Commodity recommendation method based on multi-behavior meta-path semantic mapping

Abstract

The invention discloses a commodity recommendation method based on multi-behavior meta-path semantic mapping, which comprises the following steps: constructing an e-commerce heterogeneous network node graph, and extracting a user meta-path and a corresponding instance through an automatic meta-path extraction algorithm based on a combination and screening strategy; inversely arranging the extracted user element path and the corresponding instance to obtain a commodity element path and a corresponding instance; through a commodity recommendation model based on multi-behavior meta-path semantic mapping, obtaining a user target behavior semantic embedding vector, a mapped auxiliary behavior semantic embedding vector, an initial embedding vector of a commodity and an aggregated commodity meta-path embedding vector; and calculating a commodity scoring function, constructing a final loss function of the recommendation model, and finally outputting a commodity recommendation list by the commodity recommendation model through minimizing the final loss function. According to the method, the meta-paths can be automatically extracted, the high-quality meta-path instances are extracted, the potential relation between different behavior meta-paths is mined, and the recommendation accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a commodity recommendation method based on multi-behavior meta-path semantic mapping. BACKGROUND With the rapid development of Internet e-commerce, the traditional user-commodity bipartite graph is difficult to model the complex relationship of e-commerce platform, and the heterogeneous network can better adapt to the complex relationship of e-commerce scene because it supports the modeling of multiple types of nodes and edges. In the e-commerce heterogeneous network, users, commodities, categories, etc. are abstracted as nodes, and purchase, browse, collect, belong, etc. interaction relationship is abstracted as edge.

[0002] The existing commodity recommendation method based on meta-path has low recommendation accuracy, and there are three problems in obtaining user and commodity embedding according to meta-path for recommendation: first, meta-path not only depends on the predefinition of domain experts, but also manual definition of meta-path may cause the lack of meta-path, which further leads to insufficient information extracted through meta-path. Second, most of the commodity recommendation methods based on meta-path are to extract information by aggregating meta-path instances, but there are usually multiple instances for the same meta-path, and aggregating all instances will cause redundancy of node information extraction. Third, most of the commodity recommendation methods based on meta-path ignore the potential connection between different behaviors, that is, auxiliary behaviors can indirectly deduce target behaviors. For example, a user collects a commodity, then the user is very likely to purchase the commodity in the future, that is, the purchase behavior can be indirectly deduced through the collection behavior.

[0003] Therefore, there is a need for a commodity recommendation method that can automatically extract meta-paths, extract high-quality meta-path instances, mine potential connections between different behavior meta-paths, and has high accuracy. SUMMARY

[0004] In view of this, the present application provides a commodity recommendation method based on multi-behavior meta-path semantic mapping to automatically extract meta-paths, extract high-quality meta-path instances, mine potential connections between different behavior meta-paths, and improve recommendation accuracy.

[0005] A commodity recommendation method based on multi-behavior meta-path semantic mapping, comprising: Step S1, read the interaction database of users and commodities in the e-commerce platform, construct the e-commerce heterogeneous network node graph, extract the meta-path through the meta-path automatic extraction algorithm based on combination and screening strategy, take the extracted meta-path as the user meta-path, and extract the corresponding instance in the e-commerce heterogeneous network node graph according to the user meta-path; Step S2, reverse the extracted user meta-path and its corresponding instance, take the obtained new meta-path and instance as the commodity meta-path and corresponding instance, and divide the user meta-path into user target behavior meta-path and user auxiliary behavior meta-path according to the behavior type. Step S3, input the constructed e-commerce heterogeneous network node graph, user target behavior meta-path and its corresponding instance, user auxiliary behavior meta-path and its corresponding instance, and commodity meta-path and its corresponding instance into a commodity recommendation model based on multi-behavior meta-path semantic mapping, to obtain user target behavior semantic embedding vectors and mapped auxiliary behavior semantic embedding vectors, initial embedding vectors of commodities and aggregated commodity meta-path embedding vectors; Step S4, fuse the user target behavior semantic embedding vectors and the mapped auxiliary behavior semantic embedding vectors to obtain final user embedding vectors, fuse the initial embedding vectors of commodities and the aggregated commodity meta-path embedding vectors to obtain final commodity embedding vectors, obtain a commodity scoring function based on the final user embedding vectors and the final commodity embedding vectors, and construct a final loss function of the commodity recommendation model based on the commodity scoring function, and finally output a commodity recommendation list by minimizing the final loss function.

