Multi-behavior generality fusion recommendation method and system

Abstract

The invention relates to a multi-behavior generality fusion recommendation method, which comprises the following steps of: A, acquiring behavior data of a user, including a user ID (Identity), a target article ID, an interaction type and interaction time, preprocessing the acquired data, and then constructing a training set; b, a deep learning network model based on multi-behavior generality fusion is constructed, the deep learning network model obtains generality information of different types of behaviors and fuses the generality information into final preference representation, and meanwhile learning of multiple representations is assisted through multi-angle comparative learning; training a deep learning network model by using the training set, and obtaining a required deep learning network model on the training set through minimum loss function training; and step C, inputting the preprocessed to-be-recommended user and candidate item data into the trained deep learning network model, and outputting the probability that the current user generates a target behavior on the target item. The method and the system are beneficial to improving the accuracy of recommending interested articles to the user.

Description

Technical Field

[0001] The present invention belongs to the field of recommendation systems, and in particular relates to a recommendation method and system for integrating commonalities of multiple behaviors. Background Art

[0002] In the era of information overload, how to extract information that users are interested in from massive data is a core essence. As an application practice of data mining, the purpose of the recommendation system is to achieve this essence by accurately mining the interest preferences of each user and providing each user with information that meets their own interests. Personalized recommendation systems have been widely adopted in video platforms, e-commerce, social platforms and other fields to alleviate the chain problems caused by information overload. Most existing recommendation models usually only use the behavioral data corresponding to their target tasks. In fact, other types of behaviors often contain potential user interest representations. For example, on an online retail platform, there may be multiple types of behaviors between users and items, such as browsing, adding to shopping carts, and purchasing. Different types of behaviors correspond to different dimensions of user preferences. There is a suboptimal problem in only using target behaviors to represent the overall preferences of users. How to effectively capture and integrate user preferences from different angles corresponding to multiple types of behaviors is challenging. Therefore, recent models have begun to introduce multi-type behavioral data. By using multi-type behavioral data to model the correlation within different types of data and the dependency between different types of data, multi-behavior recommendation models are constructed in the form of two paradigms. In addition, recent related work has begun to introduce contrastive learning to assist model learning.

[0003] However, although the existing models have achieved performance improvements to a certain extent, they still face some major weaknesses. First, the model still suffers from data sparsity and semantic authenticity. Although a variety of auxiliary behavior data are used to assist in modeling user preferences, their impact on the target task is smaller than that of the target behavior, and the sparsity of the target data still exists. In addition, the fusion of multiple types of behavior semantics may lead to the destruction of the user's true semantics, so it is crucial to correctly guide the fusion of behavior semantics and solve sparsity. Secondly, existing multi-behavior recommendation models usually model the contextual relationships within each behavior and the dependencies between behaviors, but lack the extraction of the commonality of users' overall behaviors. Behavioral commonality can reflect the user's attention to each item and can indirectly reflect whether the user performs the target behavior on the item. Therefore, how to effectively capture the potential commonality of behavior representation and integrate it into the user's overall preference is an important issue. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for recommending items of interest by integrating commonalities of multiple behaviors, which method and system are conducive to improving the accuracy of recommending items of interest to users.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a multi-behavior commonality fusion recommendation method, comprising the following steps:

[0006] Step A: Obtain user behavior data, including user ID, target item ID, interaction type, and interaction time, pre-process the acquired data, and then construct a training set;

[0007] Step B: construct a deep learning network model G based on the fusion of commonalities of multiple behaviors, wherein the deep learning network model G obtains common information of different types of behaviors and fuses it into the final preference representation, and assists the learning of multiple representations through multi-angle comparative learning; train the deep learning network model G using the training set, and obtain the required deep learning network model G by minimizing the loss function on the training set;

[0008] Step C: Input the preprocessed recommended user and candidate item data into the trained deep learning network model G, and output the probability that the current user will generate the target behavior for the target item.

[0009] Furthermore, the step A specifically includes the following steps:

[0010] Step A1: Obtain an existing general multi-behavior recommendation dataset; each sample in the multi-behavior recommendation dataset contains a user ID, a target item ID, and the type of interaction, whether an interaction occurs, and the interaction time;

[0011] Step A2: Preprocess each multi-behavior recommendation dataset, including: deleting outlier data; removing users with less than 5 interactions and items with less than 10 interactions to reduce the number of users and items with fewer interactions; and forming a training set with the preprocessed multi-behavior recommendation dataset.

[0012] Furthermore, the step B specifically includes the following steps:

[0013] Step B1: Divide the training set according to different behavior types and construct different types of user-item interaction graphs G k ;

[0014] Step B2: Different types of user-item interaction graphs G k Input into the corresponding initialization embedding layer to obtain the user initialization embedding corresponding to the kth behavior type Initialize embed with items Based on multiple types of initialization embeddings, we get the overall initialization embedding e corresponding to the user. u The overall initialization embedding e corresponding to the item i ;

