Method for commodity recommendation based on multi-granularity attribute set cooperative neighbor attention

Abstract

The application belongs to the technical field of e-commerce commodity cold start recommendation, and discloses a commodity recommendation method of multi-granularity attribute set cooperative neighbor attention, first, a user-user similarity relation graph is constructed by using the multi-granularity attribute set of the user, a commodity-commodity similarity relation graph is constructed by using the multi-granularity attribute set of the commodity, and the cooperative neighbor embedding representation of the user and the commodity on the multi-granularity attribute set is learned through a graph neural network; then the cooperative neighbor embedding representation on the multi-granularity attribute set is fused through an attention mechanism, and the user embedding representation and the commodity embedding representation are modeled; the score of the user to the commodity is calculated through inner product, the preference of the user attribute to the commodity attribute is learned, and a ranked commodity recommendation list is generated according to the score size. The application mines the personalized interest of the user from the multi-granularity attribute combination of the user and the multi-granularity attribute combination of the commodity, realizes the commodity recommendation of the interactive behavior cold start, and provides great support especially in the aspects of e-commerce new user and new commodity recommendation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of e-commerce commodity cold start recommendation, and particularly relates to a commodity recommendation method based on multi-granularity attribute set collaborative neighbor attention. BACKGROUND

[0002] In an e-commerce commodity recommendation system, users and commodities often have multiple attributes, and these attributes have a large amount of auxiliary information, which can depict the potential characteristics of users and commodities in different attribute spaces. For example, a user has characteristics such as "gender", "age", "occupation", and "region", and a commodity has characteristics such as "color", "type", "price", and "place of origin". Based on the attributes of the user, the rating preferences of the user can be mined, and based on the attributes of the commodity, the potential market value of the commodity can be mined. Different types of attribute sets of the user can construct different types of similar user groups, and different preference predictions of the user can be achieved. For example, users with the same "gender" are a group; users with the same "gender" and "age" are another group. Different types of attribute sets of the commodity can construct different types of similar commodity combinations, and collaborative combination recommendations of multiple types of commodities can be achieved. For example, a type of commodities with the same "color" can be sold in combination; another type of commodities with the same "color" and "type" can also be sold in combination. Therefore, based on different types of attribute set combinations, different user-user and commodity-commodity relationship graphs can be constructed, and different granularity neighbors of users and commodities can be formed. The near-neighbor users and near-neighbor commodities based on different granularity attribute views can model different types of characteristics of the user and the commodity, and further mine the rating preference interest of the user for the commodity from the user feature combination of multiple granularity attributes and the commodity feature combination of multiple granularity attributes.

[0003] In most recommendation systems, the interaction relationship between users and items is often used to model the vector representation of users and items, find similar users, and recommend similar products. In a real recommendation scenario, due to the addition of new users and new commodities, the historical data of the interaction is less, and the representation of the user and the item is difficult to obtain fully, so that the collaborative recommendation system still faces the recommendation problems of cold start and data sparseness, and the prediction accuracy of rating recommendation is low. Therefore, how to effectively utilize the rich attribute information of different types of users and commodities, realize feature modeling of users and commodities, learn rating prediction of users for commodities, and realize accurate e-commerce product recommendation has very important value, especially in the face of user and commodity cold start and data sparseness problems, which can improve the performance of recommendation. SUMMARY

[0004] In view of the problem that the existing e-commerce commodity recommendation system has poor recommendation accuracy due to the problems of cold start and data sparseness, the application provides a commodity recommendation method based on multi-granularity attribute set collaborative neighbor attention.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The present invention provides a product recommendation method based on multi-granularity attribute sets and neighbor attention, comprising the following steps:

[0007] Step S1: construct a user-user similarity graph based on a single attribute, and learn the collaborative neighbor embedding representation of the user on a single attribute based on a graph neural network;

[0008] Step S2: construct a user-user similarity graph based on the attribute subset, and learn the collaborative neighbor embedding representation of users on the attribute subset based on the graph neural network;

[0009] Step S3: construct a user-user similarity graph based on the full set of attributes, and learn the collaborative neighbor embedding representation of users on the full set of attributes based on the graph neural network;

[0010] Step S4: fuse the collaborative neighbor embedding representations on the multi-granularity attribute sets through the attention mechanism to model the user embedding representation;

[0011] Step S5: Based on steps S1-S4, similarly fuse the collaborative neighbor embedding representations on the multi-granularity attribute sets of the product to model the product embedding representation;

[0012] Step S6, calculating the user's rating of the product by inner product calculation based on the user's feature representation and the product's feature representation;

[0013] Step S7: Calculate the new user's rating of the new product, sort them according to the rating, generate a recommended product list, and complete the product recommendation.

