CN110781389A - Method and system for generating recommendations for users - Google Patents

Abstract

The present application relates to a method for generating recommendations for a user, the method comprising: generating a recommendation model, wherein the recommendation model is used for recommending articles for a user; generating a multi-view social model, the multi-view social model corresponding to a plurality of social relationship types; combining the recommendation model with the multi-view social model to arrive at a multi-view social recommendation model; training the multi-view social recommendation model; and generating a recommendation for the user using the trained multi-view social recommendation model. Related systems and computer storage media are also disclosed. The recommendation method and the recommendation device can better generate recommendations for the user.

Description

Method and system for generating recommendations for users

technical field

The present invention relates to recommendations, and more particularly, to a method and system for generating recommendations for users.

Background technique

At present, social network has become more and more popular. For example, specialized social networking sites such as WeChat and Weibo provide social networking services. In addition, services such as Alipay and Office 365 also support social networking features. In these social networks, there are social relationships between users.

With the popularity of social networks, recommendation systems based on social networks have also been developed. For example, in some social networks, content is recommended to a user based on the content viewed by the user's friends. For another example, in some social networks, products are recommended to the user based on the products concerned by the user's relatives.

However, in current social network-based recommender systems, only one type of social relationship is usually considered. Part of the reason is that some social networks have only a single type of social relationship. For example, Weibo only has a follow relationship. However, even in systems that have or have access to multiple types of network relationships for users, existing systems still only consider one type of social relationship. Considering only one social relationship makes the recommendations generated for users not sufficiently accurate.

SUMMARY OF THE INVENTION

In order to improve the accuracy of social network recommendation, the present invention provides a method and system for generating recommendation for users.

The present invention achieves the above-mentioned objects through the following technical solutions.

In one aspect, a method for generating recommendations for a user is disclosed, the method comprising:

generating a recommendation model, the recommendation model is used to recommend items for the user;

generating a multi-view social model, the multi-view social model corresponding to a plurality of social relationship types;

combining the recommendation model and the multi-view social model to obtain a multi-view social recommendation model;

training the multi-view social recommendation model; and

Recommendations are generated for the user using a trained multi-view social recommendation model.

Preferably, generating the multi-view social model includes:

Generate multiple single-view social models, and each single-view social model corresponds to a social relationship type;

The multiple single-view social models are combined to obtain a multi-view social model.

Preferably, the plurality of single-view social models correspond to a plurality of social networking services.

Preferably, each of the plurality of single-view social models corresponds to a different social networking service.

Preferably, the plurality of social relationship types include two or more of friend relationships, family relationships, communication relationships, and transfer relationships.

Preferably, combining the plurality of single-view social models includes summing objective functions of the plurality of single-view social models.

Preferably, combining the plurality of single-view social models includes averaging objective functions of the plurality of single-view social models.

Preferably, combining the plurality of single-view social models includes weighted summation of objective functions of the plurality of single-view social models.

Preferably, combining the plurality of single-view social models includes performing a pooling operation on objective functions of the plurality of single-view social models.

Preferably, combining the plurality of single-view social models includes determining a combining function for combining the plurality of single-view social models.

Preferably, using the trained multi-view social recommendation model to generate recommendations for the user comprises:

obtaining user information of the user, the user information including the user's known rating information on the item and social information of multiple social relationship types between the user and the user;

using a trained multi-view social recommendation model to determine a score for a plurality of candidate items based on the user information of the user; and

Items are recommended to the user based on the scores of the plurality of candidate items.

Preferably, the recommendation model is a matrix decomposition model.

Preferably, training the multi-view social recommendation model includes: using gradient descent method to iteratively solve the objective function of the multi-view social recommendation model to determine parameters of the multi-view social recommendation model.

Preferably, using the trained multi-view social recommendation model to generate a recommendation for the user comprises using the determined parameters to determine a score for the candidate item.

Preferably, generating a recommendation for the user includes generating a recommended commodity or recommended content for the user.

