further include steps S710 to S730. Among them, S710: Acquire the operation information in the process of user browsing products; S720: Analyze the operation information to obtain the user's interest characteristics; S730: According to the user's interest characteristics, select the items in the ranking A second eigenvector corresponding to the interest feature is found in the embedding matrix, and weighted settings are performed on the second eigenvector to generate a product ranking rule that matches the user.

In this embodiment, if the user behavior sequence feature that needs to be analyzed is the interest feature, the user's operation information during the product browsing process can be obtained, and the interest feature representing the user's interest preference can be analyzed according to the operation information. Furthermore, searching the item embedding matrix in the ranking control model according to the interest feature to obtain a second feature vector corresponding to the interest feature. Furthermore, the weighted setting is performed on the second feature vector to generate a product sorting rule matching the user. Specifically, the operation information may be text information input by the user, or operation event information performed by the user. Among them, when taking the text information input by the user as an example, for example, for the user's behavior of picking fruits, when the text information input by the user is pears, the analysis shows that the user's interest characteristics are the category characteristics of cool in nature and clearing away heat. For a category feature, the system can set or obtain all the categories contained in the category feature by using sample data for model training, and can also classify all the categories contained in the category feature through custom settings, and then These categories are weighted to get the corresponding product sorting rules. At this time, according to the product sorting rules, the fruits with the characteristics of cool nature and heat clearing included in the recommendation system can be prioritized. Taking the operation event information performed by the user as an example, all the operation events in the process of the user browsing the product can be classified into categories to form a category feature, and each operation event can be weighted and set. Wherein, the greater the weight ratio corresponding to the operation event, the higher the user's interest in the product is represented by the operation event. For example, the user's operation events are divided into four types of operation events, such as click, share, favorite, and purchase. This kind of operation event reflects that the relationship between the user's interest in the product is click<share<favorite<purchase. In this way, according to the user's operation event, it can be reflected that the user has a difference in the interest value of the product in the process of browsing the product, which better reflects the user's interest preference. At this time, the formulation of the product sorting rule may be to sort the products according to the user's interest in the products.

Further, in some specific embodiments, since the user's interest in the product will change with the change of external factors such as time, experience, demand, etc., and has greater volatility, so when analyzing the user's behavior sequence characteristics Time features can be introduced, and different weight values can be set for operations in different time periods or time periods. For example, the user's recent operations have a greater impact on the extraction of interest features, and the corresponding set weight value is larger, and the user is older. The operation of has less influence on the extraction of the feature of interest, and its corresponding weight value is smaller. Moreover, an update cycle is set for the user's interest value in the product, and the user's interest value in the product is reacquired every time a certain period of time (for example, 1 month) passes, so as to ensure that the user's interest value in the record conforms to the user's interest value in the product. The actual situation.

In some specific embodiments, please refer to FIG. 5 . FIG. 5 is a schematic flowchart of a method for generating product ranking rules based on user consumption characteristics in the recommendation-based ranking control method provided by an embodiment of the present invention.

As shown in FIG. 5 , when the user's behavior sequence feature is an interest feature, the step S700 may further include steps S740 to S760. Among them, S740: Obtain the consumption records of the user within a preset time period; S750: Analyze the consumption records to obtain the consumption characteristics of the user; A third eigenvector corresponding to the consumption feature is found from the item embedding matrix, and weighted settings are performed on the third eigenvector to generate a product ranking rule matching the user.

In this embodiment, if the user behavior sequence feature to be analyzed is a consumption feature, the consumption record representing the user's consumption preference can be analyzed according to the consumption record by acquiring the user's consumption record within a preset time period. A third feature vector corresponding to the consumption feature is obtained by searching the item embedding matrix in the ranking control model according to the consumption feature. Furthermore, the weighted setting is performed on the second feature vector to generate a product sorting rule matching the user. When the user's behavior sequence feature is a consumption feature, the corresponding third feature vector may be embodied as a single category feature such as consumption level and/or consumption frequency. Among them, for the category characteristic of consumption level, the consumption levels can be divided according to the price of the product, and each level has a corresponding consumption amount range, so as to judge the consumption preference of the user according to the collected consumption records of the user. Specifically, by obtaining the amount of consumption in each consumption record of the user within the preset time period, and counting the number of consumption records of the user falling into each consumption level within the preset time period, the statistically obtained The consumption level with a large number of consumption records is used as the consumption level characteristic of the user, and then the corresponding weight value is preset for the consumption level divided according to the price of the product, and the preset principle is the consumption level corresponding to the consumption level characteristic of the user As the center, the farther away from the consumption level of the consumption level feature, the lower the weight value corresponding to its setting. At this time, the formulation of the product sorting rules can be to sort the products according to the consumption level characteristics of the users, that is, the products corresponding to the consumption level corresponding to the consumption level characteristics of the users are prioritized, and the products of adjacent consumption levels are ranked first. Of. The consumption frequency is the number of times the user consumes products of the same or similar category within a specified time period. For the category feature of consumption frequency, the formulating of the product sorting rule may be to sort according to the degree of matching between the consumption frequency and the timeliness information of the products. For example, for insurance products, the insurance period includes 1 month, 3 months, half a year, 1 year, 5 years, 10 years, etc., if the user’s consumption frequency for a certain type of insurance product is once every six months, then the ranking control model will be Priority is given to insurance products with a timeliness of half a year, followed by insurance products of 3 months and 1 year, and insurance products of 1 month, 5 years, and 10 years.

