Recommendation Method and System Based on Subspace Trust Fusion

Abstract

The present invention relates to a recommendation method and system based on subspace trust fusion, comprising the following steps: obtaining a user set and an item set, introducing a user rating matrix based on the ratings of users for items, and introducing a user relationship matrix according to whether there is a direct social relationship between users; constructing a directed graph of direct social relationships according to the user relationship matrix, and obtaining the explicit trust degree between two users based on the directed graph of direct social relationships; obtaining the implicit trust degree between two users according to the user relationship matrix and the user rating matrix; constructing a trust network based on the explicit trust degree and the implicit trust degree; and obtaining the predicted rating of a user for an item based on the trust network. It not only considers the social relationships of individual users, but also explores the implicit relationships between users according to the rating records of users, improving the performance of item recommendation.

Description

Technical Field

[0001] The present invention relates to the field of network information technology, and in particular to a recommendation method and system based on subspace trust fusion. Background Art

[0002] The development of information technology has led to an increasing number of products flooding into people's lives, providing people with diverse lifestyles, enabling them to keep abreast of current events without leaving home, make friends through social software, shop online on e-commerce platforms, and so on. However, with the massive growth of Internet information data, people need to spend more time screening the information they are interested in. When people have no clear interest preferences, the information selection process will instead put users in a dilemma, thus reducing the information usage efficiency.

[0003] Personalized recommendation systems have alleviated the problem of information overload to a certain extent. They predict users' needs based on users' personal preferences and recommend items that users may like. Since different users have different interest preferences, the recommended content for them will also vary. Traditional recommendation systems mine the correlation between users based on users' historical records. However, with the rapid development of social networks, a trust mechanism can also be introduced into the recommendation system. Trust is the core of social networks, and there is a trust relationship between people. Therefore, the recommendation results are more credible.

[0004] Although personalized recommendation systems have been widely studied and achieved good application results, problems such as data sparsity and cold start still affect the quality of recommendations. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the technical defect of poor recommendation quality in the existing personalized recommendation system.

[0006] To solve the above technical problem, the present invention provides a recommendation method based on subspace trust fusion, including the following steps:

[0007] Collect user information and item information to obtain a user set and an item set. Introduce a user rating matrix based on users' ratings of items, and introduce a user relationship matrix according to whether there is a direct social relationship between users;

[0008] Construct a directed graph of direct social relationships according to the user relationship matrix, and obtain the explicit trust degree between two users based on the directed graph of direct social relationships;

[0009] Obtain the implicit trust degree between two users according to the user relationship matrix and the user rating matrix;

[0010] Construct a trust network based on the explicit trust degree and the implicit trust degree;

[0011] Obtain the predicted score of a user for an item based on a trust network.

[0012] Preferably, obtain the explicit trust degree between two users based on the directed graph of direct social relationships, specifically including:

[0013] Obtain the explicit trust degree between two users according to the shortest path between the users:

[0014]

[0015] where d uv is the shortest path from user u u to user u v and d max represents the maximum allowable propagation distance between users;

[0016]

[0017] where k represents the average propagation path length of users in the social relationship network, obtained from the ratio of the total number of user social relationships to the total number of users, and n represents the total number of users.

[0018] Preferably, obtain the implicit trust degree between two users according to the user relationship matrix and the user score matrix, including:

[0019] Convert the user score matrix into three sub-matrices according to a score threshold, and the three sub-matrices respectively correspond to a high score subspace, an average score subspace, and a low score subspace;

[0020] Set a score threshold, and divide items into different score subspaces according to the scores of users for the items, to obtain the item set in the high score subspace, the item set in the average score subspace, and the item set in the low score subspace;

[0021] Obtain the implicit trust degrees of any two users in the high score subspace, the average score subspace, and the low score subspace respectively;

[0022] Fuse the implicit trust degrees of any two users in the high score subspace, the average score subspace, and the low score subspace to obtain the implicit trust degree of any two users.

[0023] Preferably, the setting of the score threshold, and dividing items into different score subspaces according to the scores of users for the items, to obtain the item set in the high score subspace, the item set in the average score subspace, and the item set in the low score subspace, specifically includes:

[0024] Obtain the maximum score value r in the score matrixmax and the minimum score value r min ;

[0025] If then item i j is classified into the high - score sub - space, where r uj is the score given by user u u to item i j .

[0026] If then item i j is classified into the average - score sub - space;

[0027] If then item i j is classified into the low - score sub - space.

[0028] Preferably, obtaining the implicit trust degree of any two users in the high - score sub - space specifically includes:

[0029] User u u and user u v 's implicit trust degree in the high - score sub - space where G 1 represents the set of items in the high - score sub - space, is the set of high - score items of user u u , is the set of high - score items of user u v .

