Recommendation method and device based on decoupling learning and target behavior oriented learning

Abstract

The invention discloses a recommendation method, and particularly relates to a recommendation method and device based on decoupling learning and target behavior oriented learning. The method comprises the following steps: constructing a user project interaction graph under behaviors based on interaction conditions of users and projects under multiple behaviors; performing decoupling attribute domain representation processing on the user item interaction diagram to obtain item decomposition embedding corresponding to item nodes and user decomposition embedding corresponding to user nodes in a plurality of decoupling attribute domains; for a plurality of attribute domains corresponding to the user item interaction diagram, calculating attention scores between user nodes and adjacent item nodes under the attribute domains based on item decomposition embedding and user decomposition embedding; performing aggregation processing on the user node and the project node based on the attention score to obtain user target embedding and project target embedding under the behavior; and performing analysis processing by combining the user target embedding and the project target embedding under the plurality of behaviors to obtain a recommendation result. According to the method, a more accurate recommendation result can be obtained.

Description

Technical Field

[0001] The present invention relates to a recommendation method, and in particular to a recommendation method and device based on decoupled learning and target behavior oriented learning. Background Art

[0002] In recommendation systems, leveraging various user behaviors, such as clicks, favorites, and purchases, is crucial to alleviating the sparsity problem inherent in single behavior data. Multi-behavior recommendation refers to a method that considers multiple user behaviors (such as clicks, purchases, comments, etc.) in recommendation systems. This method can more comprehensively understand user interests and preferences, thereby improving recommendation accuracy and user satisfaction.

[0003] Existing multi-behavior recommendation methods mainly use graph network methods to model the semantic interactions between users and items under different behaviors. These graph network-based methods obtain the final user (item) embedding representation by modeling the interactive relationship between users and items in different behaviors and aggregating the embeddings learned from different user-item behaviors. Some other methods use contrastive learning to further improve model performance by adjusting user preferences between target behaviors and auxiliary behaviors. Although combining contrastive learning with graph neural networks significantly improves the performance of multi-behavior recommendation, multi-behavior recommendation still faces two main problems:

[0004] First, these methods do not consider the relationship between users' target behaviors and auxiliary behaviors for preferences of specific item attributes. Fine-grained item attributes reflect users' preferences, and there are correlations between multiple behaviors for specific item attributes.

[0005] Secondly, when learning user preferences for target behaviors, existing methods ignore the noisy data in auxiliary behaviors. The noisy data in auxiliary behaviors are mainly reflected in some item attributes that are irrelevant to the target user behavior. Even some items that seem to have no relevance to the target behavior still provide useful information for predicting the target behavior of specific item attributes.

[0006] In actual application scenarios, auxiliary behaviors such as clicks and add-to-cart that are inconsistent with the target purchase behavior have a negative impact on the recommendation accuracy of the target user's behavioral preferences. The commonly used methods in the existing technology usually only learn coarse-grained user preferences and fail to take into account the subtle attributes of the items that users pay attention to in different behaviors, resulting in poor accuracy of recommendation results. Summary of the invention

[0007] The technical problem to be solved by the present invention is that the above-mentioned methods commonly used in the prior art usually only learn coarse-grained user preferences, and fail to take into account the subtle attributes of the items that users pay attention to in different behaviors, resulting in poor accuracy of recommendation results. In order to solve the above-mentioned problem, the present invention provides a recommendation method and device based on decoupled learning and target behavior-oriented learning.

[0008] The content of the present invention includes:

[0009] In a first aspect, an embodiment of the present invention provides a recommendation method based on decoupled learning and target behavior-oriented learning, comprising:

[0010] Based on the interaction between the user and the project under the multiple behaviors, construct a user-project interaction graph under each of the behaviors, wherein the user-project interaction graph includes a user node for representing the user and a project node for representing the project, and the multiple behaviors include a target behavior and an auxiliary behavior;

[0011] Performing decoupled attribute domain representation processing on the user-project interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains;

[0012] For each of the plurality of attribute domains corresponding to the user-item interaction graph, based on the item decomposition embedding and the user decomposition embedding, calculating an attention score between the user node and its adjacent item nodes under each of the attribute domains;

[0013] Aggregate the user node and the item node based on the attention score to obtain a user goal embedding and an item goal embedding under each of the behaviors;

[0014] The user goal embedding and the item goal embedding under the multiple behaviors are combined for analysis and processing to obtain a recommendation result.