[0014] Furthermore, in step S1, a user-user similarity graph is constructed based on a single attribute, and a collaborative neighbor embedding representation of the user on a single attribute is learned based on a graph neural network. The specific steps are as follows:

[0015] Step 1.1: Users with the same attribute value on a single attribute can be considered as neighbor users under the single attribute view. Given attribute a, if user u and user u' have the same value on attribute a, denoted as f a (u) = f a (u'), then the neighbor set of user u on attribute a is defined as:

[0016]

[0017] In formula (1) represents the neighbor set of user u on attribute a, and ||a|| represents the number of attribute a.

[0018] Then, based on the user neighbor relationships on attribute a, we build a user-user attribute relationship matrix U×U. This matrix reflects users' coarse-grained collaborative interests and preferences, which influences their decision-making. In the user-user attribute relationship matrix U×U, an element value of 1 indicates that users have the same value for attribute a.

[0019] Step 1.2: The collaborative attribute interests between users can be represented by neighbor information. The collaborative neighbor interaction information between user u and user u' on attribute a is defined as:

[0020]

[0021] In formula (2) represents the neighbor interaction information between user u and user u' on attribute a, u represents the embedded representation of user u, and u' represents the embedded representation of user u'. represents the neighbor set of user u on attribute a, Represents the set of neighbors of user u' on attribute a.

[0022] In step 1.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows:

[0023]

[0024] In formula (3), Represents the neighbor interaction information between user u and user u' on attribute a. Represents the collaborative neighbor embedding representation of user u on a single attribute user-user similarity graph.

[0025] Furthermore, in step S2, a user-user similarity graph is constructed based on the attribute subset, and the collaborative neighbor embedding representation of the user on the attribute subset is learned based on the graph neural network. The specific steps are as follows:

[0026] Step 2.1 Obtain the neighbor set of user u on attribute A' Then, based on the user neighbor relationships on attribute A', a user-user attribute relationship matrix U×U is established, where the element value is A', indicating that the user and user have the same value on attribute A'. This constructs a user-user similarity relationship graph for attribute subsets.

[0027] Step 2.2: Define the collaborative neighbor interaction information between user u and user u' on attribute A' as:

[0028]

[0029] In formula (4) represents the neighbor interaction information between user u and user u' on attribute A', u represents the embedded representation of user u, and u' represents the embedded representation of user u'. represents the neighbor set of user u on attribute A', Represents the set of neighbors of user u' on attribute A'.

[0030] In step 2.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows:

[0031]

[0032] In formula (5), Represents the neighbor interaction information between user u and user u' on attribute A'. Represents the collaborative neighbor embedding representation of user u on the attribute subset user-user similarity graph.

[0033] Furthermore, in step S3, a user-user similarity graph is constructed based on the full set of attributes, and the collaborative neighbor embedding representation of the user on the full set of attributes is learned based on the graph neural network. The specific steps are as follows:

[0034] Step 3.1, obtain the neighbor set of user u on attribute A Then, through the user neighbor relationship on attribute A, a user-user attribute relationship matrix U×U is established, where the element value is A, indicating that users have the same value on attribute A. In this way, a user-user similarity relationship graph of the entire attribute set is constructed.

[0035] Step 3.2: Define the collaborative neighbor interaction information between user u and user u' on attribute A as:

[0036]

[0037] In formula (6) represents the neighbor interaction information between user u and user u' on attribute A, u represents the embedded representation of user u, and u' represents the embedded representation of user u'. represents the neighbor set of user u on attribute A, Represents the neighbor set of user u' on attribute A.