[0006] The commodity recommendation method based on multi-behavior meta-path semantic mapping provided by the application has the following beneficial effects: (1) The application does not need field experts to manually define meta-paths. Most existing commodity recommendation methods based on meta-paths manually predefine meta-paths, which not only requires prior knowledge of field experts, but also may cause the loss of meta-paths. The application proposes a meta-path automatic extraction algorithm based on combination and screening strategy, sets the length of the meta-path first, then selects nodes and edges based on the set length of the meta-path, generates an initial meta-path set through a combination strategy, and then obtains a screened meta-path set through a screening strategy. According to the meta-path, the meta-path instance is extracted in the heterogeneous network. If the instance exists, the meta-path is retained, otherwise the meta-path is removed, thereby realizing the effect of automatically extracting the meta-path.

[0007] (2) The application reduces the redundancy of node information extraction by extracting high-quality meta-path instances. Most existing commodity recommendation methods based on meta-paths extract information by aggregating meta-path instances, but there are often multiple instances for the same meta-path, and aggregating all instances will cause redundancy of node information. The method of filtering instances through the similarity between nodes has poor interpretability. The application starts from the semantic information represented by the meta-path, realizes the extraction of high-quality meta-path instances based on the cosine similarity score, can reduce the redundancy of node information, and has strong interpretability.

[0008] (3) The application can improve the accuracy of commodity recommendation by mining the potential connection between different behavior meta-paths. Existing commodity recommendation methods based on meta-paths cannot mine the potential connection between different behavior meta-paths, ignoring the inference relationship between different behaviors. The application can map the semantic of the auxiliary behavior meta-path to the semantic space of the target behavior meta-path by splicing the user auxiliary behavior initial embedding vector and the user auxiliary behavior semantic embedding vector, and then mapping through meta-learning to obtain the mapped auxiliary behavior semantic embedding vector, realizing the display modeling of the potential connection between different behavior meta-paths, and optimizing the mapped auxiliary behavior semantic embedding vector through the contrast learning loss function, which can improve the accuracy of recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 A flowchart of a commodity recommendation method based on multi-behavior meta-path semantic mapping is provided for the embodiments of the application. DETAILED DESCRIPTION

[0010] The embodiments of the application will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the embodiments of the application, and cannot be understood as a limitation of the application.

[0011] Referring to Figure 1 , the embodiments of the application provide a commodity recommendation method based on multi-behavior meta-path semantic mapping, comprising steps S1-S4: Step S1, read the interaction database of users and commodities in the e-commerce platform, construct the node graph of e-commerce heterogeneous network, extract meta-paths through the meta-path automatic extraction algorithm based on combination and screening strategy, take the extracted meta-paths as user meta-paths, and extract corresponding instances in the e-commerce heterogeneous network node graph according to the user meta-paths.

[0012] Specifically, step S1 comprises: Step S11, construct the node graph of e-commerce heterogeneous network by reading the user and commodity interaction dataset from the user and commodity interaction database of the e-commerce platform , , wherein, represents a set of nodes, The nodes in the set represent user nodes, commodity nodes, etc. represents a set of edges, and the edges in the set represent the relationship between nodes; represents a set of node types, such as user nodes, commodity nodes, commodity category nodes, etc. A set representing edge types, such as a purchase relationship between a user and a commodity, a collection relationship between a user and a commodity, a membership relationship between a commodity and a category, and the like. For a user node, the present application only considers the relationship between a user and a commodity, and does not consider the relationship between users.