[0015] Step B3: Build a behavior-aware relationship network and embed the overall initialization of the user obtained in step B2 into e u Initialize the embedding with the whole item i As the input of the first layer of the behavior-aware relation network, they are respectively input into the intra-behavior relation encoder to capture the specific contextual information within the same type, and obtain the intra-behavior relation representation of users and items. and Then all types of intra-behavior relationships of users are represented as Relationship between the behavior and the item They are input into the behavior relationship encoder to capture different types of relationship dependencies and obtain the overall behavior relationship representation corresponding to users and items. and The output of each layer is used as the input of the next layer for iterative aggregation, and the multi-layer behavior relationship representation of users and items is fused using the fully connected layer to obtain and

[0016] Step B4: Based on each type of user-item interaction graph, construct the overall user-item interaction graph G of heterogeneous relationships all ; Build a multi-view meta-network based on the overall user-item interaction graph G all The relationship between behaviors obtained from each layer of the network in step B3 is used to construct a meta-feature view of each type of user at each layer. and meta-feature views of each type of item In this way, the common information of different types is captured; the different types of meta-feature views in each layer are input into the personalized meta-networks F1 and F2 respectively, so as to capture the adaptive transformation matrix; and the different types of common information in each layer are aggregated using the adaptive transformation matrix to obtain the personalized common representation of users and items at each layer. and Construct a feature debiasing network to enhance the authenticity of the extracted common representation and improve the robustness of the common representation; for the common representation of each layer obtained, fuse it based on proportional fusion to obtain a multi-layer fused common representation and The relationship between the user's behaviors obtained in step B3 is expressed as Relationship between the behavior of the object The commonality representation of the user obtained in step B4 Commonality with items Aggregate in a proportional fusion manner to obtain the final user preference representation and item preference representation

[0017] Step B5: Construct a multi-angle contrastive learning paradigm, use uniform noise enhancement to generate corresponding enhanced views for the behavioral relationship representation between users and items at each layer in step B3; and perform contrastive learning with the original view to obtain the contrast loss of users and items based on uniform noise enhancement. and Utilize the user's final preference representation that integrates common information Relationship between user behavior and the final preference representation of the item Relationship between behaviors and items Perform contrastive learning based on feature fusion to obtain the contrast loss of users and items. and

[0018] Step B6: Perform random negative sampling on users in a batch to generate corresponding positive sample interaction sets and negative sample interaction sets; use the pairwise loss function to calculate the prediction loss; then fuse it with multiple contrast losses to obtain the final optimization target loss, and then calculate the gradient of each parameter in the deep learning network model G through the back propagation method, and use the Adam optimization algorithm to update each parameter to constrain the loss size;

[0019] Step B7: Use the final preference representation of users and items to train the model. When the loss value generated by the deep learning network model is less than the set threshold or reaches the maximum number of iterations, terminate the training of the deep learning network model G.

[0020] Furthermore, the step B2 specifically includes the following steps:

[0021] Step B21: Use the user-item interaction graph to input into the corresponding initialization embedding layer to obtain the user's specific type of initialization embedding and item-specific initialization embedding

[0022] Step B22: Aggregate the initialization embeddings of each type obtained in step B21 in the form of a list to obtain the user's overall initialization embedding and the overall initialization embedding of the item

[0023] Furthermore, the step B3 specifically includes the following steps:

[0024] Step B31: Construct a behavior-aware relationship network, which includes an intra-behavior relationship encoder and an inter-behavior relationship encoder, which are constructed by a memory weighted network and a multi-head self-attention network with residual connections, respectively, and learn the multi-hop information of users and items by iterating the number of network layers, and embed the overall initialization of the user obtained in step B22 into e u and the overall initialization embedding of the item i As the first layer input of the network, and the output of each layer of the network as the input of the next layer, for the first layer, the intra-behavior relation encoder is first used to capture each type of context information to obtain the intra-behavior relation representation corresponding to the user and the item and

[0025]

[0026]

[0027] Among them, e u With e i They are the overall initialization embeddings for users and items respectively. IntraEnc() consists of a memory weighted network, which mainly performs conversion learning between different dimensions.

[0028] Step B32: Represent the behavior relationship between the user and the item obtained in step B31 and They are input into the behavior relationship encoder to capture different types of relationship dependencies and obtain different types of behavior relationship representations corresponding to users and items. and And sum to get the overall behavior relationship between users and items and

[0029]

[0030]

[0031] Among them, IntraEnc() is composed of a multi-head self-attention mechanism, which retains the original intra-behavior relationship representation information;

[0032] Step B33: Repeat steps B31 and B32 for the number of layers in the network; thus obtaining a multi-layer overall behavior relationship representation list corresponding to users and items. and

[0033] Step B34: Based on the multi-layer behavior relationship representation list of users and items obtained in step B33, a fully connected layer is used to aggregate the multi-layer representation:

[0034]

[0035]

[0036] Among them, concat() represents the concatenation function, which concatenates the representations of multiple different layers, and Mlp1() and Mlp2() represent fully connected layers with activation functions.

[0037] Furthermore, the step B4 specifically includes the following steps:

[0038] Step B41: Based on each type of user-item interaction graph [G 1 ,G 2 ,…,G K ], constructing the overall user-item interaction graph G with heterogeneous relationships all , which contains all types of interaction edges between users and items;

[0039] Step B42: Based on the overall user-item interaction graph G obtained in step B41 all The list of relationships between the behaviors of users and items obtained by the l-th layer network in step B32 and Construct user- and item-specific meta-feature views respectively, and use them to capture the common behavioral information in different types:

[0040]

[0041]

[0042]

[0043]

[0044] in, The target representation that represents the commonality of behaviors aggregates multiple different types of relationship representations between behaviors in the same layer through residual connections. Mlp3() represents a fully connected layer with an activation function, and concat() represents a concatenation function that concatenates multiple representations to form a specific type of meta-information view, which can be used as input for the subsequent network that captures the commonality of behaviors.