In another aspect, a system for generating recommendations for users is disclosed, the system comprising:

a user information acquisition module, where the user information acquisition module is configured to acquire user information of the user, where the user information includes the user's known rating information on items and social information of multiple social relationship types between the user and the user;

A multi-view social recommendation module, the multi-view social recommendation module is used to determine the scores of multiple candidate items based on the user information of the user, and the multi-view social recommendation module includes a recommendation module and a multi-view social module, wherein:

The recommendation module implements a recommendation model, and the recommendation model is used to recommend items for users,

The multi-view social recommendation module includes multiple single-view social modules and combination modules, the multiple single-view social modules implement multiple single-view social models, and the multiple single-view social models correspond to multiple social relationship types, And the combining module is used for combining the multiple single-view social modules into the multi-view social module.

In yet another aspect, a computer-readable storage medium is disclosed storing instructions that, when executed by a computer, cause the computer to perform the method as described above.

In yet another aspect, a system is disclosed comprising means for performing the method as described above.

Compared with the prior art, the present invention can have the following beneficial effects:

The present invention can make full use of a variety of different types of social relationships to more accurately generate recommendations for users.

Description of drawings

The above summary of the present invention and the following detailed description will be better understood when read in conjunction with the accompanying drawings. It should be noted that the accompanying drawings are merely illustrative of the claimed invention. In the drawings, the same reference numbers represent the same or similar elements.

1A is a schematic diagram illustrating a social graph with a single view.

Figure IB is a schematic diagram illustrating a social graph with multiple views.

Figure 2 shows a schematic diagram of a single-view social recommendation model.

FIG. 3 shows a schematic diagram of a multi-view social recommendation model according to an embodiment of the present specification.

FIG. 4 is a flowchart illustrating a method for generating a recommendation for a user according to an embodiment of the present invention.

FIG. 5 shows a schematic diagram of a system for generating recommendations for users according to an embodiment of the present invention.

Detailed ways

The detailed features and advantages of the present invention are described in detail below in the specific embodiment, and the content is sufficient to enable any person skilled in the art to understand the technical content of the present invention and implement it accordingly, and according to the description, claims and drawings disclosed in this specification. , those skilled in the art can easily understand the related objects and advantages of the present invention.

In order to make full use of various types of social relationships in a social network system and improve the accuracy of recommendation, the present invention provides a method, system and computer-readable medium capable of generating recommendations for users with a multi-view social model.

The social network or social network service referred to in this application includes but is not limited to pure social networking sites (such as WeChat, Weibo, etc.), and also includes any other system with social network functions (such as Alipay, etc.).

The social relationships referred to in this application include, but are not limited to, friend relationships, family relationships, transfer relationships, communication relationships, and the like.

In the following, the basic concepts of social recommendation models are first introduced, followed by recommendation models and social models, respectively, followed by single-view social recommendation models and multi-view social recommendation models, and finally a multi-view social recommendation method according to an embodiment of the present invention.

social recommendation model

The so-called social recommendation model predicts the user's rating of the item based on the known relationship between users and users and the user's rating of the item, and generates recommendations based on such ratings.

The social recommendation model can be formulated in the following form:

Social recommendation model = recommendation model + social model formula (1)

See Figure 2, which shows a schematic diagram of a single-view social recommendation model. A more complete description will be given below.

The recommendation model and the social model are introduced separately below.

Recommendation model

The recommendation model described in this specification (for example, the recommendation model 202 shown in FIG. 2 ) is a machine learning model for recommending items to users. The item recommended to the user may be a commodity or other items such as content.

Various recommendation models can be employed. A common recommendation model is the latent semantic model. Matrix factorization models are the most widespread implementation of latent semantic models. The following takes the matrix decomposition model as an example to illustrate.

The core assumption of matrix factorization-based recommendation algorithms is to represent users and items with latent vectors. These latent vectors represent features shared by a subset of users and items. For example, for a user, the latent vector can be expressed as a user preference feature; for an item, the latent vector can be expressed as an item attribute feature. In many cases, the latent vector may not have practical significance, nor is it necessarily very interpretable, and each dimension does not have a definite label name, so it is called "latent vector". Through matrix decomposition, two small matrices containing latent vectors can be obtained, one represents the implicit features of the user, and the other represents the implicit features of the item. The element values of the matrix represent the degree of conformity of the corresponding user or item to the hidden factors. , both positive and negative.