[0038] In step 3.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows:

[0039]

[0040] In formula (7), Represents the neighbor interaction information on attribute A between user u and user u'. Co-neighbor embedding representation of user u on the attribute full set user-user similarity graph.

[0041] Further, the co-neighbor embedding representation of the user on the multi-granularity attribute set is fused by the attention mechanism in the step S4, and the user embedding representation is modeled, and the specific steps are as follows:

[0042] Step 4.1, based on the attribute set (a, A',..., A) of the user, a plurality of co-neighbor embedding representations of the user can be obtained

[0043] Step 4.2, in the attribute a view, the co-neighbor embedding representation is calculated The output feature score is:

[0044]

[0045] In formula (8), W h , h is the weight parameter of the neural network;

[0046] Step 4.3, the attention weight of the co-neighbor embedding representation is defined as:

[0047]

[0048] In formula (9), a is the co-neighbor embedding normalized attention weight of the user on the attribute a view, and exp is an exponential function;

[0049] Step 4.4, the co-neighbor embedding of the user on all attribute views is fused by using the attention mechanism, and the user embedding representation is defined as follows:

[0050]

[0051] In formula (10), a is the contribution degree of the co-neighbor embedding on different attribute views to the user embedding representation, and i is the attribute subset of the user.

[0052] Further, the co-neighbor embedding representation of the user on the multi-granularity attribute set is fused by the attention mechanism in the step S4, and the user embedding representation is modeled, and the specific steps are as follows:

[0053] Step 5.1, in the same way as step S1, the commodity representation of commodity v on a single attribute commodity-commodity similarity view is obtained b represents a single attribute.

[0054] Step 5.2, in the same way as step S2, the commodity representation of commodity v on the commodity-commodity similarity view of the attribute subset is obtained B' represents an attribute in the attribute subset.

[0055] Step 5.3, in the same way as step S3, obtain the commodity representation of commodity v on the commodity-commodity similarity relation view of the attribute full set B represents an attribute in the attribute full set.

[0056] Step 5.4, in the same way as step S4, obtain the final embedding representation e of the commodity v .

[0057] Further, in the step S6, the score of the user to the commodity is calculated by inner product according to the feature representation of the user and the feature representation of the commodity, and the specific steps are as follows:

[0058] We perform inner product to estimate the score of the user to the commodity. The specific score formula is as follows:

[0059] (11)

[0060] In formula (11), represents the final predicted score of the user u to the commodity v. b g represents a global bias item, b u represents a user bias item, b v represents a project bias item.

[0061] Further, in the step S7, the score of the new user to the new commodity is calculated, the recommended commodity list is generated by sorting according to the score, and the commodity recommendation is completed, and the specific steps are as follows:

[0062] For a new user u n , in the same way as steps S1-S3, the embedding representation of the new user u n is obtained. For a new commodity v m , in the same way as step S4, the embedding representation of the new commodity v m is obtained. According to formula (11), the final predicted score y of the new user u n to the new commodity v m is obtained. The commodities in the candidate set are sorted according to the score, a recommended commodity list is generated, and the commodity recommendation is completed.

[0063] The application further provides a computer readable storage medium, wherein the medium stores a computer program, and the computer program is executed to realize the multi-granularity attribute set collaborative neighbor attention commodity recommendation method.

[0064] The present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned product recommendation method based on multi-granularity attribute sets and coordinated neighbor attention is implemented.

[0065] Compared with the prior art, the present invention has the following advantages:

[0066] The method proposed in this paper distinguishes itself from existing approaches by designing a semantic embedding of users and products by mining their potential preferences through different multi-granularity attribute sets. This model then establishes a rating prediction model based on the semantic representation of these multi-granularity attribute sets. This method not only effectively improves recommendation accuracy but also alleviates the problem of poor recommendation performance in sparse scenarios, enabling product recommendations during cold-start interactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a schematic diagram of the overall model architecture of the present invention;

[0068] Figure 2 This figure compares the present invention with other graph neural network methods under different sparsity conditions on the ML-100kr dataset. DETAILED DESCRIPTION

[0069] The product recommendation method based on multi-granularity attribute sets and neighbor attention of the present invention is implemented by a computer program. The specific implementation of the technical solution proposed by the present invention will be described in detail below according to the process.