[0045] Step B43: Using two adaptive meta-transformation matrix learners, the meta-feature view lists of different types of users in each layer obtained in step B42 are respectively List of meta-traits view for different types of items Traverse and capture the adaptive transformation matrix of users and items under different types:

[0046]

[0047]

[0048] Among them, F1 and F2 represent the meta-learners for learning the adaptive transformation matrix, which are composed of two fully connected layers with activation functions;

[0049] Step B44: Use the adaptive transformation matrix obtained in step B34 to aggregate different types of common information in each layer to obtain personalized common representations of users and items in each layer:

[0050]

[0051]

[0052] Among them, δ() represents the activation function, which learns the potential common transformation representation of behavior under a specific type of representation through two adaptive transformation matrices;

[0053] Step B45: Use the constructed feature debiasing network to enhance the authenticity of the extracted common representation and improve the robustness of the common representation to capture the enhanced common representation of users and items:

[0054]

[0055]

[0056]

[0057]

[0058] Among them, Feature_Decorr() represents the feature debiasing network, which learns the unique dimension debiasing weight by removing the noise between dimensions, and enhances the original commonality representation in the form of residual connection;

[0059] Step B46: For the enhanced commonality representation of users and items in each layer obtained in step B45, they are fused by proportional fusion to obtain a multi-layer fused enhanced commonality representation, and then passed through a fully connected layer to obtain an aggregated behavior commonality representation:

[0060]

[0061]

[0062]

[0063]

[0064] Among them, Mlp4 represents the fully connected layer with activation function, α m Represents the hyperparameter that controls the ratio of fusion in each layer;

[0065] Step B47: Represent the relationship between the user's behaviors obtained in step B3 Relationship between the behavior of the object The commonality representation of the user obtained in step B4 Commonality with items Aggregate in a proportional fusion manner to obtain the final user preference representation and item preference representation

[0066]

[0067]

[0068] Among them, α f represents a hyperparameter that controls the ratio of commonality representation fusion.

[0069] Furthermore, the step B5 specifically includes the following steps:

[0070] Step B51: Generate a corresponding enhanced view for the relationship representation between the user and the item behavior at each layer in step B3 using a uniform noise enhancement method and

[0071]

[0072]

[0073]

[0074]

[0075] in, and They are obtained by averaging L enhanced representations based on controllable uniform noise enhancement to avoid the randomness of a single enhanced representation; and They represent the feature representations initialized uniformly from 0 to 1 to achieve uniform enhancement, α eps ∈R represents a hyperparameter that controls the size of the enhanced representation;

[0076] Step B52: Perform a comparative learning operation using the original image and the enhanced image. The comparative learning loss paradigm based on uniform noise enhancement at the user end and the item end is:

[0077]

[0078]

[0079] in, A similarity function is used, which uses the inner product as the similarity function; τ represents the temperature coefficient, and u' and i' represent negative samples with different indexes respectively; based on the contrast loss defined above, the consistency between the original view and the enhanced view in each layer can be maximized, and the difference between different users and objects can be maximized, thereby enhancing the robustness of the view after the fusion of multiple behavioral relationships;

[0080] Step B53: Representing the final user preference using fused common information Relationship between user behavior and the final preference representation of the item Relationship between behaviors and items Perform contrastive learning based on feature fusion to obtain the contrast loss of users and items. and

[0081]

[0082]

[0083] in, A similarity function is adopted, which uses the inner product as the similarity function; τ represents the temperature coefficient, u' and i' represent negative samples with different indexes respectively. Based on the contrast loss defined above, the similarity between the multi-relation fusion representation view and the behavior relationship representation view can be maximized, and the difference between different users and items can be maximized, thereby enhancing the robustness of the multi-relation fusion representation view.

[0084] Furthermore, the step B6 specifically includes the following steps:

[0085] Step B61: Sample the corresponding negative samples for a user in a small batch to form a corresponding positive sample interaction set Interacting with negative samples

[0086] Step B62: Use pairwise loss function Calculate the prediction loss, given u i Under the positive sample set and negative sample set, the prediction loss is defined as:

[0087]

[0088] in, Represents user u i The sth interactive item in the corresponding positive sample item sequence, and Represents user u i The sth uninteracted item in the corresponding negative sample item sequence, where θ represents the learnable parameters in the network, and the parameter λ1 and L2 regularization are used to alleviate the overfitting problem;

[0089] Step B63: Update the learning rate through the gradient optimization algorithm Adam, and use back propagation to iteratively update the model parameters to minimize the loss function to train the model G; merge the contrast loss obtained in steps B52 and B53 with the prediction loss obtained in step B61 to define the final recommendation loss:

[0090]

[0091] Among them, λ2 and λ3 represent the hyperparameters that control the intensity of the contrastive learning task.

[0092] The present invention also provides a multi-behavior commonality fusion recommendation system, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above-mentioned method steps can be implemented.