[0013] Step S12, the meta-path is a semantic relationship template in the heterogeneous graph, which defines an abstract order of node types and edge types, and the essence is a sequence formed by the alternating combination of node types and edge types. The meta-path instance is a specific implementation of the meta-path, and contains actually existing nodes and edges. The present application proposes a meta-path automatic extraction algorithm based on a combination and screening strategy. The length of the meta-path is first set to , and nodes and edges are selected. Then, an initial meta-path set is generated through a combination strategy, and a screened meta-path set is obtained through a screening strategy. The combination strategy is to first determine the order of the edges of the meta-path, and then add legal nodes at both ends of the edges, that is, the edges are matched with the connected nodes. The screening strategy is to select the meta-path with the user and commodity nodes at the head and tail, and to select the meta-path with the most node and edge types.

[0014] Step S13, according to the meta-path, the meta-path instance is extracted in the e-commerce heterogeneous network node graph. If the instance does not exist, the meta-path is removed. Otherwise, the meta-path is retained as the final extracted user meta-path, and the extracted instance is taken as the instance corresponding to the user meta-path.

[0015] Step S2, the extracted user meta-path and the corresponding instance are reversed, and the obtained new meta-path and instance are taken as the commodity meta-path and the corresponding instance. The user meta-path is divided into a user target behavior meta-path and a user auxiliary behavior meta-path according to the behavior type.

[0016] Specifically, step S2 includes the following steps. Step S21, the user meta-path is reversed, and the obtained new meta-path is taken as the commodity meta-path of the commodity. The head node of the commodity meta-path is a commodity node, and the tail node is a user node. The instance corresponding to the user meta-path is reversed, and the obtained new meta-path instance is taken as the meta-path instance of the commodity . The formula for reversing is:

[0017]

[0018] wherein, represents the user meta-path, representing the instance of the meta-path corresponding to the user , representing the reverse operation; Step S22, according to the type of the first edge in the user meta-path, the user meta-path is divided into a user target behavior meta-path and a user auxiliary behavior meta-path, wherein, the present application is used for commodity purchase recommendation, therefore, the purchase behavior meta-path is taken as the user target behavior meta-path, representing the target behavior meta-path of the user , the remaining behavior (collection, adding to the shopping cart, etc.) meta-path other than the purchase behavior is taken as the user auxiliary behavior meta-path, representing the first auxiliary behavior meta-path of the user , representing the category index of the auxiliary behavior, , , representing the number of categories of auxiliary behaviors.

[0019] Step S3, input the constructed e-commerce heterogeneous network node graph, the user target behavior meta-path and the corresponding instance thereof, the user auxiliary behavior meta-path and the corresponding instance thereof, and the commodity meta-path and the corresponding instance thereof into the commodity recommendation model based on multi-behavior meta-path semantic mapping, to obtain the user target behavior semantic embedding vector and the mapped auxiliary behavior semantic embedding vector, the initial embedding vector of the commodity and the aggregated commodity meta-path embedding vector.

[0020] Wherein, the commodity recommendation model comprises a high-quality user meta-path instance extraction module, an initial embedding extraction module, a user and commodity embedding extraction module, a user auxiliary behavior semantic mapping module and a contrast learning module; Step S3 specifically comprises: Step S31, in the high-quality user meta-path instance extraction module, high-quality user meta-path instances are extracted based on the embedding vector of the instance and the embedding vector of the meta-path; Step S32, in the initial embedding extraction module, the user auxiliary behavior initial embedding vector and the initial embedding vector of the commodity are extracted by the FunkSVD matrix decomposition algorithm; Step S33, in the user and commodity embedding extraction module, based on the high-quality user meta-path instance, the user target behavior semantic embedding vector, the user auxiliary behavior semantic embedding vector and the aggregated commodity meta-path embedding vector are obtained by aggregating the meta-path embedding vector; Step S34, in the user auxiliary behavior semantic mapping module, after splicing the user auxiliary behavior initial embedding vector and the user auxiliary behavior semantic embedding vector, mapping is performed through meta-learning to obtain the mapped auxiliary behavior semantic embedding vector; Step S35: In the contrastive learning module, the mapped auxiliary behavior semantic embedding vector is optimized by the contrastive learning loss function.