[0070] like Figure 1 As shown, the product recommendation method based on multi-granularity attribute sets and neighbor attention of the present invention includes the following steps:

[0071] Step S1: Build a user-user similarity graph based on a single attribute and learn the collaborative neighbor embedding representation of the user on a single attribute based on a graph neural network. The specific steps are as follows:

[0072] Step 1.1: Users with the same attribute value on a single attribute can be considered as neighbor users under a single attribute view. Given attribute a, if user u and user u' have the same value on attribute a, denoted as f a (u) = f a (u'), then the neighbor set of user u on attribute a is defined as:

[0073]

[0074] In formula (1) represents the neighbor set of user u on attribute a, and ||a|| represents the number of attribute a.

[0075] Further, through the user neighbor relationship on attribute a, a user-user attribute relationship matrix UxU is established. The user-user attribute relationship matrix on attribute a reflects the coarse-grained collaborative interest preference of the user, which has a certain influence on the decision of the user. In the user-user attribute relationship matrix UxU, the element value is 1, indicating that the user and the user have the same value on attribute a.

[0076] Step 1.2, the collaborative attribute interest between users can be represented by neighbor information, and the collaborative neighbor interaction information between user u and user u' on attribute a is defined as:

[0077]

[0078] In formula (2) represents the neighbor interaction information between user u and user u' on attribute a, u represents the embedding representation of user u, and u' represents the embedding representation of user u'. represents the neighbor set of user u on attribute a, represents the neighbor set of user u' on attribute a.

[0079] Step 1.3, considering all similar neighbors, the collaborative user attribute representation of the collaborative neighbor is defined based on the graph neural network as:

[0080]

[0081] In formula (3) represents the neighbor interaction information between user u and user u' on attribute a. represents the collaborative neighbor embedding representation of user u on the single attribute user-user similarity graph.

[0082] Step S2, according to the attribute subset, a user-user similarity graph is constructed, and a collaborative neighbor embedding representation of the user on the attribute subset is learned based on the graph neural network, and the specific steps are as follows:

[0083] Step 2.1, the neighbor set of user u on attribute A' is obtained Further, through the user neighbor relationship on attribute A', a user-user attribute relationship matrix UxU is established, and the element value is A', indicating that the user and the user have the same value on attribute A'. Thus, an attribute subset user-user similarity graph is constructed.

[0084] Step 2.2, the collaborative neighbor interaction information between user u and user u' on attribute A' is defined as:

[0085]

[0086] In formula (4) represents the neighbor interaction information between user u and user u' on attribute A', u represents the embedded representation of user u, and u' represents the embedded representation of user u'. represents the neighbor set of user u on attribute A', Represents the set of neighbors of user u' on attribute A'.

[0087] In step 2.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows:

[0088]

[0089] In formula (5), Represents the neighbor interaction information between user u and user u' on attribute A'. Represents the collaborative neighbor embedding representation of user u on the attribute subset user-user similarity graph.

[0090] Step S3: Build a user-user similarity graph based on the full set of attributes, and learn the collaborative neighbor embedding representation of users on the full set of attributes based on the graph neural network. The specific steps are as follows:

[0091] Step 3.1, obtain the neighbor set of user u on attribute A Then, through the user neighbor relationship on attribute A, a user-user attribute relationship matrix U×U is established, where the element value is A, indicating that users have the same value on attribute A. In this way, a user-user similarity relationship graph of the entire attribute set is constructed.

[0092] Step 3.2: Define the collaborative neighbor interaction information between user u and user u' on attribute A as:

[0093]

[0094] In formula (6) represents the neighbor interaction information between user u and user u' on attribute A, u represents the embedded representation of user u, and u' represents the embedded representation of user u'. represents the neighbor set of user u on attribute A, Represents the neighbor set of user u' on attribute A.

[0095] In step 3.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows:

[0096]

[0097] In formula (7), Represents the neighbor interaction information on attribute A between user u and user u'. Co-neighbor embedding representation of user u on the attribute full set user-user similarity graph.