[0093] Compared with the prior art, the present invention has the following beneficial effects: the present invention provides a multi-behavior commonality fusion recommendation method and system. Different from the previous methods, the present invention fully considers the behavioral commonality information existing in the multi-behavior data, explicitly captures its characteristics and integrates them into the final preference representation; and constructs a multi-angle comparative learning paradigm to promote the effective learning of representation from two angles. The experimental results show that the performance of the present invention is better than the existing commonly used models, and can effectively improve the accuracy of recommending items of interest to users. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 is a flow chart of a method implementation of an embodiment of the present invention;

[0095] Figure 2 It is an architecture diagram of the deep learning network model G based on the fusion of multi-behavior commonalities in an embodiment of the present invention. DETAILED DESCRIPTION

[0096] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0097] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0098] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0099] like Figure 1 As shown, this embodiment provides a multi-behavior commonality fusion recommendation method, including the following steps:

[0100] Step A: Obtain user behavior data, including user ID, target item ID, interaction type, and interaction time, pre-process the acquired data, and then construct a training set;

[0101] Step B: construct a deep learning network model G based on the fusion of commonalities of multiple behaviors, wherein the deep learning network model G obtains common information of different types of behaviors and fuses it into the final preference representation, and assists the learning of multiple representations through multi-angle comparative learning; train the deep learning network model G using the training set, and obtain the required deep learning network model G by minimizing the loss function on the training set;

[0102] Step C: Input the preprocessed recommended user and candidate item data into the trained deep learning network model G, and output the probability that the current user will generate the target behavior for the target item.

[0103] In this embodiment, the step A specifically includes the following steps:

[0104] Step A1: Obtain an existing general multi-behavior recommendation dataset; each sample in the multi-behavior recommendation dataset contains a user ID, a target item ID, and the type of interaction, whether an interaction occurs, and the interaction time;

[0105] Step A2: Preprocess each multi-behavior recommendation dataset, including: deleting outlier data; removing users with less than 5 interactions and items with less than 10 interactions to reduce the number of users and items with fewer interactions; and forming a training set with the preprocessed multi-behavior recommendation dataset.

[0106] Figure 2 : is an architecture diagram of a deep learning network model G based on multi-behavior commonality fusion in this embodiment. In this embodiment, the specific implementation steps of step B are as follows.

[0107] Step B1: Divide the training set according to different behavior types and construct different types of user-item interaction graphs G k .

[0108] Step B2: Different types of user-item interaction graphs G k Input into the corresponding initialization embedding layer to obtain the user initialization embedding corresponding to the kth behavior type Initialize embed with items Based on multiple types of initialization embeddings, we get the overall initialization embedding e corresponding to the user. u The overall initialization embedding e corresponding to the item i .

[0109] The step B2 specifically comprises the following steps:

[0110] Step B21: Use the user-item interaction graph to input into the corresponding initialization embedding layer to obtain the user's specific type of initialization embedding and item-specific initialization embedding

[0111] Step B22: Aggregate the initialization embeddings of each type obtained in step B21 in the form of a list to obtain the user's overall initialization embedding and the overall initialization embedding of the item

[0112] Step B3: Build a behavior-aware relationship network and embed the overall initialization of the user obtained in step B2 into e u Initialize the embedding with the whole item i As the input of the first layer of the behavior-aware relation network, they are respectively input into the intra-behavior relation encoder to capture the specific contextual information within the same type, and obtain the intra-behavior relation representation of users and items. and Then all types of intra-behavior relationships of users are represented as Relationship between the behavior and the item They are input into the behavior relationship encoder to capture different types of relationship dependencies and obtain the overall behavior relationship representation corresponding to users and items. and The output of each layer is used as the input of the next layer for iterative aggregation. A simple fully connected layer is used to fuse the multiple layers of the behavior relationship representations of users and items to obtain and

[0113] The step B3 specifically comprises the following steps:

[0114] Step B31: Construct a behavior-aware relationship network, which includes an intra-behavior relationship encoder and an inter-behavior relationship encoder, which are constructed by a memory weighted network and a multi-head self-attention network with residual connections, respectively, and learn the multi-hop information of users and items by iterating the number of network layers, and embed the overall initialization of the user obtained in step B22 into e u and the overall initialization embedding of the item i As the first layer input of the network, and the output of each layer of the network as the input of the next layer, for the first layer, the intra-behavior relation encoder is first used to capture each type of context information to obtain the intra-behavior relation representation corresponding to the user and the item and

[0115]

[0116]

[0117] Among them, e u With e i They are the overall initialization embeddings for users and items respectively. InterEnc() consists of a memory weighted network, which mainly performs conversion learning between different dimensions.

[0118] Step B32: Represent the behavior relationship between the user and the item obtained in step B31 and They are input into the behavior relationship encoder to capture different types of relationship dependencies and obtain different types of behavior relationship representations corresponding to users and items. and And sum to get the overall behavior relationship between users and items and

[0119]

[0120]

[0121] Among them, InterEnc() is composed of a multi-head self-attention mechanism, which retains the original intra-behavior relationship representation information.