[0030] Preferably, fusing the implicit trust degrees of any two users in the high - score sub - space, the implicit trust degrees in the average - score sub - space, and the implicit trust degrees in the low - score sub - space to obtain the implicit trust degree of any two users specifically includes:

[0031] The implicit trust degree of any two users u u and u v is where represents the implicit trust degree of user u u and user u v in the high - score sub - space, represents the implicit trust degree of user u u and user u v in the average - score sub - space, represents the implicit trust degree of user u u and user u v in the low - score sub - space.

[0032] Preferably, collecting user information and item information to obtain a user set and an item set, introducing a user rating matrix based on the ratings of users for items, and introducing a user relationship matrix according to whether there is a direct social relationship between users, specifically including:

[0033] Obtain a user set U = {u1, u2,..., u n} and an item set I = {i1, i2,..., i m}, where n and m respectively represent the total number of users and the total number of items;

[0034] The user rating matrix R = [r uj n×m , where r uj is the rating of user u u for item i j ;

[0035] The user relationship matrix M, with a size of n×n, and the matrix elements take values of 0 or 1, representing the user relationship value.

[0036] Preferably, constructing a trust network based on explicit trust and implicit trust, specifically including:

[0037] The trust degree of the trust network where t explicit (u u , u v ) is the explicit trust degree between user u u and user u v , and t implicit (u u , u v ) represents the implicit trust degree between users.

[0038] Preferably, obtaining the predicted rating of a user for an item based on the trust network, specifically including:

[0039] The predicted rating

[0040] where is the average rating of the target user u u , N u represents the top N nodes adjacent to the target user u u in the trust network, w uv represents the trust degree between the target user u u and the neighbor user u v , and r vj is the rating value of the neighbor user u v for item i j .

[0041] ​The present invention discloses a recommendation system based on subspace trust fusion, including:

[0042] An information collection module, which is used to collect user information and item information, obtain a user set and an item set, introduce a user rating matrix based on the ratings of users for items, and introduce a user relationship matrix according to whether there is a direct social relationship between users;

[0043] An explicit trust degree calculation module, which constructs a directed graph of direct social relationships according to the user relationship matrix and obtains the explicit trust degree between two users based on the directed graph of direct social relationships;

[0044] An implicit trust degree calculation module, which obtains the implicit trust degree between two users according to the user relationship matrix and the user rating matrix;

[0045] A trust network construction module, which constructs a trust network based on the explicit trust degree and the implicit trust degree;

[0046] A rating prediction module, which obtains the predicted rating of a user for an item based on the trust network.

[0047] The above technical solution of the present invention has the following advantages compared with the prior art:

[0048] The present invention constructs a trust network according to the explicit trust and implicit trust between users, then finds the trusted domain of the target user through the trust network, and finally makes a final prediction based on the neighbors in the trusted domain. The present invention constructs a trust network from two aspects, not only considering the personal social relationships of users, but also exploring the implicit relationships between users according to the rating records of users. Therefore, the present invention can well improve the performance of item recommendation. Description of the Drawings

[0049] Figure 1 It is a schematic flowchart of the present invention. Detailed Embodiments

[0050] The following further describes the present invention with reference to the drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the embodiments cited do not limit the present invention.

[0051] Referring to Figure 1 as shown, the present invention discloses a recommendation method based on subspace trust fusion, including the following steps:

[0052] Step 1: Collect user information and item information to obtain a user set and an item set. Introduce a user rating matrix based on the ratings of users for items, and introduce a user relationship matrix according to whether there is a direct social relationship between users. Specifically, it includes:

[0053] Obtain the user set U = {u1, u2,..., u n} and the item set I = {i1, i2,..., i m}, where n and m represent the total number of users and the total number of items respectively.

[0054] The user rating matrix R = [r uj n×m where r uj is the rating of user u u for item i j . The larger the value of r uj , the more user u u likes item i j . If the value of r uj is 0, it means that user u u has not rated item i j .

[0055] The user relationship matrix M, with a size of n×n, and the matrix elements take values of 0 or 1, representing the user relationship value. If there is a direct social relationship between users, the user relationship value is 1; otherwise, the user relationship value is 0.

[0056] Step 2: Construct a directed graph of direct social relationships based on the user relationship matrix, and obtain the explicit trust degree between two users based on the directed graph of direct social relationships. Specifically, it includes:

[0057] Use a directed graph to represent the social relationships in the user relationship matrix. The nodes in the directed graph represent users, and the edges represent the existence of direct social relationships between users. For any two users u u and u v , in the directed graph, if there is an edge from user u u to user u v , it means that user u u can spread to user u v . The shortest path between user u u and user u v can be obtained according to the number of edges passed through during the spread. If there is no edge from user u u to user u v , the shortest path is set to 0.