Suppose there is a user set u and an item set v. There is also a matrix R of user ratings for items, where each element R _ij in matrix R represents the rating of user _ui for item v _j . Using a matrix factorization algorithm, the rating matrix can be decomposed into a user-latent factor matrix and an item latent factor matrix. Assuming that the latent vectors of user _ui and item _vj are denoted by U _i and V _j , respectively, an example of the objective function of a simplified recommendation model can be expressed as follows:

The above objective function aims to minimize the error between the true rating and the predicted rating.

It can be appreciated that the above objective function is a simple example, and other objective functions may also be used, such as adding a regular term, which will not be repeated here.

Equation 2 can be solved using methods such as gradient descent. For example, U and V can be obtained by iteration using gradient descent. Subsequently, for a specific user, user information of the specific user may be obtained, such as the user's known rating information for various items, and the user and user social information. The candidate items can then be scored using the obtained U and V based on the user information. Based on the ratings of each candidate item, a recommendation to the user can be generated.

Other recommendation models may also be employed. For example, examples of recommendation models that can be employed include, but are not limited to, domain-based collaborative filtering models, other latent semantic models such as latent semantic analysis, topic models, and the like. Preferably, the recommendation model adopted in this specification is a user-item relationship-based recommendation model or a user-based recommendation model.

Single View Social Model

Referring to Figure 1A, a prior art social graph with a single view is shown. As shown in Figure 1A, the social graph has only one view, ie, only represents a single social relationship type (eg, friend relationship). For example, in FIG. 1A, user 102 is friends with users 106 and 110, and user 104 is friends with users 106 and 108. It should be pointed out that although the two-way relationship is shown in FIG. 1A , those skilled in the art can appreciate that the social relationship may have a one-way relationship (eg, one-way follow, one-way transfer, etc.).

In some examples, social relationships may have strength. For example, where the relationship between user 102 and user 106 is a friend relationship, the strength of the social relationship may be based on the pro-progress of user 102 and user 106 . For example, the affinity progress may be determined based on similarity of user 102 and user 106, frequency of interactions, greetings used, and the like.

As another example, where the relationship between user 102 and user 106 is an affiliation, the strength of the social relationship may be based on the kinship between user 102 and user 106 . For example, a father-son relationship may be considered to have greater strength than an uncle-nephew relationship.

For another example, where the relationship between the user 102 and the user 106 is a transfer relationship, the strength of the social relationship may be based on the transfer frequency, the transfer amount, and the like between the user 102 and the user 106 .

It can be understood that the stronger the relationship between the two users, the better the effect will be when recommending the other user based on the relevant information of one of the users (such as the user's preference for items, etc.).

In the above example, the strength of the relationship between users is used as an example for description. Those skilled in the art will appreciate that social relationships may also have other dimensions, such as similarity between two users. Two similar users may have similar preferences for items, so that social network-based recommendations can be made using a social model that reflects the similarity between users.

To describe such a single view social graph, there are a number of social models (eg single view social model 204 shown in Figure 2 or single view social models 312, 314...316 shown in Figure 3). For example, assuming that S represents a user-user social relationship matrix, each element _Sif of this matrix S represents the strength and/or similarity of the social relationship between user _ui and user _uf , then a type of social The objective function of the model can be expressed as follows:

There are also social models that employ other constraints. For example, in another social model that can constrain the average relationship between user _ui and user _uf , the objective function of such a model can be expressed as follows:

These single-view social models are well known to those skilled in the art and will not be described in detail here.

Taking the above example as an example, based on the known relationship strength and/or similarity of the users in the single-view social model, the unknown relationship strength and/or similarity of two users may be predicted.

Single View Social Recommendation Model

Referring to FIG. 2, a schematic diagram of a single-view social recommendation model 206 is shown. As shown in FIG. 2, a single-view social recommendation model 206 may be implemented by combining a recommendation model (based on user-item relationships) 202 and a single-view social model (based on user-user relationships) 204.

Through the above single-view social recommendation model, not only the user-item relationship, but also the user-user relationship in a single view is considered when generating recommendations, so that compared with the pure recommendation model, the single-view social recommendation model Recommendation models provide a new dimension to achieve better recommendations.