[0122] Step B33: Repeat steps B31 and B32 for the number of layers in the network; thus obtaining a multi-layer overall behavior relationship representation list corresponding to users and items. and

[0123] Step B34: Based on the multi-layer behavior relationship representation list of users and items obtained in step B33, a fully connected layer is used to aggregate the multi-layer representation:

[0124]

[0125]

[0126] Among them, concat() represents the concatenation function, which concatenates the representations of multiple different layers, and Mlp1() and Mlp2() represent fully connected layers with activation functions.

[0127] Step B4: Based on each type of user-item interaction graph, construct the overall user-item interaction graph G of heterogeneous relationships all ; Build a multi-view meta-network based on the overall user-item interaction graph G all The relationship between behaviors obtained from each layer of the network in step B3 is used to construct a meta-feature view of each type of user at each layer. and meta-feature views of each type of item In this way, the common information of different types is captured; the different types of meta-feature views in each layer are input into the personalized meta-networks F1 and F2 respectively, so as to capture the adaptive transformation matrix; and the different types of common information in each layer are aggregated using the adaptive transformation matrix to obtain the personalized common representation of users and items at each layer. and Construct a feature debiasing network to enhance the authenticity of the extracted common representation and improve the robustness of the common representation; for the common representation of each layer obtained, fuse it based on proportional fusion to obtain a multi-layer fused common representation and The relationship between the user's behaviors obtained in step B3 is expressed as Relationship between the behavior of the object The commonality representation of the user obtained in step B4 Commonality with items Aggregate in a proportional fusion manner to obtain the final user preference representation and item preference representation

[0128] The step B4 specifically comprises the following steps:

[0129] Step B41: Based on each type of user-item interaction graph [G 1 ,G 2 ,…,G K ], constructing the overall user-item interaction graph G with heterogeneous relationships all , which contains all types of interaction edges between users and items.

[0130] Step B42: Based on the overall user-item interaction graph G obtained in step B41all The list of relationships between the behaviors of users and items obtained by the l-th layer network in step B32 and Construct user- and item-specific meta-feature views respectively, and use them to capture the common behavioral information in different types:

[0131]

[0132]

[0133]

[0134]

[0135] in, The target representation that represents the commonality of behaviors aggregates multiple different types of relationship representations between behaviors in the same layer through residual connections. Mlp3() represents a fully connected layer with an activation function, and concat() represents a concatenation function, which concatenates multiple representations to form a specific type of meta-information view to provide input for the subsequent network that captures the commonality of behaviors.

[0136] Step B43: Using two adaptive meta-transformation matrix learners, the meta-feature view lists of different types of users in each layer obtained in step B42 are respectively List of meta-traits view for different types of items Traverse and capture the adaptive transformation matrix of users and items under different types:

[0137]

[0138]

[0139] Among them, F1 and F2 represent meta-learners for learning adaptive transformation matrices, which are composed of two fully connected layers with activation functions.

[0140] Step B44: Use the adaptive transformation matrix obtained in step B34 to aggregate different types of common information in each layer to obtain personalized common representations of users and items in each layer:

[0141]

[0142]

[0143] Among them, δ() represents the activation function, and through two adaptive transformation matrices, the potential common behavior transformation representation under a specific type of representation is learned.

[0144] Step B45: Use the constructed feature debiasing network to enhance the authenticity of the extracted common representation and improve the robustness of the common representation to capture the enhanced common representation of users and items:

[0145]

[0146]

[0147]

[0148]

[0149] Among them, Feature_Decorr() represents the feature debiasing network, which learns the unique dimension debiasing weights by removing the noise between dimensions, and enhances the original common representation in the form of residual connection.

[0150] Step B46: For the enhanced commonality representation of users and items in each layer obtained in step B45, they are fused by proportional fusion to obtain a multi-layer fused enhanced commonality representation, and then passed through a fully connected layer to obtain an aggregated behavior commonality representation:

[0151]

[0152]

[0153]

[0154]

[0155] Among them, Mlp4 represents the fully connected layer with activation function, α m represents the hyperparameter controlling the ratio of each layer fusion. Step B47: The relationship between the user behaviors obtained in step B3 is represented as Relationship between the behavior of the object The commonality representation of the user obtained in step B4 Commonality with items Aggregate in a proportional fusion manner to obtain the final user preference representation and item preference representation

[0156]

[0157]

[0158] Among them, α f represents a hyperparameter that controls the ratio of commonality representation fusion.

[0159] Step B5: Construct a multi-angle contrastive learning paradigm, use uniform noise enhancement to generate corresponding enhanced views for the behavioral relationship representation between users and items at each layer in step B3; and perform contrastive learning with the original view to obtain the contrast loss of users and items based on uniform noise enhancement. and Utilize the user's final preference representation that integrates common information Relationship between user behavior and the final preference representation of the item Relationship between behaviors and items Perform contrastive learning based on feature fusion to obtain the contrast loss of users and items. and

[0160] The step B5 specifically comprises the following steps:

[0161] Step B51: Generate a corresponding enhanced view for the relationship representation between the user and the item behavior at each layer in step B3 using a uniform noise enhancement method and

[0162]

[0163]

[0164]

[0165]

[0166] in, and They are obtained by averaging L enhanced representations based on controllable uniform noise enhancement to avoid the randomness of a single enhanced representation; and They represent the feature representations initialized uniformly from 0 to 1 to achieve uniform enhancement, α eps ∈R represents a hyperparameter that controls the size of the enhanced representation, which is adjusted by eps , we can avoid the problem that the enhanced view deviates greatly from the original view and thus discards the real semantics. The enhanced view can retain most of the information of the original representation and some variance through this data augmentation method. Based on this enhancement method, we can construct an effective and simple comparison view.