[0058] Obtain the explicit trust degree between two users according to the shortest path between users: ​

[0059]

[0060] Among them, d uv is the shortest path from user u u to user u v . d max represents the maximum allowable propagation distance between users. The propagation distance is the length of the shortest path between two users, which is approximately equal to the average path length between users in the social relationship network.

[0061]

[0062] Among them, k represents the average propagation path length of users in the social relationship network, which is obtained according to the ratio of the total number of user social relationships to the total number of users, and n represents the total number of users.

[0063] Step 3: Obtain the implicit trust degree between two users according to the user relationship matrix and the user rating matrix, including:

[0064] S31: Convert the user rating matrix into three sub-matrices according to the rating threshold. The three sub-matrices respectively correspond to the high-rating sub-space, the average-rating sub-space, and the low-rating sub-space. The sizes of the three sub-matrices are the same as the user rating matrix, and respectively represent the item situations of high ratings, average ratings, and low ratings of users.

[0065] S32: Set the rating threshold, and divide the items into different rating sub-spaces according to the ratings of users on the items, to obtain the item set of the high-rating sub-space, the item set of the average-rating sub-space, and the item set of the low-rating sub-space; specifically including:

[0066] Obtain the maximum rating value r max and the minimum rating value r min in the rating matrix;

[0067] If then item i j is divided into the high-rating sub-space, where r uj is the rating of user u u on item i j ;

[0068] If then item i j is divided into the average-rating sub-space;

[0069] If then item i j is divided into the low-rating sub-space.

[0070] S33. By partitioning the user rating matrix into subspaces such that the rating values in each rating subspace are similar, when measuring the implicit trust degree between users, instead of considering the specific rating values of users for items, it is calculated based on the number of items rated by users. Obtain the implicit trust degrees of any two users in the high-rating subspace, the average-rating subspace, and the low-rating subspace respectively.

[0071] Among them, obtaining the implicit trust degree of any two users in the high-rating subspace specifically includes:

[0072] User u u and user u v 's implicit trust degree in the high-rating subspace Among them, G 1 represents the set of items in the high-rating subspace, is the set of high-score items of user u u , and is the set of high-score items of user u v .

[0073] In addition, the calculation methods of implicit trust degrees in different subspaces are the same and will not be elaborated here.

[0074] S34. Fuse the implicit trust degrees of any two users in the high-rating subspace, the average-rating subspace, and the low-rating subspace to obtain the implicit trust degree of any two users, specifically including:

[0075] The implicit trust degree of any two users u u and u v where Among them, represents the implicit trust degree of user u u and user u v in the high-rating subspace, represents the implicit trust degree of user u u and user u v in the average-rating subspace, represents the implicit trust degree of user u u and user u v in the low-rating subspace.

[0076] In the present invention, steps two and three can be performed in any order.

[0077] Step four: The trust network is jointly constructed based on explicit trust and implicit trust and is represented by a weighted directed graph. In the trust network, users are nodes, and the trust degree between users is determined by the weight of the edge. Based on the explicit trust degree and the implicit trust degree, construct the trust network, specifically including: ​

[0078] Trust Degree of Trust Network Among them, t explicit (u u , u v ) is the explicit trust degree between user u u and user u v , and t implicit (u u , u v ) represents the implicit trust degree between users.

[0079] Step 5: Given any target user u u ∈U, for item i j that the target user has not rated, this module needs to predict the rating value of the target user u u for item i j , and determine whether to recommend item i j to the target user according to the size of the rating value. Obtain the predicted rating of the user for the item based on the trust network, specifically including:

[0080] Predicted Rating

[0081] Among them, is the average rating of the target user u u , N u represents the first N nodes adjacent to the target user u u in the trust network, w uv represents the trust degree between the target user u u and the neighbor user u v , r vj is the rating value of the neighbor user u v for item i j .

[0082] The present invention also discloses a recommendation system based on subspace trust fusion, including an information collection module, an explicit trust degree calculation module, an implicit trust degree calculation module, a trust network construction module, and a rating prediction module.