Continuing the above example, based on the above formula 1, formula 2 and formula 3, it can be determined that the objective function of the single-view social recommendation model can be expressed as follows:

The above single-view social recommendation model can also use the gradient descent method to solve the objective function through continuous updating of multiple iterations, so as to train the above-mentioned single-view social recommendation model. A trained single-view social recommendation model can be used to provide recommendations for new users.

Multi-view social model

Referring to Figure IB, a social graph with multiple views is shown. Unlike the social graph in FIG. 1A , which has a single view, the social graph shown in FIG. 1B has multiple views, eg, each social graph may correspond to a social relationship type. Preferably, different views correspond to different social relationship types. For example, maybe one view corresponds to friend relationships in a social network, another view corresponds to transfer relationships, yet another view corresponds to communication relationships, and so on. Preferably, the plurality of views are from more than one social networking service. In some examples, one or more of the multiple views are from the same social networking service. For example, Alipay can provide both a view associated with a friend relationship and a view associated with a transfer relationship. In other examples, each of the plurality of views is from a different social networking service. For example, the communication relationship can be obtained from the phone book, the payment relationship can be obtained from Alipay, and so on. Preferably, the social graph IB constitutes a heterogeneous network.

For example, as shown in Figure IB, user 102 has not only a friend relationship with users 106 and 110, but also a transfer relationship, but user 102 and user 108 only have a transfer relationship. For another example, as shown in FIG. 1B , user 104 and user 106 only have a friend relationship, user 104 and user 108 only have a communication relationship, and a single user 104 and user 106 have a friend relationship and a communication relationship. Likewise, although two-way relationships are shown in FIG. 1B , those skilled in the art will appreciate that the social relationships may have one-way relationships (eg, one-way follow, one-way transfer, etc.).

Similar to FIG. 1A , the various types of social relationships shown in FIG. 1B can have strengths. An example of the strength of the social relationship has been described above with reference to FIG. 1A and will not be repeated here.

It can be understood that the multi-view social graph of FIG. 1B contains more abundant information than the single-view social graph of FIG. 1A . For example, in the example of FIG. 1B , assuming that the strength of the communication relationship between user 104 and user 108 is the same as the strength of the communication relationship between user 104 and user 110, the additional A friend relationship may mean that user 104 has a more intimate relationship with user 108 than with user 110 . Of course, this is just a simple example, and those skilled in the art can mine more complex information based on multiple views of the social graph.

Likewise, in addition to the strength of the relationship between users, the relationship between users may have other dimensions, such as similarity between users and the like.

To take advantage of this richer information, the present application proposes a multi-view social model to describe a multi-view social graph. In an embodiment of the present invention, a multi-view social model (eg, multi-view social model 304 of FIG. 3 ) is a combination of multiple single-view social models (eg, single-view social model 1 312 of FIG. 3 , single-view social model 1 312 , single-view Social Model 2 314,...Single View Social Model M 316) is constructed. Preferably, each single-view social model may correspond to a social relationship type. Preferably, the social relationship types corresponding to different single-view social models may be the same or different, but the multiple single-view social models correspond to at least two social relationship types. The plurality of social relationship types may include, for example, two or more of friend relationships, family relationships, communication relationships, and transfer relationships. Preferably, the plurality of single-view social models correspond to a plurality of social networking services. Preferably, each of the plurality of single-view social models is based on a different social networking service. Alternatively, two or more of the multiple single-view social models are based on the same social networking service (eg, the Alipay social networking service may provide both transfer and friend social relationship types, which can be used to generate two single-view social networking services) Model).

For example, assuming m∈{1,2,…,M} to represent M single-view social models, each single-view social model representing a type of social relationship, the objective function of the multi-view social model can be expressed as follows :

表示用户i和用户f之间的单视图社交模型m所表示的社交关系的强度和/或两个用户的相似度。

表示用组合函数

来组合多个单视图社交模型m∈{1,2,…,}。所述组合函数

可以是各种函数，即可用各种不同方式来组合该多个单视图社交模型。in

Represents the strength of the social relationship and/or the similarity of the two users represented by the single-view social model m between user i and user f.

Represents a composite function

to combine multiple single-view social models m∈{1,2,…,}. the combined function

There can be various functions, ie, the multiple single-view social models can be combined in various different ways.