[0098] Step S4, the co-neighbor embedding representation on the multi-granularity attribute set is fused through the attention mechanism to model the user embedding representation, and the specific steps are as follows:

[0099] Step 4.1, based on the attribute set (a, A',..., A) of the user, we can obtain multiple co-neighbor embedding representations of the user

[0100] Step 4.2, in the attribute a view, the co-neighbor embedding representation is calculated The output feature score is:

[0101]

[0102] In formula (8), W h , h is the weight parameter of the neural network;

[0103] Step 4.3, the attention weight of the co-neighbor embedding representation is defined as:

[0104]

[0105] In formula (9), is the normalized attention weight of the co-neighbor embedding of the user in the attribute a view, and exp is the exponential function;

[0106] Step 4.4, the co-neighbor embedding of the user in all attribute views is fused using the attention mechanism, and the user embedding representation is defined as follows:

[0107]

[0108] In formula (10), is the contribution degree of the co-neighbor embedding on different attribute views to the user embedding representation, and i is the attribute subset of the user.

[0109] Step S5, according to steps S1-S4, the co-neighbor embedding representation on the multi-granularity attribute set of the commodity is also fused to model the commodity embedding representation, and the specific steps are as follows:

[0110] Step 5.1, in the same way as step S1, the commodity representation of commodity v on a single attribute commodity-commodity similarity view is obtained b represents a single attribute.

[0111] Step 5.2, in the same way as step S2, the commodity representation of commodity v on the commodity-commodity similarity view of the attribute subset is obtained B' represents the attributes in the attribute subset.

[0112] Step 5.3: In the same way as step S3, obtain the product representation of product v on the attribute set product-product similarity relationship view B represents an attribute in the entire attribute set.

[0113] Step 5.4: Use the same method as step S4 to obtain the final embedding representation of the product e v .

[0114] Step S6: Based on the user's feature representation and the product's feature representation, the user's rating of the product is calculated by inner product, and a rating prediction model is constructed. The specific steps are as follows:

[0115] We perform inner product to estimate the user's rating of the product. The specific rating formula is as follows:

[0116] (11)

[0117] In formula (11), Indicates the final predicted rating of user u for product v. b g Represents the global bias item, b u represents the user's bias term, b v Represents the bias term of the project.

[0118] Step S7: Calculate the new user's rating of the new product, sort it by rating, and generate a list of recommended products. The specific steps are as follows:

[0119] For new users n , the same method as steps S1-S3, obtain the new user u n Embedded representation of For new products m , the same method as step S4, obtain the new product v m Embedded representation of According to formula (11), get the new user u n For new products m The final predicted score y is obtained. The products in the candidate set are sorted according to their scores, and a list of recommended products is generated to complete the product recommendation.

[0120] To verify the effectiveness of our method, we conducted experiments on the ML-100kr dataset (https: / / grouplens.org / datasets / movielens / ). The dataset information is shown in Table 1:

[0121] Table 1 Dataset

[0122]

[0123] The experiment was conducted on a dataset split into a training set, a validation set, and a test set in an 8:1:1 ratio. RMSE was used as the evaluation metric. To verify the effectiveness and advancement of the proposed technical solution, several existing rating recommendation prediction model methods were selected for comparison: GCN, NGCF, GAT, and LightGCN. The experimental results are shown in Table 2:

[0124] Table 2 Experimental results

[0125]

[0126] It can be seen from the results in Table 2 that the technical solution of the present invention can obtain detection results with better accuracy and reliability than existing methods when predicting user ratings of products.

[0127] At the same time, the present invention randomly blocks the scoring labels, thereby setting different sparsity ratios for the data, and observing the performance of the RMSE index of the present invention and the corresponding graph neural network on the processed data set. The experimental results are as follows Figure 2 As shown. Figure 2 As data sparsity increases, the RMSE of several methods increases significantly, indicating that the accuracy of user rating predictions decreases, demonstrating that sparsity can affect the quality of user and item embeddings. Furthermore, it can be found that the proposed method achieves lower RMSE than the corresponding graph neural network method under varying data sparsity conditions, demonstrating superior rating prediction performance.

