CN104915391A - Article recommendation method based on trust relationship - Google Patents

Abstract

The present invention provides a method for recommending items based on a trust relationship, which includes the following steps: establishing a user model and an item model, establishing an old user-item scoring model, and a user-user trust model, and determining the recommended user; according to the user-user trust model, The initial trust model is defined as direct trust; for old users, the similarity between the recommended user and other old users is calculated according to the old user-item scoring model, and the most similar user to the recommended user is obtained by combining the trust relationship of the recommended user For new users, calculate the most similar user set of the recommended user based on the trust relationship between new users and old users; then use the collaborative filtering algorithm to predict item ratings on the most similar user set, and use high-scoring items as recommendations. The invention can solve the data sparsity problem and the user cold start problem in the recommendation system.

Description

An item recommendation method based on trust relationship

technical field

The invention belongs to the field of artificial intelligence, and in particular relates to an item recommendation method based on a trust relationship.

Background technique

With the development of information technology and the Internet, various data such as user information and product information have grown exponentially, and "information overload" has appeared. Both information consumers and information producers have encountered great challenges: as information consumers , How to find the information you are interested in from a large amount of information is a very difficult thing; as an information producer, how to make the information you produce stand out and attract the attention of the majority of users is also a very difficult thing. The recommended method is an important tool to solve this contradiction.

At present, almost all large and medium-sized websites, such as Facebook, Amazon, Netflix, Taobao, Douban, etc., have used various forms of recommendation methods to varying degrees and achieved remarkable results. Recommendation methods have become a very important technology in e-commerce and social networks, and have also produced huge economic benefits. Recommendation methods have been greatly developed in both theory and practice.

At present, various recommendation algorithms have emerged, and the main methods are content-based recommendation and collaborative filtering recommendation.

Content-based recommendation is to construct a personal interest graph and document based on the user's historical behavior (such as browsing records, evaluation of items, etc.), calculate the similarity between the recommended item and the user's interest document, and recommend the item with high similarity to the user. For example, on the Pandora music website, its musicians and researchers personally listened to tens of thousands of songs from different singers, and then marked the different characteristics of the songs (such as melody, rhythm, arrangement and lyrics, etc.), and then Pandora will The annotated genes calculate the similarity of the songs and recommend other music to the user that is genetically similar to the music he previously liked. Obviously, this method relies on the accurate description of project attributes, requires a large amount of collection of project attribute information, has low performance, and usually has problems of cold start and scalability.

Another mainstream method is collaborative filtering recommendation. It was first used in the mail method Tapestry. Collaborative filtering algorithms are also divided into user-based collaborative filtering and item-based collaborative filtering. The user-based collaborative filtering algorithm recommends to the user other items that are similar to his interests, and the item-based collaborative filtering algorithm recommends to the user items that are similar to the items he liked before. For example, the user-based collaborative filtering algorithm only needs to calculate the similarity between users based on the user's rating information on items in the past, and then can use those users who are highly similar to the target user (recommended user) to like recommend things to target users. Similarly, the item-based collaborative filtering algorithm can calculate the similarity between items according to the users who like the items, and then make recommendations.

The collaborative filtering algorithm makes up for the shortcomings of the content-based recommendation algorithm. With its simple and efficient characteristics, it is widely used in many fields. However, this method only relies on the user's historical rating information and does not use other data. There are also problems such as cold start and data sparseness.

In recent years, researchers at home and abroad have done a lot of research on trust-based computing and personalized recommendation, mainly using trust relationships to improve the performance of personalized recommendation, which alleviates the data sparsity problem of recommendation to a certain extent, and has achieved certain results. Research results. However, the existing methods mainly focus on the calculation and reasoning of users' explicit trust relationship, and many valuable implicit trust relationships are often ignored, and only the trust relationship on the current recommendation platform is considered.

Therefore, there is a need for a recommendation method that more effectively addresses the problems of cold-start users and data sparsity.

Contents of the invention

The present invention makes improvements to the problems existing in the above-mentioned prior art, that is, the technical problem to be solved by the present invention is to provide an item recommendation method based on a trust relationship, which can solve the data sparsity problem and the cold-start user problem in the recommendation system.

In order to solve the problems of the technologies described above, the present invention provides the following technical solutions:

An item recommendation method based on a trust relationship, comprising the following steps:

S1. Establish user model and item model, establish old user-item scoring model, user-user trust model, and determine recommended users;

S2. According to the user-user trust model, the initial trust model is defined as direct trust; the system defines that if user A directly trusts user B, user B directly trusts user C, and user A has no direct trust relationship with user C in the initial state, then define User A indirectly trusts user C, which is 1-step transfer; and so on, 2-step transfer and 3-step transfer, and trust decay is added in the transfer process; all trust relationships of recommended users are calculated;

S3. For old users, calculate the similarity between the recommended user and other old users according to the old user-item scoring model, and combine the trust relationship of the recommended user to obtain the most similar user set of the recommended user; for new users, According to the trust relationship between the new user and the old user, the most similar user set of the recommended user is calculated; then the collaborative filtering algorithm is used to predict the item rating on the most similar user set, and the high-scoring item is recommended as a recommendation.

The establishment of the user model: according to the user history records, including the items purchased and the ratings of the items and the connection between the users to establish the user model; for new users, the system imports the connections between this user and other old users on the Internet, Build user models.