In one embodiment of the invention, the combined function can be a summation function

It can be appreciated that although the above combination function adopts a common summation function, a weighted summation function can also be used to assign different weight values to different single-view social models. For example, where certain social network views are more valuable for recommendation than others, these social network views may be given higher weight values.

可以是求平均函数

In another embodiment of the present invention, the combined function

can be an averaging function

可以是池化函数

In yet another embodiment of the present invention, the combined function

Can be a pooling function

The pooling function is a common function in convolutional networks, and those skilled in the art understand its meaning and implementation, and will not be described in detail here.

Those skilled in the art can understand that other combining functions can also be used to combine multiple views.

Which type of combination function to use can be selected by the developer according to needs.

It should be noted that, although in the above example, each view is represented by the same single-view social model (eg, a matrix factorization model), it can be appreciated that different social models can be employed for different views. In this case, multiple single-view social models can still be combined in the manner exemplified above. For example, operations such as summing, weighted summing, averaging, and pooling can be performed on multiple different single-view social models.

Multi-view social recommendation model

In order to make full use of the richer information provided by the multi-view social graph proposed by the present invention to generate recommendations and improve the accuracy of the recommendation, the present invention proposes a multi-view social recommendation model.

Referring to FIG. 3, it shows a schematic diagram of a multi-view social recommendation model according to an embodiment of the present specification. As shown in Figure 3, by combining a recommendation model (based on user-item relationships) and a multi-view social model (based on multiple types of user-user relationships), a variety of different user relationships can be fully utilized, breaking down different types of The barriers between social relationships further increase the number of dimensions utilized, thereby providing better recommendation effects than the single-view social recommendation model.

Specifically, the multi-view social recommendation model 306 combines the recommendation model 302 with the multi-view social model 304 to generate recommendations for the user.

Continuing the above example, replacing the social model in formula (1) with a multi-view social model, the formula of the multi-view social recommendation model can be obtained:

Multi-view social recommendation model = recommendation model + multi-view social model formula (10)

Taking the above example objective function as an example, based on the above Equation 10, Equation 2 and Equation 6, it can be determined that the objective function of the multi-view social recommendation model can be expressed as follows:

等等。The above combined function can be the averaging function in Equation 7 above Summation function pooling function

and many more.

As already explained above, which combination function is chosen can be chosen by the developer according to his needs.

Generating Recommendations Based on Multi-View Social Recommendation Models

After the multi-view social recommendation model is obtained, the multi-view social recommendation model can be trained with the training data, and the multi-view social recommendation model is iteratively updated. After training the multi-view social recommendation model, the trained social recommendation model may be used to generate recommendations for the user.

Taking the example objective function above as an example, the objective function of the multi-view social recommendation model introduced above can be solved using the gradient descent method. For example, for the objective function shown in Equation 11, it can be derived, and by updating and iterating, U and V can be obtained, and the parameters of the multi-view social recommendation model can be obtained. Using this parameter, based on the user latent vector of the specific user and the item latent vector of the specific item, the rating of the specific user for the specific item can be predicted according to the formula R=U ^T V. Recommendations can then be made based on the particular user's ratings of items in the candidate item set. For example, an item with the highest rating may be recommended to the user. Alternatively, a number of top rated items may be recommended to the user. For example, the top three rated items may be recommended to the user.

It should be pointed out that although items (commodities) are used as examples in the above examples, it should be understood that the recommended items may be products (such as physical products or virtual products) or content (such as audio, video, Weibo posts, etc.) ).

Multi-view social recommendation method

Referring to FIG. 4, a flowchart of a method 400 for generating recommendations for a user according to an embodiment of the present invention is shown.

The method 400 may include: at step 402, generating a recommendation model for recommending items for a user. As already described in detail above, the recommendation model may employ any suitable known recommendation model, such as a matrix factorization model.