In some specific embodiments, in addition to the user's behavior sequence features, the sample data for training the recommendation model also includes user portrait feature data. This embodiment can also combine the user's portrait features and user behavior sequence features to generate a Match the product collation. Specifically, the user's portrait features are constructed by acquiring the user's personal information, and the user's portrait features are also used as a judgment index for product recommendation and ranking. Wherein, the user profile features may be composed of multiple feature tags, such as gender, age, city, education level, occupation, and so on. The above feature tags are obtained by directly extracting from user information, wherein the user information may be information filled in by the user when registering, or information contained in other accounts associated with the user, etc. The user portrait feature is relatively stable as an inherent attribute of the user, so after extracting the user portrait feature of the user, it may not need to be maintained and updated for a long period of time. Therefore, the user portrait features can be directly obtained from the information filled in by the user when registering or the information contained in the associated account, but there may be a problem of incomplete user information. In this case, according to the user's existing basic information, the user's information that is closest to the user's existing basic information can be found in the database to make up for the basic characteristics that the user has not filled in, so as to form a complete user's basic characteristics .

The recommendation-based ranking control method provided by the above embodiment obtains the product features corresponding to the product through the knowledge map of the product, and then embedding the product feature as an intermediate parameter into the preset neural network model for training and generating intelligent The recommended ranking control model, and then obtain the user’s behavior data and obtain the product ranking rules matching the user from the ranking control model based on the user’s behavior data, and finally generate the product ranking rules according to the ranking control model Sort the products to generate a corresponding sorted list of products and recommend them to users. The above method solves the cold start problem of product recommendation by introducing the knowledge map of the product to embed the product features obtained from the knowledge map into the neural network model for model training. The entity attribute class embedding vector feature searches and analyzes the user's behavior sequence characteristics, and generates a product ranking rule that matches the user based on the behavior sequence characteristics, which overcomes the high-dimensional and sparse user portrait features that make it difficult for the system to meet the user's needs. The question of effect.

In order to solve the above technical problem, an embodiment of the present invention further provides a recommendation-based sorting control device. Please refer to FIG. 6 for details. FIG. 6 is a basic structural block diagram of a recommendation-based sorting control device provided by an embodiment of the present invention.

As shown in FIG. 6 , a ranking control device based on recommendation includes: an acquisition module 10 , a processing module 20 and an execution module 30 . Wherein, the acquisition module 10 is used to acquire product features, wherein the product features include feature vectors of the product obtained by performing feature learning on the product knowledge map; the processing module 20 is used to use the product features Embedded in the neural network model preset by the recommendation system for training to generate a sorting control model for intelligent recommendation, wherein the sorting control model generates product sorting rules that match the user according to the user's behavior data; the execution module 30 is used to sort the products according to the product sorting rules generated by the sorting control model to generate a corresponding product sorting list and recommend them to the user.

In some specific embodiments, the recommendation-based ranking control device further includes: a first acquisition submodule, a first processing submodule, and a first execution submodule. Wherein, the first acquisition submodule is used to acquire the knowledge map of the product; the first processing submodule is used to extract the entity attribute parameters of the product from the knowledge map; the first execution submodule is used to The entity attribute parameters of the product are converted into feature vectors in the form of embedding.

In some specific embodiments, the recommendation-based ranking control device further includes: a second execution submodule. The second execution sub-module is used for summarizing the feature vectors of the products stored in the recommendation system to generate product features, wherein the product features are embodied as an item embedding matrix.

In some specific embodiments, the recommendation-based ranking control device further includes: a second acquisition submodule, a second processing submodule, a third processing submodule, and a third execution submodule. Wherein, the second acquisition submodule is used to acquire user behavior data; the second processing submodule is used to analyze user behavior sequence characteristics according to the behavior data; the third processing submodule is used to obtain user behavior data from the The item embedding matrix of the sorting control model finds out the first feature vector that matches the user's behavior sequence feature; the third execution submodule is used to A weighted setting is performed on the first feature vector to generate a product sorting rule matching the user.