The beneficial effects of the present invention are: based on a collaborative filtering algorithm, the present invention combines user historical information and trust relationships between users, and conducts trust transfer to the trust network, discovers potential indirect trust relationships between users, and solves the problem of data sparseness. For cold-start users, the system transfers trust by importing user contacts from other websites, so as to achieve high-accuracy recommendations.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the description, and are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the attached picture:

Figure 1 is a schematic diagram of the principle of trust transfer;

Fig. 2 is a flowchart of an item recommendation system based on trust relationship.

Detailed ways

As shown in Figure 1-2, the present invention discloses a method for recommending items based on a trust relationship, which includes the following steps:

S1. Establish user model and item model, establish old user-item scoring model, user-user trust model, and determine recommended users;

S2. According to the user-user trust model, the initial trust model is defined as direct trust; the system defines that if user A directly trusts user B, user B directly trusts user C, and user A has no direct trust relationship with user C in the initial state, then define User A indirectly trusts user C, which is 1-step transfer; and so on, 2-step transfer and 3-step transfer, and trust decay is added in the transfer process; all trust relationships of recommended users are calculated;

S3. For old users, calculate the similarity between the recommended user and other old users according to the old user-item scoring model, and combine the trust relationship of the recommended user to obtain the most similar user set of the recommended user; for new users, According to the trust relationship between the new user and the old user, the most similar user set of the recommended user is calculated; then the collaborative filtering algorithm is used to predict the item rating on the most similar user set, and the high-scoring item is recommended as a recommendation.

Aiming at the currently widely used recommendation system, the present invention is based on a collaborative filtering algorithm, combines user history information and trust relationships between users, and transfers trust to a trust network to enhance the effect of recommendation. The present invention will be described in detail below in conjunction with examples.

Figure 1 is a description of the trust transfer theory of the present invention, each node represents a user, and the arrow represents a direct trust relationship, for example, user A directly trusts user B and user D, and user B directly trusts user C. According to the theory of trust transmission, user A indirectly trusts users C and E through 1-step transmission; indirectly trusts user F through 2-step transmission; and indirectly trusts user G through 3-step transmission. For trust relationships that can be transmitted through multiple paths, the shortest path is adopted: for example, user A to user C can pass through A-B-C or A-D-E-C. According to the principle of the shortest path, A-B-C is adopted, that is, one-step transmission.

And trust decay is added on the basis of trust transfer, assuming that the initial direct trust value is t, and the trust decay parameter is θ, then the indirect trust value transferred through 1 step is t*θ, and the indirect trust value transferred through 2 steps is t *θ*θ, and so on. In theory, passing 2 steps to 3 steps is the best.

Fig. 2 is a flowchart of a recommendation system based on trust relationship. The recommended method includes the following steps:

Step 1, the recommendation system starts;

Step 2, determine the recommended user;

Step 3, determine whether the recommended user has historical records (such as purchased items, rating items, etc.), if yes, go to step 4; if not, it is a new user, then go to step 6;

Step 4, import user-item rating matrix (a form of representation of user history);

Step 5, calculate the similarity between the recommended user and other users in the system according to the information imported in step 4 (Pearson similarity can be used);

Step 6, import the user-user trust matrix (a representation of the relationship between users), for old users, it is the trust relationship between old users and old users in the system; for new users, It is the trust relationship between new users and old users on other integrated platforms;

Step 7, perform trust transfer on the trust relationship imported in step 6 according to the trust transfer theory proposed above, and save the result in the trust relationship;

Step 8, calculate the trust value (including direct trust and indirect trust) of the recommended user to other users in the system;

Step 9: Integrate the similarity calculated in step 5 and the trust value calculated in step 8 to obtain the degree of association between the recommended user and other users in the system; if the recommended user does not have a history record, take step 8 alone The trust value calculates the degree of association between the recommended user and other users in the system;

Step 10, according to the degree of association calculated in step 9, the adjacent user topMatches of the recommended user is obtained (ie, the N users with the highest degree of association with the recommended user);

Step 11, according to the adjacent user topMatches in step 10, use the collaborative filtering algorithm to calculate the user's "predictive score" for the item;

Step 12, arrange the items according to the "predicted score value" from high to low;

Step 13, using the first N items as recommended items;

Step 14, end of recommendation.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still understand the foregoing embodiments The recorded technical solutions are modified, or some of the technical features are equivalently replaced. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (2)

1. based on an item recommendation method for trusting relationship, it is characterized in that, comprise the steps:

S1, set up user model and object model, set up old user-article Rating Model, user-user trust model, determine recommended user;

S2, according to user-user trust model, initial trust model is defined as direct trust; If system definition user A directly trusts user B, user B directly trusts user C, and user A does not have direct trusting relationship in original state and user C, and so define user A and indirectly trust user C, this situation is 1 step transmission; 2 step transmission and 3 step transmission by that analogy, and in transmittance process, add trust decay; Calculate all trusting relationships of recommended user;

S3, for old user, calculate the similarity of recommended user and other old users according to old user-article Rating Model, in conjunction with the trusting relationship of recommended user, calculate the most similar users collection of recommended user; For new user, calculate the most similar users collection of recommended user according to trusting relationship between new user and old user; Then on most similar users collection, utilize collaborative filtering to predict article scoring, article of being marked by height are as recommendation.

2. a kind of item recommendation method based on trusting relationship according to claim 1, is characterized in that, the foundation of described user model: according to user's historical record, comprises the article of purchase and sets up user model to the contact between the scoring of article and user; For new user, contacting on system introducing internet between this user and other old users, sets up user model.