[0015] Optionally, performing decoupled attribute domain representation processing on the user-item interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains includes:

[0016] Constructing a plurality of decoupled attribute domains, and obtaining the user initial embedding corresponding to the user node and the project initial embedding corresponding to the project node;

[0017] The item initial embedding is decomposed in each of the attribute domains to obtain the item decomposed embedding in the attribute domain, and the user decomposed embedding in each of the attribute domains is determined based on the user initial embedding.

[0018] Optionally, decomposing the initial embedding of the item in each of the attribute domains to obtain the decomposed embedding of the item in the attribute domain includes:

[0019] In each of the attribute domains, the initial embedding of the item is projected into the embedding of the attribute domain using the pre-learned projection matrix to obtain the item decomposition embedding in the attribute domain. The item decomposition embedding in the attribute domain is:

[0020]

[0021] in, The item decomposition embedding e is used to represent item node i in attribute domain a. i The initial embedding of the item used to characterize item node i, Used to characterize the projection matrix, which is a learnable parameter matrix, and ||||2 is used to characterize the L2 norm.

[0022] Optionally, the calculating, based on the item decomposition embedding and the user decomposition embedding, an attention score between the user node and its adjacent item nodes in each attribute domain comprises:

[0023] Determine an average of the item decomposition embeddings of all adjacent item nodes corresponding to the user node in the attribute domain as a query vector corresponding to the user node;

[0024] The query vector corresponding to the user node is used to calculate the attention score between the user node and its adjacent item nodes in the attribute domain.

[0025] Optionally, the attention score between the user node and its adjacent item node is:

[0026]

[0027] in, It is used to represent the query vector corresponding to the user node u in the attribute domain a under the target behavior. are pre-learned parameters, [;] are used to represent the concatenation of vectors, Item decomposition embedding used to characterize item node i under attribute domain a.

[0028] Optionally, the aggregating the user node and the item node based on the attention score to obtain a user goal embedding and an item goal embedding under each of the behaviors includes:

[0029] Based on the attention score, the user node and its adjacent project nodes are aggregated to obtain a user aggregate embedding, and the project node and its adjacent user nodes are aggregated to obtain a project aggregate embedding;

[0030] The user aggregate embeddings in all the attribute domains under the same user-item interaction graph are added together to obtain the user target embedding under the behavior, and the project aggregate embeddings in all the attribute domains under the same user-item interaction graph are added together to obtain the project target embedding under the behavior.

[0031] Optionally, the item is a commodity, the target behavior is purchase, and the auxiliary behavior includes at least one of click, collection and comment;

[0032] The interaction between the user and the item under the multiple behaviors is used to characterize the situation in which the user performs the target behavior and the auxiliary behavior on the commodity, and the attribute domain is used to characterize the attributes of the commodity.

[0033] In a second aspect, an embodiment of the present invention provides a recommendation device based on decoupled learning and target behavior-oriented learning, including:

[0034] A construction module, configured to construct a user-item interaction graph under each of the behaviors based on the interaction between the user and the item under the multiple behaviors, wherein the user-item interaction graph includes a user node for representing the user and an item node for representing the item, and the multiple behaviors include a target behavior and an auxiliary behavior;

[0035] A decoupled representation module, used to perform decoupled attribute domain representation processing on the user-project interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains;

[0036] an attention calculation module, configured to calculate, for each of the plurality of attribute domains corresponding to the user-item interaction graph, an attention score between the user node and its adjacent item nodes under each of the attribute domains based on the item decomposition embedding and the user decomposition embedding;

[0037] An aggregation processing module, used for performing aggregation processing on the user node and the project node based on the attention score to obtain a user goal embedding and a project goal embedding under each of the behaviors;

[0038] The analysis and processing module is used to analyze and process the user goal embedding and the project goal embedding under multiple behaviors to obtain recommendation results.

[0039] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps in the recommendation method based on decoupled learning and target behavior-oriented learning as described in the first aspect.

[0040] In a fourth aspect, an embodiment of the present invention provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the recommendation method based on decoupled learning and target behavior-oriented learning as described in the first aspect.

[0041] The beneficial effect of the present invention is that in this embodiment, the user-item interaction under different behaviors is characterized in multiple decoupled attribute domains, and the user's preferences are reflected through fine-grained item attributes. The user's preferences for different item attributes are learned based on a decoupled graph convolutional network to capture the fine-grained semantic interactions between users and items in different attribute spaces. Through the method provided in this embodiment, the subtle attributes of the items that users pay attention to in different behaviors are understood, which can improve the understanding of user preferences and improve the quality of recommendations in complex behavioral environments, thereby obtaining more accurate recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Attached Figure 1 A flowchart of a recommendation method based on decoupled learning and target behavior oriented learning provided by an embodiment of the present invention;