The method 400 may further include, in step 404, generating a multi-view social model, the multi-view social model corresponding to a plurality of social relationship types. Specifically, generating a multi-view social model may include: generating a plurality of single-view social models, wherein each single-view social model may correspond to a social relationship type. Preferably, the social relationship types corresponding to different single-view social models may be the same or different, but the multiple single-view social models correspond to at least two social relationship types. As described above, the plurality of single-view social models may correspond to a plurality of social networking services. Preferably, each of the plurality of single-view social models corresponds to a different social networking service. Alternatively, two or more of the plurality of single-view social models may be based on the same social networking service. Preferably, the plurality of social relationship types include two or more of friend relationships, family relationships, communication relationships, and transfer relationships.

Generating a multi-view social model may include: further comprising: combining the plurality of single-view social models to obtain a multi-view social model.

Single-view social models can be composed according to composition functions. For example, combining the plurality of single-view social models may include summing the objective functions of the plurality of single-view social models (eg, Equation 7 above), summing a weighted sum, averaging (eg, Equation 8 above), performing Pooling operations (eg Equation 9 above). Combining the plurality of single-view social models may include determining a composition function for combining the plurality of single-view social models. Preferably, a selection of a composition function for composing a single view social model may be received from a developer.

The method 400 may further include: in step 406, combining the recommendation model and the multi-view social model to obtain a multi-view social recommendation model. The way of combining the recommendation model and the multi-view social model may be the same as the way of combining the recommendation model and the single-view social model, for example.

Optionally, the method 400 may further include: in step 408, training the multi-view social recommendation model. For example, training the multi-view social recommendation model may include: using a gradient descent method to iteratively solve the objective function of the multi-view social recommendation model to determine parameters of the multi-view social recommendation model. This parameter can later be used to determine the score of the candidate item to be recommended.

The method 400 may also include, at step 410, generating a recommendation for the user using the trained multi-view social recommendation model.

This step may include: acquiring user information of the user, where the user information includes known rating information of the user on the item and social information of multiple types of social relations between the user and the user. The user information can be represented, for example, as a vector representation of the user, such as a one-hot vector or the like.

This step may further include using a trained multi-view social recommendation model to determine a score for each of the plurality of candidate items based on the user information of the user. For example, based on the parameters of the multi-view social recommendation model, the user's rating for the item may be solved.

This step may further include: recommending items to the user based on the scores of the plurality of candidate items. For example, an item with the highest rating may be recommended to the user. Alternatively, a number of top rated items may be recommended to the user. For example, the top three rated items may be recommended to the user. The recommendation may include, but is not limited to, a recommendation for a commodity or a recommendation for content, and the like.

Using the trained recommendation model to generate recommendations can be performed in any manner known to those skilled in the art, and the process thereof will not be described in detail here.

Multi-view social recommender system

Referring to FIG. 5, there is shown a schematic diagram of a system 500 for generating recommendations for users according to an embodiment of the present invention.

As shown in FIG. 5 , the system 500 may include a user information acquisition module 502 . The user information acquisition module 502 may acquire user information, where the user information includes known rating information of the user on the item and social information of multiple social relationship types between the user and the user. The user information may be, for example, a vector representation of the user obtained based on the user ID.

The system 500 may also include a multi-view social recommendation module 506 . The multi-view social recommendation module 506 may include a recommendation module 508 and a multi-view social module 510 . The recommendation module 508 may implement the recommendation model 302 as described above. The multi-view social recommendation module 510 may include a plurality of single-view social modules 512 and combination modules 514 . Each single-view social module 512 may implement a single-view social model. The combining module 514 is used to combine the plurality of single-view social modules 512 into a multi-view social module 510, for example, any one of the various combining functions described above can be used to perform combining. The multi-view social recommendation module 506 may generate recommendations for users based on user information from the user information acquisition module 502 .

The present application also discloses a computer-readable storage medium comprising computer-executable instructions stored thereon that, when executed by a processor, cause the processor to perform various embodiments described herein Methods.

The present application also discloses a system that can include means for performing the methods of the various embodiments described herein.

It will be appreciated that the methods according to various embodiments of the present invention may be implemented in software, firmware or a combination thereof.

It is understood that the specific order or hierarchy of steps in the disclosed methods is an illustration of exemplary processes. Based upon design preferences, it is understood that the specific order or hierarchy of steps in the methods may be rearranged. The accompanying method claims present elements of the various steps in a sample order, and are not meant to be limited to the specific order or hierarchy presented unless specifically recited herein.