[0021] In this embodiment, step S31 specifically includes: In the high-quality user metapath instance extraction module, high-quality metapath instances are extracted based on user feature preference similarity. The embedding vector of a metapath instance can represent the semantic information of the metapath, and since metapaths are extracted through user behavior, the semantic information of the metapath can represent the user's feature preferences under that behavior. However, there may be multiple instances extracted from the metapath, and different instances may contain common nodes. Using the embedding vector obtained by aggregating all instances as the semantic information of the metapath would cause redundancy in node information extraction. Therefore, instances containing redundant information should be filtered out before using metapath instances, i.e., high-quality metapath instances should be extracted.

[0022] The high-quality user metapath instance extraction module extracts high-quality metapath instances only for instances obtained from the same user metapath. Specifically, it first extracts the embedding vector of each instance of a single user metapath using a convolutional neural network, with the expression as follows:

[0023] in, Indicates user The The first element of the path The embedding vector of each instance, It is a convolutional neural network. For the first The encoding matrix of each instance, These are the parameters of the convolutional neural network; Then, by aggregating the embedding vectors of each instance on average, we obtain the embedding vector of the metapath, expressed as:

[0024] in, For users The Embedding vector of element path, For users The The total number of instances of a single path; Next, remove the first metapath. The embedding vectors of the remaining instances are re-aggregated by average to obtain the result after removing the first instance. Metapath embedding vector after each instance and calculate and cosine similarity score The expression is:

[0025] in, The cosine similarity function; A high cosine similarity score indicates that the deleted instance has little impact on the user preferences reflected in the metapath, meaning the instance is of lower quality. Conversely, a low score indicates a higher quality instance. Therefore, instances are sorted in ascending order of cosine similarity score, and the top instances are selected. Each instance serves as a high-quality user metapath instance extracted from the data.

[0026] Step S32 specifically includes: In the initial embedding extraction module, bipartite graphs of different behaviors are extracted from the heterogeneous network node graph of e-commerce. The bipartite graphs are then transformed into interaction matrices corresponding to the behaviors, which are the target behavior interaction matrices. and auxiliary behavior interaction matrix The interaction matrix is ​​decomposed using the FunkSVD matrix factorization algorithm to obtain the initial embedding vectors of user-assisted behaviors and products.

[0027] Step S33 specifically includes: In the user and product embedding extraction module, based on high-quality user meta-path instances, the embedding vectors of similar behavior meta-path instances are aggregated through an attention mechanism, expressed as:

[0028]

[0029]

[0030] in, For the first Item path instance for user Importance score For activation function, This represents the parameter matrix of the feedforward neural network. For users The The first element of the path Embedding vectors of high-quality instances, For the first The weight of a single path instance. Indicates the first The attention mechanism of the size calculates the first The weight of a single path instance. For normalization function, It is a mathematical constant The index, is the user's behavior semantic embedding vector, which is divided into target behavior semantic embedding vector and auxiliary behavior semantic embedding vector according to the behavior type; is an average operation; There may be multiple meta-paths of the same behavior, so the embedding vectors of all meta-paths of the same behavior are aggregated again using the attention mechanism to obtain the user's behavior semantic embedding vector, expressed as:

[0031]

[0032]

[0033] wherein, is the importance score of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the i-th meta-path weight; is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, is the embedding vector of the i-th meta-path of the same behavior of the user, The embedding vector of the commodity meta-path instance is extracted using the convolutional neural network again, and then the embedding vector of the commodity meta-path instance is aggregated through the attention mechanism to obtain the embedding vector of the commodity meta-path. Finally, the embedding vector of the aggregated commodity meta-path is obtained by aggregating the embedding vectors of the commodity meta-paths through the attention mechanism.

[0034] The step S34 specifically comprises: In the user auxiliary behavior semantic mapping module, the auxiliary behavior can be indirectly derived to the target behavior. In order to measure the derivation relationship, the module maps the user's auxiliary behavior semantic embedding into the target behavior semantic embedding space through meta-learning.