[0167] Step B52: Perform a comparative learning operation using the original image and the enhanced image. The comparative learning loss paradigm based on uniform noise enhancement at the user end and the item end is:

[0168]

[0169]

[0170] in, A similarity function is adopted, which uses the inner product as the similarity function; τ represents the temperature coefficient, u' and i' represent negative samples with different indexes respectively; based on the contrast loss defined above, the consistency between the original view and the enhanced view in each layer can be maximized, and the difference between different users and objects can be maximized, thereby enhancing the robustness of the view after the fusion of multiple behavioral relationships.

[0171] Step B53: Representing the final user preference using fused common information Relationship between user behavior and the final preference representation of the item Relationship between behaviors and items Perform contrastive learning based on feature fusion to obtain the contrast loss of users and items. and

[0172]

[0173]

[0174] in, A similarity function is adopted, which uses the inner product as the similarity function; τ represents the temperature coefficient, u' and i' represent negative samples with different indexes respectively. Based on the contrast loss defined above, the similarity between the multi-relation fusion representation view and the behavior relationship representation view can be maximized, and the difference between different users and items can be maximized, thereby enhancing the robustness of the multi-relation fusion representation view.

[0175] Step B6: Randomly negatively sample the users in a batch to generate corresponding positive sample interaction sets and negative sample interaction sets; use the pairwise loss function to calculate the prediction loss; then fuse it with multiple contrast losses to obtain the final optimized target loss, and then calculate the gradient of each parameter in the deep learning network model G through the back propagation method, and use the Adam optimization algorithm to update each parameter to constrain the loss size.

[0176] The step B6 specifically comprises the following steps:

[0177] Step B61: Sample the corresponding negative samples for a user in a small batch to form a corresponding positive sample interaction set Interacting with negative samples

[0178] Step B62: Use pairwise loss function Calculate the prediction loss, given u i Under the positive sample set and negative sample set, the prediction loss is defined as:

[0179]

[0180] in, Represents user u i The sth interactive item in the corresponding positive sample item sequence, and Represents user u i The sth uninteracted item in the corresponding negative sample item sequence, where θ represents the learnable parameters in the network, and the parameters λ1 and L2 regularization are used to alleviate the overfitting problem.

[0181] Step B63: Update the learning rate through the gradient optimization algorithm Adam, and use back propagation to iteratively update the model parameters to minimize the loss function to train the model G; merge the contrast loss obtained in steps B52 and B53 with the prediction loss obtained in step B61 to define the final recommendation loss:

[0182]

[0183] Among them, λ2 and λ3 represent the hyperparameters that control the intensity of the contrastive learning task.

[0184] Step B7: Use the final preference representation of users and items to train the model. When the loss value generated by the deep learning network model is less than the set threshold or reaches the maximum number of iterations, terminate the training of the deep learning network model G.

[0185] In this example, the present method (MMNCL) and the existing commonly used method were tested on the same platform, and the results are shown in Table 1. The test results show that the performance of the present method is better than that of the existing method.

[0186] Table 1 Comparison of test results between this method and existing methods

[0187]

[0188] This embodiment also provides a multi-behavior commonality fusion recommendation system, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the above-mentioned method steps can be implemented.

[0189] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0190] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0191] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0192] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0193] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A multi-behavior commonality fusion recommendation method, characterized in that: The following steps are involved: Step A: Obtain user behavior data, including user ID, target item ID, interaction type, and interaction time, pre-process the acquired data, and then construct a training set; Step B: construct a deep learning network model G based on the fusion of commonalities of multiple behaviors, wherein the deep learning network model G obtains common information of different types of behaviors and fuses it into the final preference representation, and assists the learning of multiple representations through multi-angle comparative learning; train the deep learning network model G using the training set, and obtain the required deep learning network model G by minimizing the loss function on the training set; Step C: Input the preprocessed recommended user and candidate item data into the trained deep learning network model G, and output the probability that the current user will generate the target behavior for the target item.

2. The multi-behavior commonality fusion recommendation method according to claim 1, characterized in that: The step A specifically comprises the following steps: Step A1: Obtain an existing general multi-behavior recommendation dataset; each sample in the multi-behavior recommendation dataset contains a user ID, a target item ID, and the type of interaction, whether an interaction occurs, and the interaction time; Step A2: Preprocess each multi-behavior recommendation dataset, including: deleting outlier data; removing users with less than 5 interactions and items with less than 10 interactions to reduce the number of users and items with fewer interactions; and forming a training set with the preprocessed multi-behavior recommendation dataset.