It will be understood that the description of an element herein in the singular or the representation of only one in a drawing does not imply that the number of the element is limited to one. Furthermore, modules or elements described or illustrated herein as separate may be combined into a single module or element, and modules or elements described or illustrated herein as a single module or element may be split into multiple modules or elements.

It should also be understood that the terms and expressions used herein are for descriptive purposes only, and the present invention should not be limited to these terms and expressions. The use of these terms and expressions is not intended to exclude any equivalents of those shown and described (or portions thereof), and it should be recognized that various modifications that may exist should also be included within the scope of the claims. Other modifications, changes and substitutions may also exist. Accordingly, the claims should be deemed to cover all such equivalents.

Also, it should be pointed out that although the present invention has been described with reference to the current specific embodiments, those skilled in the art should realize that the above embodiments are only used to illustrate the present invention, without departing from the present invention. Various equivalent changes or substitutions can also be made under the spirit of the present invention. Therefore, as long as the changes and modifications to the above-mentioned embodiments are within the scope of the essential spirit of the present invention, they will fall within the scope of the claims of the present application.

Claims (15)

1. A method for generating recommendations for a user, the method comprising:

generating a recommendation model, wherein the recommendation model is used for recommending articles for a user;

generating a multi-view social model, the multi-view social model corresponding to a plurality of social relationship types;

combining the recommendation model with the multi-view social model to arrive at a multi-view social recommendation model;

training the multi-view social recommendation model; and

generating recommendations for the user using a trained multi-view social recommendation model.

2. The method of claim 1, wherein generating the multi-view social model comprises:

generating a plurality of single-view social models, each single-view social model corresponding to a social relationship type;

combining the plurality of single-view social models to arrive at a multi-view social model.

3. The method of claim 2, wherein the plurality of single-view social models correspond to a plurality of social networking services.

4. The method of claim 2, wherein each of the plurality of single-view social models corresponds to a different social networking service.

5. The method of claim 1, wherein the plurality of social relationship types comprise two or more of a friendship, an affinity, a correspondence, and a transfer relationship.

6. The method of claim 2, wherein combining the plurality of single-view social models comprises summing, weighted summing, averaging, or performing a pooling operation on objective functions of the plurality of single-view social models.

7. The method of claim 1, wherein combining the plurality of single-view social models comprises determining a combining function for combining the plurality of single-view social models.

8. The method of claim 1, wherein generating a recommendation for the user using a trained multi-view social recommendation model comprises:

acquiring user information of the user, wherein the user information comprises known scoring information of the user on an article and social information of a plurality of social relation types between the user and the user;

determining a score for a plurality of candidate items using a trained multi-view social recommendation model based on user information of the user; and

recommending items to the user based on the scores of the plurality of candidate items.

9. The method of claim 1, wherein the recommendation model is a matrix decomposition model.

10. The method of claim 1, wherein training the multi-view social recommendation model comprises:

iteratively solving an objective function of the multi-view social recommendation model using a gradient descent method to determine parameters of the multi-view social recommendation model.

11. The method of claim 10, wherein generating a recommendation for the user using a trained multi-view social recommendation model comprises:

a score for the candidate item is determined using the determined parameters.

12. The method of claim 1, wherein generating recommendations for the user comprises generating recommended goods or recommended content for the user.

13. A system for generating recommendations for a user, the system comprising:

the system comprises a user information acquisition module, a user information acquisition module and a user information processing module, wherein the user information acquisition module is used for acquiring user information of a user, and the user information comprises known scoring information of the user on an article and social information of a plurality of social relationship types between the user and the user;

a multi-view social recommendation module to determine scores for a plurality of candidate items based on user information of the user, the multi-view social recommendation module comprising a recommendation module and a multi-view social module, wherein:

the recommendation module implements a recommendation model for recommending items for a user,

the multi-view social recommendation module includes a plurality of single-view social modules that implement a plurality of single-view social models corresponding to a plurality of social relationship types, and a combining module to combine the plurality of single-view social modules into the multi-view social module.

14. A computer-readable storage medium storing instructions that, when executed by a computer, cause the computer to perform the method of any of claims 1-12.

15. A system comprising means for performing the method of any one of claims 1-12.

Images