[0043] Attached Figure 2 A diagram of the architecture of a recommendation method based on decoupled learning and target behavior-oriented learning provided by an embodiment of the present invention;

[0044] Attached Figure 3 A schematic diagram of a user-project interaction diagram acquisition process provided by an embodiment of the present invention;

[0045] Attached Figure 4a A schematic diagram of a decoupled graph convolution part provided by an embodiment of the present invention;

[0046] Attached Figure 4b for Figure 4a A partial enlarged schematic diagram of the corresponding part of the attribute domain 1;

[0047] Attached Figure 4c for Figure 4a A local enlarged schematic diagram of the corresponding part of the attribute domain K;

[0048] Attached Figure 5 A schematic diagram of a decoupled contrast learning part provided by an embodiment of the present invention;

[0049] Attached Figure 6 A schematic diagram of a recommendation device based on decoupled learning and target behavior oriented learning provided by an embodiment of the present invention;

[0050] Attached Figure 7 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In the embodiments of the present application, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In the embodiments of the present application, the term "multiple" refers to two or more, and other quantifiers are similar. The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first" and "second" are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] See also Figure 1 , Figure 1 A schematic diagram of a recommendation method based on decoupled learning and target behavior-oriented learning provided by an embodiment of the present invention, wherein the method specifically comprises the following steps:

[0054] Step 101, based on the interaction between users and items under multiple behaviors, construct a user-item interaction graph under each of the behaviors, wherein the user-item interaction graph includes a user node for representing the user and an item node for representing the item, and the multiple behaviors include a target behavior and an auxiliary behavior;

[0055] Step 102, performing decoupled attribute domain representation processing on the user-item interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains;

[0056] Step 103, for each of the plurality of attribute domains corresponding to the user-item interaction graph, based on the item decomposition embedding and the user decomposition embedding, calculating the attention score between the user node and its adjacent item nodes under each of the attribute domains;

[0057] Step 104, performing aggregation processing on the user node and the item node based on the attention score to obtain a user goal embedding and an item goal embedding under each of the behaviors;

[0058] Step 105: analyzing and processing the user goal embedding and the item goal embedding under the multiple behaviors to obtain a recommendation result.

[0059] See also Figure 2 The embodiment of the present application also provides a recommendation method framework based on decoupled learning and target behavior-oriented learning, which specifically includes four modules: a decoupled attribute domain representation model for implementing step 102, a target behavior-guided attention mechanism module for implementing step 103, a decoupled graph attention network module for implementing step 103, and a denoising contrast learning module for training.

[0060] The following will introduce the different steps or modules in detail. In the specific implementation, user embedding can also be called user features or user representations, and item embedding can also be called item features or item representations. As a specific embodiment, the item can be the item targeted by the user's behavior, for example, when the behavior is purchase, the item is the purchased item.

[0061] See also Figure 3 In step 101, the interaction between users and items under multiple behaviors is first obtained. In a specific embodiment, the number of users and items can be multiple. The multiple behaviors include target behaviors and auxiliary behaviors. In some embodiments, the number of auxiliary behaviors is one or more. For example, in Figure 3 In the specific embodiment shown, the behaviors include purchase, add-to-cart and click, wherein purchase is the target behavior and add-to-cart and click are auxiliary behaviors.

[0062] In step 101, a user-item interaction graph under different behaviors is established respectively. In the user-item interaction graph under the behavior, if a user interacts with an item under the behavior (for example, user A purchases item A, that is, user A and item A interact under the purchase behavior), the user node representing the user and the item node representing the item are connected. Figure 3 In the specific embodiment shown, corresponding user-item interaction graphs are obtained for the three behaviors of purchase, add-to-cart and click.

[0063] In step 102, the project nodes and user nodes in the user-project interaction graph are projected into multiple decoupled attribute domains to characterize the attractiveness of different project attributes to users under different behaviors. The correlation between user preferences in different behaviors is mainly reflected in different project attributes, and the noise data in auxiliary behaviors is also closely related to project attributes. Through this step, we can pay attention to the user's preferences for project attributes under different behaviors, so focusing on specific project attributes can obtain more accurate embedding.

[0064] Optionally, in some embodiments, step 102 includes:

[0065] Constructing a plurality of decoupled attribute domains, and obtaining the user initial embedding corresponding to the user node and the project initial embedding corresponding to the project node;

[0066] The item initial embedding is decomposed in each of the attribute domains to obtain the item decomposed embedding in the attribute domain, and the user decomposed embedding in each of the attribute domains is determined based on the user initial embedding.