[0083] The information collection module is used to collect user information and item information, obtain a user set and an item set, introduce a user rating matrix based on the ratings of users for items, and introduce a user relationship matrix according to whether there is a direct social relationship between users;

[0084] The explicit trust degree calculation module constructs a directed graph of direct social relationships according to the user relationship matrix, and obtains the explicit trust degree between two users based on the directed graph of direct social relationships;

[0085] The implicit trust degree calculation module obtains the implicit trust degree between two users according to the user relationship matrix and the user rating matrix;

[0086] The trust network construction module constructs a trust network based on explicit trust degree and implicit trust degree;

[0087] The scoring prediction module obtains the predicted score of the user for the item based on the trust network.

[0088] Next, the technical solutions in the present invention will be further described in conjunction with specific embodiments.

[0089] The present invention selects the FilmTrust dataset for testing. This dataset contains 1508 users, 2071 movies, and 35497 movie rating records. The rating range of this dataset is between 0.5 and 4. The size of the rating value represents the preference degree of the user for the movie. When the rating is 0.5, it means the user does not like the movie. When the rating is 4, it means the user likes the movie very much. This dataset can form a rating matrix of size 1508 * 2071, with a sparsity of 98.8%, indicating that the user rating records are very few. If a user has not rated a certain movie, the rating value is set to 0. In order to verify the performance of the present invention, a five-fold cross-validation method is adopted. The dataset is randomly divided into five parts. Each time an experiment is conducted, one of them is selected as the test set, and the remaining four parts are used as the training set. The final result takes the average value of the five experiments.

[0090] Processing the specific data through the above scoring method, in this embodiment, N = 60.

[0091] The effect of the present invention can be verified through the following experiment: Adopting the five-fold cross-validation method, the FilmTrust dataset is randomly divided into five parts. Each time an experiment is conducted, one of them is selected as the test dataset, and the remaining four parts are used as the training dataset. The experiment is conducted five times, and the final result takes the average value of the five experiments. In the recommendation prediction, the comparison methods include the present invention, the collaborative filtering recommendation algorithm (UCF) using Pearson similarity, the recommendation algorithm based on the trust mechanism (TARS), and the trust-based collaborative filtering recommendation (UTCF).

[0092] The absolute value mean error index (MAE), the squared error index (RMSE), and the recall rate (Recall) are used to evaluate the recommendation effect:

[0093]

[0094]

[0095]

[0096] Among them, V = {(u u , i j ) | uu ∈ U, i j ∈ T}, represents the predicted value of user u u for item i j , r uj represents the true value of user u u for item i j in the test set. is the recommended list generated for user u u according to the prediction result, and is the set of items that user u u truly likes in the test set. Table 1 shows the comparison of the recommended results of three algorithms.

[0097] Table 1

[0098] Recommendation MAE RMSE Recall UCF 0.8976 1.1512 0.4011 TARS 0.6958 1.9089 0.2025 UTCF 0.7206 0.9174 0.5049 The present invention 0.6139 0.8103 0.5693

[0099] As can be seen from Table 1, the recommendation performance of the present invention is significantly better than other comparison methods.

[0100] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0102] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1The functions specified in one or more boxes.

[0103] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the steps of the functions specified in Figure 1 one process or more processes and / or boxes Figure 1 the functions specified in one box or more boxes.

[0104] Obviously, the above embodiments are only examples given for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. And the obvious changes or modifications derived therefrom still fall within the protection scope of the present invention.

Claims

1. A recommendation method based on subspace trust fusion, characterized in that, It includes the following steps: Collect user information and item information, obtain a user set and an item set, introduce a user rating matrix based on the ratings of items by users, and introduce a user relationship matrix according to whether there is a direct social relationship between users; Construct a directed graph of direct social relationships based on the user relationship matrix, and obtain the explicit trust degree between two users based on the directed graph of direct social relationships; According to the user relationship matrix and the user rating matrix, obtain the implicit trust degree between two users, including: Convert the user rating matrix into three sub-matrices according to a rating threshold, and the three sub-matrices respectively correspond to a high-rating subspace, an average-rating subspace, and a low-rating subspace; Set a rating threshold, and divide items into different rating subspaces according to the ratings of items by users, to obtain an item set in the high-rating subspace, an item set in the average-rating subspace, and an item set in the low-rating subspace, specifically including: Obtain the maximum rating value r in the rating matrix max and the minimum rating value r min ; If then item i j is assigned to the high - rated subspace, where r uj is the rating of item i u by user u j ; If then item i j is assigned to the average score subspace; If then item i j is assigned to the low-scoring subspace; Obtain the implicit trust degree between any two users in the high-rating subspace, the implicit trust degree between any two users in the average-rating subspace, and the implicit trust degree between any two users in the low-rating subspace; Fuse the implicit trust degree between any two users in the high-rating subspace, the implicit trust degree between any two users in the average-rating subspace, and the implicit trust degree between any two users in the low-rating subspace, to obtain the implicit trust degree between any two users; Construct a trust network based on the explicit trust degree and the implicit trust degree; Obtain the predicted rating of a user for an item based on the trust network.