In some specific embodiments, the user's behavior sequence feature in the recommendation-based ranking control device is a single category feature of the user, at least embodied as one of the following: interest feature, consumption feature.

In some specific embodiments, when the user's behavior sequence feature is an interest feature, the recommendation-based ranking control device further includes: a second acquisition submodule, a fourth processing submodule, and a fourth execution submodule. Wherein, the third obtaining sub-module is used to obtain operation information in the process of users browsing products; the fourth processing sub-module is used to analyze the operation information to obtain the user's interest characteristics; the fourth The execution sub-module is used to find a second feature vector corresponding to the feature of interest from the item embedding matrix in the ranking control model according to the feature of interest of the user, so as to set the weight of the second feature vector Generate product collations that match this user.

In some specific embodiments, when the user's behavior sequence is characterized by consumption, the recommendation-based ranking control device further includes: a fourth acquisition submodule, a fifth processing submodule, and a fifth execution submodule. Wherein, the fourth obtaining submodule is used to obtain the consumption records of the user within a preset time period; the fifth processing submodule is used to analyze the consumption records to obtain the consumption characteristics of the user; the fifth execution submodule It is used to find out a third feature vector corresponding to the consumption feature from the item embedding matrix in the ranking control model according to the consumption feature of the user, so as to weight the third feature vector to generate a The user matches the product sorting rules.

The recommendation-based ranking control device described in the above-mentioned embodiments obtains the product features corresponding to the product through the product knowledge map, and then uses the product feature as an intermediate parameter to embed it into the preset neural network model for training and generates a realization The ranking control model of intelligent recommendation, and then obtain the user's behavior data and the product ranking rules matching the user from the ranking control model according to the user's behavior data, and finally according to the product ranking generated by the ranking control model The rules sort the products to generate a corresponding sorted list of products and recommend them to users. The above method solves the cold start problem of product recommendation by introducing the knowledge map of the product to embed the product features obtained from the knowledge map into the neural network model for model training. The entity attribute class embedding vector feature searches and analyzes the user's behavior sequence characteristics, and generates a product ranking rule that matches the user based on the behavior sequence characteristics, which overcomes the high-dimensional and sparse user portrait features that make it difficult for the system to meet the user's needs. The question of effect.

To solve the above technical problem, an embodiment of the present invention further provides a computer device. Please refer to FIG. 7 for details. FIG. 7 is a basic structural block diagram of a computer device provided by an embodiment of the present invention.

As shown in FIG. 7, a schematic diagram of the internal structure of the computer equipment. As shown in FIG. 7, the computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected through a system bus. Wherein, the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions, and the database can store control information sequences, and when the computer-readable instructions are executed by the processor, the processor can realize a A recommendation-based ranking control method. The processor of the computer equipment is used to provide computing and control capabilities and support the operation of the entire computer equipment. Computer-readable instructions may be stored in the memory of the computer device, and when the computer-readable instructions are executed by the processor, the processor may execute a ranking control method based on recommendations. The network interface of the computer device is used for connecting and communicating with the terminal. Those skilled in the art can understand that the structure shown in Figure 7 is only a block diagram of a part of the structure related to the solution of this application, and does not constitute a limitation to the computer equipment on which the solution of this application is applied. The specific computer equipment can be More or fewer components than shown in the figures may be included, or some components may be combined, or have a different arrangement of components.

In this embodiment, the processor is used to execute the specific functions of the acquisition module 10, the processing module 20 and the execution module 30 in FIG. 6, and the memory stores program codes and various data required for executing the above modules. The network interface is used for data transmission between user terminals or servers. The memory in this embodiment stores the program codes and data needed to execute all submodules in the sequence control device based on the recommendation, and the server can call the program code and data of the server to execute the functions of all submodules.

The computer device described in the above embodiment acquires the product features corresponding to the product through the knowledge graph of the product, and then uses the product feature as an intermediate parameter to embed it into the preset neural network model for training and generates a ranking for intelligent recommendation Control the model, and then obtain the user’s behavior data and obtain the product ranking rules matching the user from the ranking control model according to the user’s behavior data, and finally sort the products according to the product ranking rules generated by the ranking control model Sorting generates a corresponding sorted list of products and recommends them to users. The above method solves the cold start problem of product recommendation by introducing the knowledge map of the product to embed the product features obtained from the knowledge map into the neural network model for model training. The entity attribute class embedding vector feature searches and analyzes the user's behavior sequence characteristics, and generates a product ranking rule that matches the user based on the behavior sequence characteristics, which overcomes the high-dimensional and sparse user portrait features that make it difficult for the system to meet the user's needs. The question of effect.

The present invention also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, one or more processors execute the recommendation-based Sequence the steps of the control method.