[0067] In this embodiment, several decoupled attribute domains are constructed to represent different project attribute domains. The attribute domain is an implicit space, which is different from the explicit project category and user attributes. It is a latent semantic space that represents implicit potential item attributes. For example, when the item is a clothing item, the attribute domain can represent the clothing's style, color, price, material and other attributes. In the specific implementation, the number of attribute domains is not limited here, and each user-item interaction diagram corresponds to multiple attribute domains.

[0068] It should be understood that, unless otherwise specified, the attribute domain referred to in each embodiment includes each attribute domain corresponding to each user-item interaction diagram (ie, all behaviors), and the steps performed in all attribute domains are the same and will not be described in detail.

[0069] In some embodiments, each user and project is associated with an ID embedding. is the embedding matrix initialized by the user embedding, Q∈R N×D is the embedding matrix initialized with item embeddings, where M represents the number of users, N represents the number of items, and D represents the embedding size. Formally, given a one-hot encoded embedding matrix for a user and a one-hot encoded embedding matrix of the items The details are as follows:

[0070]

[0071]

[0072] Among them, e uis the user initial embedding of user u, is a one-hot encoding vector corresponding to user u, e i is the item initial embedding of item i, is a one-hot encoded vector of item i.

[0073] In each attribute domain, the item decomposed embedding is obtained by decomposing the initial item embedding. Assume that A is A=1, 2, ..., |A|, A is a hyperparameter representing the number of decoupled attribute domains, and use the projection matrix to transform the above obtained e i Projected into the embedding of attribute a∈A. Specifically, in some embodiments, decomposing the initial embedding of the item in each of the attribute domains to obtain the decomposed embedding of the item in the attribute domain includes:

[0074] In each of the attribute domains, the initial embedding of the item is projected into the embedding of the attribute domain using the pre-learned projection matrix to obtain the item decomposition embedding in the attribute domain. The item decomposition embedding in the attribute domain is:

[0075]

[0076] in, The item decomposition embedding e is used to represent item node i in attribute domain a. i The initial embedding of the item used to characterize item node i, is used to characterize the projection matrix, the projection matrix is ​​a learnable parameter matrix, and || ||2 is used to characterize the L2 norm. In some embodiments, is the projection matrix of the attribute domain shared by all projects.

[0077] It should be understood that in this embodiment, the initial embedding of the input graph attention network is the same as the item decomposition embedding, namely:

[0078]

[0079] Optionally, in some embodiments, the specific manner of determining the user decomposed embedding in each attribute domain based on the user initial embedding is as follows:

[0080] Using the transformation matrix Get the user decomposition embedding and use it as the initial embedding of the input graph attention network as follows:

[0081]

[0082] Since different attribute domains should contain different information about item attributes, otherwise multiple item embeddings degenerate into equivalent to one item embedding, it is necessary to separate the item embeddings in different attribute domains. In order to achieve better model capacity and interpretability and avoid redundant information, independence loss is introduced in this embodiment to constrain the independence of different attribute domains and prevent them from degenerating into a single attribute domain.

[0083] See also Figure 4a-4c Specifically, in this embodiment, mutual information can be used to encourage the separation of item decomposition embeddings in different attribute domains. For each item i, the mutual score is as follows:

[0084]

[0085] where s(·) is a function that measures the similarity of item decomposition embeddings corresponding to two identical items in different attribute domains, τ represents the temperature in the softmax function as a hyperparameter, and A is the set of attribute domains.

[0086] In some embodiments, s(·) is set to a cosine similarity function as follows:

[0087]

[0088] The final dependency loss consists of the mutual scores of all items, which is calculated as follows:

[0089]

[0090] like Figure 2 As shown, in step 103, an attention mechanism based on target behavior guided learning is designed for each attribute domain, so as to modify the message passing of the graph neural network according to the user's preference for the target behavior. For all adjacent project nodes of a user in a specific attribute domain, if the project embedding in the attribute domain is closer to the user's target behavior preference, the information transmitted by the project node to the user node in the attribute domain should have a higher importance weight than other adjacent project nodes.

[0091] Optionally, in some embodiments, step 103 includes:

[0092] Determine an average of the item decomposition embeddings of all adjacent item nodes corresponding to the user node in the attribute domain as a query vector corresponding to the user node;

[0093] The query vector corresponding to the user node is used to calculate the attention score between the user node and its adjacent item nodes in the attribute domain.