[0035] In this embodiment, the user auxiliary behavior semantic embedding vector obtained in the step S33 and the user auxiliary behavior initial embedding vector obtained in the step S32 are spliced first, and then the spliced embedding vector is input into the multilayer perceptron. Then, the embedding vector output by the multilayer perceptron is sliced and reshaped to obtain a learnable parameter matrix and a bias parameter vector. Finally, the mapped auxiliary behavior semantic embedding vector is obtained through the mapping function, expressed as: ​​​

[0036]

[0037]

[0038]

[0039] in, It is the embedding vector processed by a multilayer perceptron. It is a multilayer perceptron. User The Semantic embedding vectors of auxiliary behaviors, User The The initial embedding vector for each auxiliary behavior. This indicates a splicing operation. User The The learnable mapping parameter matrix of the auxiliary behaviors. This represents a vector reshaping operation; This indicates a slicing operation, before extraction. One component; User The A learnable mapping bias parameter vector for each auxiliary behavior; This indicates a slicing operation, extracting the first slice. The final portion; The mapped user The Auxiliary behavior semantic embedding vector, Represents a non-linear activation function. This represents the normalization function.

[0040] In step S35, positive samples consist of embedding pairs consisting of the target behavior semantic embedding vector of the same user and the auxiliary behavior semantic embedding vector after each mapping, while negative samples consist of embedding pairs consisting of the target behavior semantic embedding vectors of different users.

[0041] In this embodiment, the contrastive learning loss function The expression is:

[0042]

[0043] in, For the first Auxiliary behavior loss function Indicates the number of user nodes. For users Target behavior semantic embedding vector Indicates the temperature coefficient. For users Target behavior semantic embedding vector.

[0044] Step S4: Fuse the user target behavior semantic embedding vector with the mapped auxiliary behavior semantic embedding vector to obtain the final user embedding vector. Fuse the initial product embedding vector with the aggregated product meta-path embedding vector to obtain the final product embedding vector. Based on the final user embedding vector and the final product embedding vector, obtain the product rating function. At the same time, construct the final loss function of the product recommendation model based on the product rating function. By minimizing the final loss function, the product recommendation model finally outputs the product recommendation list.

[0045] Specifically, step S4 includes: Step S41: Fuse the user target behavior semantic embedding vector with the mapped auxiliary behavior semantic embedding vector to obtain the final user embedding vector, expressed as:

[0046] in, For users The corresponding final user embedding vector; Step S42: Merge the initial embedding vector of the product with the aggregated product metapath embedding vector to obtain the final product embedding vector, expressed as:

[0047] in, For goods The corresponding final product embedding vector, For the goods Initial embedding vector, For goods The corresponding aggregated product metapath embedding vector; Step S43: Based on the final user embedding vector and the final product embedding vector, the product rating function is obtained, expressed as:

[0048] in, Indicates user Purchase goods The predicted score Indicates transpose; Step S44: Set the items that the user has purchased as positive samples and the items that have not been purchased as negative samples, and randomly sample from the positive and negative samples respectively. The sample combination is a positive and negative sample pair, and a pair-wise marginal loss function is used for target optimization, and the expression is:

[0049] wherein, is a pair-wise marginal loss function of the commodity purchase recommendation task, is a predicted score of the user purchasing the first positive sample, is a predicted score of the user purchasing the first negative sample, is a predicted score of the user purchasing the first negative sample, max represents taking the maximum value; The contrast learning loss function and the pair-wise marginal loss function are weighted and fused to obtain a final loss function of the commodity recommendation model

[0050] wherein, is a hyperparameter for controlling the contrast learning intensity, is a hyperparameter for controlling the weight decay intensity, is a parameter of the commodity recommendation model, represents the square of the Frobenius norm of the parameter matrix, and is used to realize regularization to prevent model overfitting; By minimizing the final loss function, the commodity recommendation model finally outputs a commodity recommendation list, and the commodity recommendation list shows the commodities that the user is most likely to purchase.

[0051] In summary, according to the commodity recommendation method based on multi-behavior meta-path semantic mapping according to the above embodiment, the following beneficial effects are obtained: (1) The present application does not need to manually define the meta-path by the field experts. The existing commodity recommendation method based on meta-path mostly predefines the meta-path manually, which not only needs the prior knowledge of the field experts, but also may cause the loss of the meta-path. The present application proposes an automatic extraction algorithm of the meta-path based on the combination and screening strategy, sets the length of the meta-path first, then selects the nodes and edges based on the set length of the meta-path, generates an initial meta-path set through the combination strategy, and obtains the screened meta-path set through the screening strategy, extracts the meta-path instance in the heterogeneous network according to the meta-path, and if the instance exists, the meta-path is retained, otherwise the meta-path is removed, thereby realizing the effect of automatically extracting the meta-path.