2. The recommendation method based on subspace trust fusion according to claim 1, characterized in that Obtain the explicit trust degree between two users based on the directed graph of direct social relationships, specifically including: Obtain the explicit trust degree between two users according to the shortest path between users: Among them, d uv is the shortest path from user u u to user u v , and d max represents the maximum allowable propagation distance between users; Wherein, k represents the average propagation path length of users in the social relationship network, obtained according to the ratio of the total number of user social relationships to the total number of users, and n represents the total number of users.

3. The recommendation method based on subspace trust fusion according to claim 1, wherein Obtain the implicit trust degree between any two users in the high-rating subspace, specifically including: User u u and User u v Implicit trust degree in the polymer space Among them, G 1 represents the set of items in the polymer space, is the set of high-score items of User u u and is the set of high-score items of User u v and 4. The recommendation method based on subspace trust fusion according to claim 1, wherein The step of fusing the implicit trust degree between any two users in the high-rating subspace, the implicit trust degree between any two users in the average-rating subspace, and the implicit trust degree between any two users in the low-rating subspace, to obtain the implicit trust degree between any two users, specifically including: The implicit trust degree between any two users u u and u v is wherein, represents the implicit trust degree between user u u and user u v in the high - rating subspace, represents the implicit trust degree between user u u and user u v in the average - rating subspace, represents the implicit trust degree between user u u and user u v in the low - rating subspace.

5. The recommendation method based on subspace trust fusion according to claim 1, wherein The step of collecting user information and item information, obtaining a user set and an item set, introducing a user rating matrix based on the ratings of items by users, and introducing a user relationship matrix according to whether there is a direct social relationship between users, specifically including: Obtain a user set U = {u1, u2,..., u n} and an item set I = {i1, i2,..., i m}, where n and m respectively represent the total number of users and the total number of items; The user rating matrix R = [r uj n×m , where r uj is the rating given by user u u to item i j ;​ The user relationship matrix M, with a size of n×n, and the matrix elements take values of 0 or 1, representing the user relationship value.

6. The recommendation method based on subspace trust fusion according to claim 1, characterized in that The step of constructing a trust network based on the explicit trust degree and the implicit trust degree, specifically including: Trust Degree of Trust Network Among them, t explicit (u u , u v ) is the explicit trust degree between user u u and user u v , and t implicit (u u , u v ) represents the implicit trust degree between users.

7. The recommendation method based on subspace trust fusion according to claim 1, wherein The step of obtaining the predicted rating of a user for an item based on the trust network, specifically including: Predicted score Among them, is the average score of the target user u u , N u represents the top N nodes adjacent to the target user u u in the trust network, w uv represents the trust degree between the target user u u and the neighbor user u v , r vj is the neighbor user u v 's score for item i j .

8. A recommendation system based on subspace trust fusion, characterized in that Including: An information collection module, which is used to collect user information and item information, obtain a user set and an item set, introduce a user rating matrix based on the ratings of items by users, and introduce a user relationship matrix according to whether there is a direct social relationship between users; An explicit trust degree calculation module, which constructs a directed graph of direct social relationships according to the user relationship matrix, and obtains the explicit trust degree between two users based on the directed graph of direct social relationships; An implicit trust degree calculation module, which obtains the implicit trust degree between two users according to the user relationship matrix and the user rating matrix, including: Converting the user rating matrix into three sub-matrices according to the rating threshold, and the three sub-matrices respectively correspond to the high-rating subspace, the average-rating subspace, and the low-rating subspace; Setting the rating threshold, and dividing the items into different rating subspaces according to the ratings of the users on the items, to obtain the item set in the high-rating subspace, the item set in the average-rating subspace, and the item set in the low-rating subspace, specifically including: Obtain the maximum rating value r in the rating matrix max and the minimum rating value r min ; If then item i j is assigned to the high - rating subspace, where r uj is the rating of user u u for item i j ; If then item i j is partitioned into the average score subspace; If then item i j is assigned to the low-score subspace; Obtaining the implicit trust degrees of any two users in the high-rating subspace, the average-rating subspace, and the low-rating subspace respectively; Fusing the implicit trust degrees of any two users in the high-rating subspace, the average-rating subspace, and the low-rating subspace to obtain the implicit trust degree of any two users; A trust network construction module, which constructs a trust network based on the explicit trust degree and the implicit trust degree; A rating prediction module, which obtains the predicted rating of the user on the item based on the trust network.

Images