3. The multi-behavior commonality fusion recommendation method according to claim 1, characterized in that: The step B specifically comprises the following steps: Step B1: Divide the training set according to different behavior types and construct different types of user-item interaction graphs G k ; Step B2: Different types of user-item interaction graphs G k Input into the corresponding initialization embedding layer to obtain the user initialization embedding corresponding to the kth behavior type Initialize embed with items Based on multiple types of initialization embeddings, we get the overall initialization embedding e corresponding to the user. u The overall initialization embedding e corresponding to the item i ; Step B3: Build a behavior-aware relationship network and embed the overall initialization of the user obtained in step B2 into e u Initialize the embedding with the whole item i As the input of the first layer of the behavior-aware relation network, they are respectively input into the intra-behavior relation encoder to capture the specific contextual information within the same type, and obtain the intra-behavior relation representation of users and items. and Then all types of intra-behavior relationships of users are represented as Relationship between the behavior and the item They are input into the behavior relationship encoder to capture different types of relationship dependencies and obtain the overall behavior relationship representation corresponding to users and items. and The output of each layer is used as the input of the next layer for iterative aggregation, and the multi-layer behavior relationship representation of users and items is fused using the fully connected layer to obtain and Step B4: Based on each type of user-item interaction graph, construct the overall user-item interaction graph G of heterogeneous relationships all ; Build a multi-view meta-network based on the overall user-item interaction graph G all The relationship between behaviors obtained from each layer of the network in step B3 is used to construct a meta-feature view of each type of user at each layer. and meta-feature views of each type of item In this way, the common information of different types is captured; the different types of meta-feature views in each layer are input into the personalized meta-networks F1 and F2 respectively, so as to capture the adaptive transformation matrix; and the different types of common information in each layer are aggregated using the adaptive transformation matrix to obtain the personalized common representation of users and items at each layer. and Construct a feature debiasing network to enhance the authenticity of the extracted common representation and improve the robustness of the common representation; for the common representation of each layer obtained, fuse it based on proportional fusion to obtain a multi-layer fused common representation and The relationship between the user's behaviors obtained in step B3 is expressed as Relationship between the behavior of the object The commonality representation of the user obtained in step B4 Commonality with items Aggregate in a proportional fusion manner to obtain the final user preference representation and item preference representation Step B5: Construct a multi-angle contrastive learning paradigm, use uniform noise enhancement to generate corresponding enhanced views for the behavioral relationship representation between users and items at each layer in step B3; and perform contrastive learning with the original view to obtain the contrast loss of users and items based on uniform noise enhancement. and Utilize the user's final preference representation that integrates common information Relationship between user behavior and the final preference representation of the item Relationship between behaviors and items Perform contrastive learning based on feature fusion to obtain the contrast loss of users and items. and Step B6: Perform random negative sampling on users in a batch to generate corresponding positive sample interaction sets and negative sample interaction sets; use the pairwise loss function to calculate the prediction loss; then fuse it with multiple contrast losses to obtain the final optimization target loss, and then calculate the gradient of each parameter in the deep learning network model G through the back propagation method, and use the Adam optimization algorithm to update each parameter to constrain the loss size; Step B7: Use the final preference representation of users and items to train the model. When the loss value generated by the deep learning network model is less than the set threshold or reaches the maximum number of iterations, terminate the training of the deep learning network model G.

4. The multi-behavior commonality fusion recommendation method according to claim 3 is characterized in that: The step B2 specifically comprises the following steps: Step B21: Use the user-item interaction graph to input into the corresponding initialization embedding layer to obtain the user's specific type of initialization embedding and item-specific initialization embedding Step B22: Aggregate the initialization embeddings of each type obtained in step B21 in the form of a list to obtain the user's overall initialization embedding and the overall initialization embedding of the item 5. The multi-behavior commonality fusion recommendation method according to claim 4 is characterized in that: The step B3 specifically comprises the following steps: Step B31: Construct a behavior-aware relationship network, which includes an intra-behavior relationship encoder and an inter-behavior relationship encoder, which are constructed by a memory weighted network and a multi-head self-attention network with residual connections, respectively, and learn the multi-hop information of users and items by iterating the number of network layers, and embed the overall initialization of the user obtained in step B22 into e u and the overall initialization embedding of the item i As the first layer input of the network, and the output of each layer of the network as the input of the next layer, for the first layer, the intra-behavior relation encoder is first used to capture each type of context information to obtain the intra-behavior relation representation corresponding to the user and the item and Among them, e u With e i They are the overall initialization embeddings for users and items respectively. IntraEnc() consists of a memory weighted network, which mainly performs conversion learning between different dimensions. Step B32: Represent the behavior relationship between the user and the item obtained in step B31 and They are input into the behavior relationship encoder to capture different types of relationship dependencies and obtain different types of behavior relationship representations corresponding to users and items. and And sum to get the overall behavior relationship between users and items and Among them, InterEnc() is composed of a multi-head self-attention mechanism, which retains the original intra-behavior relationship representation information; Step B33: Repeat steps B31 and B32 for the number of layers in the network; thus obtaining a multi-layer overall behavior relationship representation list corresponding to users and items. and Step B34: Based on the multi-layer behavior relationship representation list of users and items obtained in step B33, a fully connected layer is used to aggregate the multi-layer representation: Among them, concat() represents the concatenation function, which concatenates the representations of multiple different layers, and Mlp1() and Mlp2() represent fully connected layers with activation functions.