Those of ordinary skill in the art can understand that realizing all or part of the processes in the methods of the above embodiments can be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. During execution, it may include the processes of the embodiments of the above-mentioned methods. Wherein, the aforementioned storage medium may be a nonvolatile storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM).

It should be understood that although the various steps in the flow chart of the accompanying drawings are displayed sequentially according to the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but may be executed at different times, and the order of execution is also It is not necessarily performed sequentially, but may be performed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

The above descriptions are only part of the embodiments of the present invention. It should be pointed out that those skilled in the art can make some improvements and modifications without departing from the principles of the present invention. It should be regarded as the protection scope of the present invention.

Claims (10)

1. A recommendation based ranking control method, comprising the steps of:

Obtaining product features, wherein the product features comprise feature vectors of a product obtained by feature learning of a knowledge graph of the product;

embedding the product characteristics into a neural network model preset by a recommendation system for training to generate a sequencing control model for realizing intelligent recommendation, wherein the sequencing control model is used for generating a product sequencing rule matched with a user according to behavior data of the user;

And sorting the products according to the product sorting rule generated by the sorting control model to generate a corresponding product sorting list and recommending the product sorting list to the user.

2. The recommendation-based ranking control method of claim 1 wherein product features are obtained, wherein the product features include a step of feature vectors of the product obtained by feature learning of a knowledge graph of the product, further comprising:

acquiring a knowledge graph of a product;

extracting entity attribute parameters of the product from the knowledge-graph;

and converting the entity attribute parameters of the product into a feature vector in an embedding form.

3. The recommendation-based ranking control method of claim 2 wherein, after the step of converting the entity attribute parameters of the product into feature vectors in embedding form, further comprising:

Summarizing the feature vectors of the products stored in the recommendation system to generate product features, wherein the product features are embodied as item embedding matrixes.

4. The recommendation-based ranking control method according to claim 3, wherein the step of embedding the product features into a neural network model preset by a recommendation system for training to generate a ranking control model for realizing intelligent recommendation is further followed by the step of generating product ranking rules matching with the user according to the behavior data of the user:

acquiring behavior data of a user;

Analyzing the behavior sequence characteristics of the user according to the behavior data;

Finding out a first feature vector matched with the behavior sequence feature of the user from the item embedding matrix of the sequencing control model;

and performing weighting setting on the first feature vector according to the behavior sequence features of the user and a preset weighting rule to generate a product sorting rule matched with the user.

5. The recommendation-based ranking control method according to claim 4, wherein the behavior sequence feature of the user is a single category feature of the user, and is at least one of the following: interest features, consumption features.

6. The recommendation-based ranking control method according to claim 5, wherein when the behavior sequence feature of the user is an interest feature, the step of performing weighting setting on the first feature vector according to the behavior sequence feature of the user and a preset weighting rule to generate a product ranking rule matching with the user further comprises:

acquiring operation information of a user in a product browsing process;

Analyzing the operation information to obtain the interest characteristics of the user;

And searching a second feature vector corresponding to the interest feature from the item embedding matrix in the sorting control model according to the interest feature of the user, and performing weighting setting on the second feature vector to generate a product sorting rule matched with the user.

7. the recommendation-based ranking control method according to claim 5, wherein when the behavior sequence feature of the user is a consumption feature, the step of performing weighting setting on the first feature vector according to the behavior sequence feature of the user and a preset weighting rule to generate a product ranking rule matching with the user further comprises:

Acquiring consumption records of a user within a preset time period;

Analyzing the consumption record to obtain the consumption characteristics of the user;

And searching a third feature vector corresponding to the consumption feature from the item embedding matrix in the sorting control model according to the consumption feature of the user, and performing weighting setting on the third feature vector to generate a product sorting rule matched with the user.

8. A recommendation based ranking control apparatus comprising:

the system comprises an acquisition module, a processing module and a display module, wherein the acquisition module is used for acquiring product characteristics, and the product characteristics comprise characteristic vectors of a product, which are obtained by characteristic learning of a knowledge graph of the product;

the processing module is used for embedding the product characteristics into a neural network model preset by a recommendation system for training to generate a sequencing control model for realizing intelligent recommendation, wherein the sequencing control model is used for generating a product sequencing rule matched with a user according to behavior data of the user;

And the execution module is used for sequencing the products according to the product sequencing rule generated by the sequencing control model to generate a corresponding product sequencing list and recommending the product sequencing list to the user.

9. A computer device comprising a memory and a processor, the memory having stored therein computer-readable instructions that, when executed by the processor, cause the processor to perform the recommendation-based ranking control method of any of claims 1 to 7.

10. A storage medium storing computer-readable instructions which, when executed by one or more processors, cause the one or more processors to perform the recommendation-based ranking control method of any of claims 1 to 7 above.