[0052] (2) The present application reduces the redundancy of node information extraction by extracting high-quality meta-path instances. Existing meta-path-based commodity recommendation methods mostly extract information by aggregating meta-path instances, but there are often multiple instances of the same meta-path, and aggregating all instances will cause redundancy of node information. The method of filtering instances based on the similarity between nodes has poor interpretability. The present application extracts high-quality meta-path instances based on cosine similarity scores from the semantic information represented by the meta-path, which not only reduces the redundancy of node information, but also has strong interpretability.

[0053] (3) The present application can improve the accuracy of commodity recommendation by mining the potential relationship between different behavior meta-paths. Existing meta-path-based commodity recommendation methods mostly cannot mine the potential relationship between different behavior meta-paths, ignoring the inference relationship between different behaviors. The present application can map the semantic embedding vector of the auxiliary behavior to the semantic space of the target behavior meta-path by concatenating the initial embedding vector of the user auxiliary behavior with the semantic embedding vector of the user auxiliary behavior and then mapping it through meta-learning, which can realize the explicit modeling of the potential relationship between different behavior meta-paths, and the mapped auxiliary behavior semantic embedding vector can be optimized through the contrastive learning loss function, which can improve the accuracy of recommendation.

[0054] The above-described embodiments only express several embodiments of the present application, and the description is relatively specific and detailed, but it should not be understood as limiting the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of protection of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A product recommendation method based on multi-behavior meta-path semantic mapping, characterized in that, include: Step S1: Read the user and product interaction database in the e-commerce platform, construct the e-commerce heterogeneous network node graph, extract the meta-paths using the meta-path automatic extraction algorithm based on combination and filtering strategies, use the extracted meta-paths as user meta-paths, and extract the corresponding instances in the e-commerce heterogeneous network node graph based on the user meta-paths. Step S2: Invert the extracted user meta paths and their corresponding instances, use the new meta paths and instances as product meta paths and their corresponding instances, and divide the user meta paths into user target behavior meta paths and user auxiliary behavior meta paths according to the behavior type. Step S3: Input the constructed e-commerce heterogeneous network node graph, user target behavior meta-path and its corresponding instance, user auxiliary behavior meta-path and its corresponding instance, product meta-path and its corresponding instance into the product recommendation model based on multi-behavior meta-path semantic mapping to obtain user target behavior semantic embedding vector, mapped auxiliary behavior semantic embedding vector, initial embedding vector of product and aggregated product meta-path embedding vector. Step S4: Fuse the user target behavior semantic embedding vector with the mapped auxiliary behavior semantic embedding vector to obtain the final user embedding vector. Fuse the initial product embedding vector with the aggregated product meta-path embedding vector to obtain the final product embedding vector. Based on the final user embedding vector and the final product embedding vector, obtain the product rating function. At the same time, construct the final loss function of the product recommendation model based on the product rating function. By minimizing the final loss function, the product recommendation model finally outputs the product recommendation list.

2. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 1, characterized in that, Step S1 specifically includes: Step S11: Construct the node graph of the e-commerce heterogeneous network by reading the user and product interaction dataset from the user and product interaction database of the e-commerce platform. , ,in, Represents a set of nodes. Describe the set of edges. A collection representing node types. A set representing the edge type; Step S12 proposes an automatic meta-path extraction algorithm based on a combination and filtering strategy. First, the meta-path length is set to... Select Each node and The edges are then combined to generate an initial meta-path set, and then a filtered meta-path set is obtained through a filtering strategy. The combination strategy is to first determine the order of the meta-path edges, and then add valid nodes to both ends of the edges, that is, the edges must match the connected nodes. The filtering strategy is to select meta-paths with user and product nodes as the head and tail nodes respectively, and select meta-paths with the most node and edge types. Step S13: Extract meta-path instances from the heterogeneous network node graph of e-commerce based on the meta-path. If the instance does not exist, remove the meta-path; otherwise, retain the meta-path and use it as the final extracted user meta-path. Use the extracted instance as the instance corresponding to the user meta-path.

3. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 2, characterized in that, Step S2 specifically includes: Step S21, will the user The original metapath is reversed, and the resulting new metapath is used as the product. Product metapath The head node of the product metapath is the product node, and the tail node is the user node; for users... The instance corresponding to the meta path is reversed, and the resulting new meta path instance is used as the product. Metapath instance The inverse formula is: in, Indicates user metapath, Indicates user The instance corresponding to the metapath, Indicates the reverse operation; Step S22: Based on the type of the first edge in the user meta-path, the user meta-path is divided into user target behavior meta-path and user auxiliary behavior meta-path, wherein the purchase behavior meta-path is regarded as the user target behavior meta-path. Indicates user The target behavior meta-path is defined, and other behavior meta-paths besides the purchase behavior are treated as user auxiliary behavior meta-paths. Indicates user The A type of auxiliary behavior metapath, in which, An index representing the types of auxiliary behaviors. , This indicates the number of types of auxiliary behaviors.

4. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 3, characterized in that, The product recommendation model includes a high-quality user metapath instance extraction module, an initial embedding extraction module, a user and product embedding extraction module, a user auxiliary behavior semantic mapping module, and a contrastive learning module. Step S3 specifically includes: Step S31: In the high-quality user metapath instance extraction module, high-quality user metapath instances are extracted based on the instance embedding vector and the metapath embedding vector. Step S32: In the initial embedding extraction module, the initial embedding vectors of user auxiliary behavior and products are extracted using the FunkSVD matrix factorization algorithm. Step S33: In the user and product embedding extraction module, based on high-quality user meta-path instances, the user target behavior semantic embedding vector, user auxiliary behavior semantic embedding vector, and aggregated product meta-path embedding vector are obtained by aggregating meta-path embedding vectors. Step S34: In the user-assisted behavior semantic mapping module, the initial embedding vector of user-assisted behavior and the semantic embedding vector of user-assisted behavior are concatenated and then mapped through meta-learning to obtain the mapped semantic embedding vector of assisted behavior. Step S35: In the contrastive learning module, the mapped auxiliary behavior semantic embedding vector is optimized by the contrastive learning loss function.

5. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 4, characterized in that, Step S31 specifically includes: In the high-quality user metapath instance extraction module, the embedding vector of each instance of a single user metapath is first extracted using a convolutional neural network, expressed as: in, Indicates user The The first element of the path The embedding vector of each instance, It is a convolutional neural network. For the first The encoding matrix of the instances, These are the parameters of the convolutional neural network; Then, by aggregating the embedding vectors of each instance on average, we obtain the embedding vector of the metapath, expressed as: in, For users The Embedding vector of element path, For users The The total number of instances of a single path; Next, remove the first metapath. The embedding vectors of the remaining instances are re-aggregated by average to obtain the result after removing the first instance. Metapath embedding vector after each instance and calculate and cosine similarity score The expression is: in, The cosine similarity function; Instances are sorted in ascending order of cosine similarity score, and the top instances are selected. Each instance serves as a high-quality user metapath instance extracted from the data.

6. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 5, characterized in that, Step S32 specifically includes: In the initial embedding extraction module, bipartite graphs of different behaviors are extracted from the heterogeneous network node graph of e-commerce. The bipartite graphs are then transformed into interaction matrices corresponding to the behaviors, which are the target behavior interaction matrices. and auxiliary behavior interaction matrix The interaction matrix is ​​decomposed using the FunkSVD matrix factorization algorithm to obtain the initial embedding vectors of user-assisted behaviors and products.

7. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 6, characterized in that, Step S33 specifically includes: In the user and product embedding extraction module, based on high-quality user meta-path instances, the embedding vectors of similar behavior meta-path instances are aggregated through an attention mechanism, expressed as: in, For the first Item path instance for user Importance score For activation function, This represents the parameter matrix of the feedforward neural network. For users The The first element of the path Embedding vectors of high-quality instances, For the first The weight of a single path instance. Indicates the first The attention mechanism of the size calculates the first The weight of a single path instance. For normalization function, It is a mathematical constant The index, It is a user that aggregates high-quality instance embedding vectors using an attention mechanism. The Embedding vector of element path, For averaging operations; Then, the attention mechanism is used to aggregate all meta-path embedding vectors of the same type of behavior, thereby obtaining the user's behavior semantic embedding vector, expressed as: in, For users The first of the same behaviors Path pairs Importance score It is the feature vector of the attention parameters. User The first of the same behaviors Embedding vector of element path, It is the first The weight of each element path; User The behavioral semantic embedding vectors are divided into target behavioral semantic embedding vectors and auxiliary behavioral semantic embedding vectors according to the behavioral type. Then, a convolutional neural network is used to extract the embedding vectors of the product metapath instances. Then, the embedding vectors of the product metapath instances are aggregated through an attention mechanism to obtain the embedding vector of the product metapath. Finally, the embedding vectors of the product metapath are aggregated through an attention mechanism to obtain the aggregated product metapath embedding vector.

8. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 7, characterized in that, Step S34 specifically includes: In the user-assisted behavior semantic mapping module, the user-assisted behavior semantic embedding vector obtained in step S33 and the initial user-assisted behavior embedding vector obtained in step S32 are first concatenated. Then, the concatenated embedding vector is input into the multilayer perceptron. Next, the embedding vector output by the multilayer perceptron is sliced ​​and reshaped to obtain the learnable parameter matrix and bias parameter vector. Finally, the mapped auxiliary behavior semantic embedding vector is obtained through the mapping function, the expression of which is: in, It is the embedding vector processed by a multilayer perceptron. It is a multilayer perceptron. User The Semantic embedding vectors of auxiliary behaviors, User The The initial embedding vector for each auxiliary behavior. This indicates a splicing operation. User The The learnable mapping parameter matrix of the auxiliary behaviors. This represents a vector reshaping operation; This indicates a slicing operation, before extraction. One component; User The A learnable mapping bias parameter vector for each auxiliary behavior; This indicates a slicing operation, extracting the first slice. The final portion; The mapped user The Auxiliary behavior semantic embedding vector, Represents a non-linear activation function. This represents the normalization function.

9. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 8, characterized in that, In step S35, the contrastive learning loss function is used. The expression is: in, For the first Auxiliary behavior loss function Indicates the number of user nodes. For users Target behavior semantic embedding vector Indicates the temperature coefficient. For users Target behavior semantic embedding vector.

10. The product recommendation method based on multi-behavior meta-path semantic mapping according to claim 9, characterized in that, Step S4 specifically includes: Step S41: Fuse the user target behavior semantic embedding vector with the mapped auxiliary behavior semantic embedding vector to obtain the final user embedding vector, expressed as: in, For users The corresponding final user embedding vector; Step S42: Merge the initial embedding vector of the product with the aggregated product metapath embedding vector to obtain the final product embedding vector, expressed as: in, For goods The corresponding final product embedding vector, For the goods Initial embedding vector, For goods The corresponding aggregated product metapath embedding vector; Step S43: Based on the final user embedding vector and the final product embedding vector, the product rating function is obtained, expressed as: in, Indicates user Purchase goods The predicted score Indicates transpose; Step S44: Set the items that the user has purchased as positive samples and the items that have not been purchased as negative samples, and randomly sample from the positive and negative samples respectively. Each sample combination consists of positive and negative sample pairs. The objective is optimized using a pairwise marginal loss function, expressed as: in, The pairwise marginal loss function for the product purchase recommendation task. For users Purchase the Positive samples The predicted score For users Purchase the negative samples The predicted score This indicates taking the maximum value; The final loss function of the product recommendation model is obtained by weighted fusion of the contrastive learning loss function and the pairwise marginal loss function. The expression is: in, It is a hyperparameter that controls the intensity of contrastive learning. It is a hyperparameter that controls the intensity of weight decay. These are the parameters of the product recommendation model. This represents the square of the Frobenius norm of the parameter matrix; By minimizing the final loss function, the product recommendation model ultimately outputs a list of recommended products.

Images