[0128] Example 2

[0129] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the product recommendation method based on multi-granularity attribute sets and coordinated neighbor attention of the above-mentioned embodiment 1 is implemented.

[0130] Example 3

[0131] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the product recommendation method based on multi-granularity attribute sets and coordinated neighbor attention of the above-mentioned embodiment 1 is implemented.

[0132] The above embodiments are preferred implementations of the present invention, but the implementation of the present invention is not limited to the above embodiments. For ordinary technicians in this field, several modifications and improvements can be made without departing from the principles of the present invention, which are all included in the scope of protection of the present invention.

[0133] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A product recommendation method based on multi-granularity attribute sets and neighbor attention, characterized by: The following steps are involved: Step S1: construct a user-user similarity graph based on a single attribute, and learn the collaborative neighbor embedding representation of the user on a single attribute based on a graph neural network; Step S2: construct a user-user similarity graph based on the attribute subset, and learn the collaborative neighbor embedding representation of users on the attribute subset based on the graph neural network; Step S3: construct a user-user similarity graph based on the full set of attributes, and learn the collaborative neighbor embedding representation of users on the full set of attributes based on the graph neural network; Step S4: fuse the collaborative neighbor embedding representations on the multi-granularity attribute sets through the attention mechanism to model the user embedding representation; In step S4, the specific steps of modeling the user embedding representation by fusing the collaborative neighbor embedding representations on the multi-granularity attribute sets through the attention mechanism are as follows: Step 4.1, based on the user's attribute set ( , ,…, ), we can get multiple collaborative neighbor embedding representations of the user ( , ,…, ); Step 4.2, in Properties On the view, calculate the collaborative neighbor embedding representation The output feature score is: （8） In formula (8), 、 is the weight parameter of the neural network; Step 4.3, define collaborative neighbor embedding representation The attention weight is: （9） In formula (9), For users in properties Co-neighbor embedding normalized attention weights on views, is an exponential function; In step 4.4, the attention mechanism is used to fuse the user's collaborative neighbor embeddings on all attribute views, and the user embedding representation is defined as follows: （10） In formula (10), is the contribution of collaborative neighbor embedding to user embedding representation on different attribute views, is a subset of the user's attributes; Step S5: Based on steps S1-S4, similarly fuse the collaborative neighbor embedding representations on the multi-granularity attribute sets of the product to model the product embedding representation; Step S6: Calculate the user's rating of the product through the inner product based on the user's embedded representation and the product's embedded representation; Step S7: Calculate the new user's rating of the new product, sort it by rating, generate a list of recommended products, and complete the product recommendation; In step S7, the new user's rating of the new product is calculated, and the products are sorted according to the rating to generate a list of recommended products. The specific steps to complete the product recommendation are as follows: For New Users , the same method as steps S1-S3, get new users Embedded representation of For new products , the same method as step S4, obtain new products Embedded representation of According to the scoring formula , acquire new users For new products Final prediction score ; Sort the products in the candidate set according to their scores, generate a list of recommended products, and complete the product recommendation.

2. The product recommendation method based on multi-granularity attribute sets and neighbor attention according to claim 1 is characterized by: In step S1, the user-user similarity graph is constructed based on a single attribute, and the specific steps of learning the collaborative neighbor embedding representation of the user on a single attribute based on the graph neural network are as follows: Step 1.1: Users with the same attribute value on a single attribute can be considered as neighbor users under a single attribute view. If the user and users In the properties The values ​​on are the same, denoted as , then define the user In the properties The neighbor set on is: （1） In formula (1) Represents a user In the properties The neighbor set on Representation attributes The number of Then through the attributes User neighbor relationships on the network, establish a user-user attribute relationship matrix ; property The user-user attribute relationship matrix on reflects the user's coarse-grained collaborative interest preference and has a certain impact on the user's decision-making; the user-user attribute relationship matrix In the example, the element value is , indicating that the user and the user have The values ​​on are the same; Step 1.2: The collaborative attribute interests between users can be represented by neighbor information, defining user and users Between attributes The collaborative neighbor interaction information on is: （2） In formula (2) Represents a user and users Between attributes Neighbor interaction information on Representative User The embedding representation of Representative User Embedded representation of Represents a user In the properties The neighbor set on Represents a user In the properties The set of neighbors on ; In step 1.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows: （3） In formula (3), Represents a user and users Between attributes Neighbor interaction information on Represents a user Collaborative neighbor embedding representation on single attribute user-user similarity graph.