6. The multi-behavior commonality fusion recommendation method according to claim 5, characterized in that: The step B4 specifically comprises the following steps: Step B41: Based on each type of user-item interaction graph [G 1 ,G 2 ,…,G K ], constructing the overall user-item interaction graph G with heterogeneous relationships all , which contains all types of interaction edges between users and items; Step B42: Based on the overall user-item interaction graph G obtained in step B41 all The list of relationships between users and items obtained by the l-th layer network in step B32 and Construct user- and item-specific meta-feature views respectively, and use them to capture the common behavioral information in different types: in, The target representation that represents the commonality of behaviors aggregates multiple different types of relationship representations between behaviors in the same layer through residual connections. Mlp3() represents a fully connected layer with an activation function, and concat() represents a concatenation function that concatenates multiple representations to form a specific type of meta-information view, which can be used as input for the subsequent network that captures the commonality of behaviors. Step B43: Use two adaptive meta-transformation matrix learners to respectively transform the meta-feature view lists of different types of users in each layer obtained in step B42 List of meta-traits view for different types of items Traverse and capture the adaptive transformation matrix of users and items under different types: Among them, F1 and F2 represent the meta-learners for learning the adaptive transformation matrix, which are composed of two fully connected layers with activation functions; Step B44: Use the adaptive transformation matrix obtained in step B34 to aggregate different types of common information in each layer to obtain personalized common representations of users and items in each layer: Among them, δ() represents the activation function, and through two adaptive transformation matrices, the potential common transformation representation of behavior under a specific type of representation is learned; Step B45: Use the constructed feature debiasing network to enhance the authenticity of the extracted common representation and improve the robustness of the common representation to capture the enhanced common representation of users and items: Among them, Feature_Decorr() represents the feature debiasing network, which learns the unique dimension debiasing weight by removing the noise between dimensions, and enhances the original commonality representation in the form of residual connection; Step B46: For the enhanced commonality representation of users and items in each layer obtained in step B45, they are fused by proportional fusion to obtain a multi-layer fused enhanced commonality representation, and then passed through a fully connected layer to obtain an aggregated behavior commonality representation: Among them, Mlp4 represents the fully connected layer with activation function, α m Represents the hyperparameter that controls the ratio of fusion in each layer; Step B47: Represent the relationship between the user's behaviors obtained in step B3 Relationship between behavior and items The commonality representation of the user obtained in step B4 Commonality with items Aggregate in a proportional fusion manner to obtain the final user preference representation and item preference representation Among them, α f represents a hyperparameter that controls the ratio of commonality representation fusion.

7. The multi-behavior commonality fusion recommendation method according to claim 6, characterized in that: The step B5 specifically comprises the following steps: Step B51: Generate a corresponding enhanced view for the relationship representation between the user and the item behavior at each layer in step B3 using a uniform noise enhancement method and in, and They are obtained by averaging L enhanced representations based on controllable uniform noise enhancement to avoid the randomness of a single enhanced representation; and They represent the feature representations initialized uniformly from 0 to 1 to achieve uniform enhancement, α eps ∈R represents a hyperparameter that controls the size of the enhanced representation; Step B52: Perform a comparative learning operation using the original image and the enhanced image. The comparative learning loss paradigm based on uniform noise enhancement at the user end and the object end is: in, A similarity function is used, which uses the inner product as the similarity function; τ represents the temperature coefficient, and u' and i' represent negative samples with different indexes respectively; based on the contrast loss defined above, the consistency between the original view and the enhanced view in each layer can be maximized, and the difference between different users and objects can be maximized, thereby enhancing the robustness of the view after the fusion of multiple behavioral relationships; Step B53: Representing the final user preference using fused common information Relationship between user behavior and the final preference representation of the item Relationship between behaviors and items Perform contrastive learning based on feature fusion to obtain the contrast loss of users and items. and in, A similarity function is adopted, which uses the inner product as the similarity function; τ represents the temperature coefficient, u' and i' represent negative samples with different indexes respectively. Based on the contrast loss defined above, the similarity between the multi-relation fusion representation view and the behavior relationship representation view can be maximized, and the difference between different users and items can be maximized, thereby enhancing the robustness of the multi-relation fusion representation view.

8. The multi-behavior commonality fusion recommendation method according to claim 7, characterized in that: The step B6 specifically comprises the following steps: Step B61: Sample the corresponding negative samples for a user in a small batch to form a corresponding positive sample interaction set Interacting with negative samples Step B62: Use pairwise loss function Calculate the prediction loss, given u i Under the positive sample set and negative sample set, the prediction loss is defined as: in, Represents user u i The sth interactive item in the corresponding positive sample item sequence, and Represents user u i The sth uninteracted item in the corresponding negative sample item sequence, where θ represents the learnable parameters in the network, and the parameter λ1 and L2 regularization are used to alleviate the overfitting problem; Step B63: Update the learning rate through the gradient optimization algorithm Adam, and use back propagation to iteratively update the model parameters to minimize the loss function to train the model G; merge the contrast loss obtained in steps B52 and B53 with the prediction loss obtained in step B61 to define the final recommendation loss: Among them, λ2 and λ3 represent the hyperparameters that control the intensity of the contrastive learning task.

9. A multi-behavior commonality fusion recommendation system, characterized in that: The method comprises a memory, a processor, and computer program instructions stored in the memory and executable by the processor. When the processor executes the computer program instructions, the method steps as described in any one of claims 1 to 8 can be implemented.

Images