3. The product recommendation method based on multi-granularity attribute sets and neighbor attention according to claim 1 is characterized by: In step S2, the user-user similarity graph is constructed based on the attribute subset, and the specific steps of learning the collaborative neighbor embedding representation of the user on the attribute subset based on the graph neural network are as follows: Step 2.1, get the user In the properties Neighborhood Set , and then through the attributes User neighbor relationships on the network, establish a user-user attribute relationship matrix , the element value is , indicating that the user and the user have The values ​​on are the same, thus constructing the attribute subset user-user similarity relationship graph; Step 2.2, define users and users Between attributes The collaborative neighbor interaction information on is: （4） In formula (4) Represents a user and users Between attributes Neighbor interaction information on Representative User The embedding representation of Representative User Embedded representation of Represents a user In the properties The set of neighbors on ; Represents a user In the properties The set of neighbors on ; In step 2.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows: （5） In formula (5), Represents a user and users Between attributes Neighbor interaction information on Represents a user Collaborative neighbor embedding representation on attribute subset user-user similarity graph.

4. The product recommendation method based on multi-granularity attribute sets and neighbor attention according to claim 1 is characterized by: In step S3, a user-user similarity graph is constructed based on the full set of attributes, and the specific steps of learning the collaborative neighbor embedding representation of the user on the full set of attributes based on the graph neural network are as follows: Step 3.1, get the user In the properties Neighborhood Set , and then through the attributes User neighbor relationships on the network, establish a user-user attribute relationship matrix , the element value is , indicating that the user and the user have The values ​​on are the same; thus constructing a user-user similarity relationship graph of the entire attribute set; Step 3.2, define users and users Between attributes The collaborative neighbor interaction information on is: （6） In formula (6) Represents a user and users Between attributes Neighbor interaction information on Representative User The embedding representation of Representative User Embedded representation of Represents a user In the properties The neighbor set on Represents a user In the properties The set of neighbors on ; In step 3.3, considering all similar neighbors, the collaborative user attributes of collaborative neighbors are defined based on the graph neural network as follows: （7） In formula (7), Represents a user and users Between attributes Neighbor interaction information on Represents a user Collaborative neighbor embedding representation on user-user similarity graph over the entire attribute set.

5. The product recommendation method based on multi-granularity attribute sets and neighbor attention according to claim 1 is characterized by: In step S5, based on steps S1-S4, the collaborative neighbor embedding representations on the multi-granularity attribute sets of the product are similarly fused. The specific steps for modeling the product embedding representation are: Step 5.1: Get the product in the same way as step S1 Product representation on a single attribute product-product similarity relationship view , Represents a single attribute; Step 5.2: Get the product in the same way as step S2 Product representation on attribute subset product-product similarity relationship view , Represents an attribute in an attribute subset; Step 5.3, same method as step S3, get the product Product representation on the attribute set product-product similarity relationship view , Represents an attribute in the entire attribute set; Step 5.4: Use the same method as step S4 to obtain the final embedding representation of the product. .

6. The product recommendation method based on multi-granularity attribute sets and neighbor attention according to claim 1, characterized in that: In step S6, the specific steps of calculating the user's rating of the product through the inner product based on the user's embedded representation and the product's embedded representation are as follows: The inner product is used to estimate the user's rating of the product; the specific rating formula is as follows: （11） In formula (11), Represents a user For products The final prediction score, Represents the global bias item, represents the user's bias term, Represents the bias term of the project.

7. A computer-readable storage medium, characterized in that: The medium stores a computer program, which, when executed, implements the product recommendation method based on multi-granularity attribute sets and coordinated neighbor attention as described in any one of claims 1 to 6.

8. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for recommending products based on multi-granularity attribute sets and coordinated neighbor attention is implemented as described in any one of claims 1 to 6.

Images