[0094] In this embodiment, in order to learn the attention scores of user node u and its adjacent project node i in attribute domain a, a query vector q that can reflect the target behavior preference of user node u is first learned. Specifically, in some embodiments, a query vector is calculated in each attribute domain in each user behavior, and different query vectors need to be calculated for different behaviors. The query vector is the average of all item representations that the user has interacted with, so it can represent the user's preference. The average of the item decomposition embeddings of all item nodes that interact with user node u in attribute domain a is determined as the query vector, as follows:

[0095]

[0096] in, It is used to represent the query vector corresponding to the user node u in the attribute domain a under the target behavior. Represents the nth item node adjacent to the user node u in the attribute domain a. The e and Figure 4b Corresponding to the target historical behavior in.

[0097] Then, the learned query vector can be used to calculate the attention score of the user node and its adjacent item nodes. Optionally, in some embodiments, the attention score between the user node and its adjacent item nodes is:

[0098]

[0099] in, It is used to represent the query vector corresponding to the user node u in the attribute domain a under the target behavior. are pre-learned parameters, [;] are used to represent the concatenation of vectors, Item decomposition embedding used to characterize item node i under attribute domain a.

[0100] It should be understood that in each attribute domain, the attention score of each user node and its adjacent project node can be calculated. After obtaining the attention score of each user node and its adjacent project node, it is equivalent to obtaining the attention score of each project node and its adjacent user node.

[0101] Optionally, in some embodiments, step 104 includes:

[0102] Based on the attention score, the user node and its adjacent project nodes are aggregated to obtain a user aggregate embedding, and the project node and its adjacent user nodes are aggregated to obtain a project aggregate embedding;

[0103] The user aggregate embeddings in all the attribute domains under the same user-item interaction graph are added together to obtain the user target embedding under the behavior, and the project aggregate embeddings in all the attribute domains under the same user-item interaction graph are added together to obtain the project target embedding under the behavior.

[0104] like Figure 2 As shown in the figure, after obtaining the attention scores between the user node and its adjacent project nodes, the graph attention network is used for aggregation processing. For each attribute domain, the attention guided by the target behavior is incorporated into the graph neural network to design the graph attention network (GAT), and then the GAT is used to aggregate the neighbors of the user (project) node and distinguish the attractiveness of different project attributes to users.

[0105] In some embodiments, based on the LightGCN method, an aggregation function under specific auxiliary behaviors is designed as follows:

[0106]

[0107]

[0108] in, Used to represent the neighbor nodes of user node u, It is used to represent the neighbor nodes of item node i; l is used to represent the number of layers of the graph attention network, Used to represent (l+1) th The user node u of the layer is embedded in the user decomposition of the attribute domain a. Used to represent (l+1) th The item node i of the layer is embedded in the item decomposition of the attribute domain a. It is used to represent the attention scores of user u and adjacent item node i in the attribute domain a of layer l.

[0109] The corresponding attributes of each domain under the target behavior The value of is set to 1. For the attribute domain {1,2,…,A} under the auxiliary behavior k, the graph attention network of L layers can be obtained:

[0110]

[0111]

[0112] in, is used to represent the embedding of the user node u in the attribute domain a under the auxiliary behavior k obtained through the L-layer graph attention network, It is used to represent the embedding of item node i in attribute domain a under auxiliary behavior k obtained through an L-layer graph attention network.

[0113] In the above way, the embeddings corresponding to the same user node in all attribute domains under the same behavior are summed to obtain the user goal embedding of the user node under the behavior. Similarly, the embeddings corresponding to the same project node in all attribute domains under the same behavior are summed to obtain the project goal embedding of the project node under the behavior.

[0114] In the application embodiment, a target behavior-guided graph attention mechanism is designed in the graph convolutional network to obtain a graph attention network, and the target user behavior preference can guide the learning of user-item interaction learning in auxiliary behavior. Incorporating the attention mechanism into the graph convolutional network can effectively extract the project attributes that users are interested in auxiliary behavior, while reducing the influence of irrelevant project attributes.

[0115] Optionally, in some embodiments, the combining the user goal embedding and the item goal embedding under the multiple behaviors for analysis and processing to obtain a recommendation result includes:

[0116] Determine a user final embedding based on the user target embedding under all the behaviors, and determine a project final embedding based on the project target embedding under all the behaviors;

[0117] The recommendation result is obtained by performing analysis and processing based on the user final embedding and the item final embedding.

[0118] After obtaining the user target embedding and target item embedding for each of the K behaviors {1, 2, ..., K}, the user final embedding and item final embedding for all behaviors can be further determined. In some embodiments, the user final embedding and item final embedding are:

[0119]

[0120]

[0121] in, and represents the weight matrix, and In the specific implementation, the final recommendation result can be obtained by analyzing and processing based on the final embedding of the user and the final embedding of the item. The specific process can be found in the description of the relevant technology and will not be repeated here.

[0122] This embodiment also provides a recommendation model, including the above-mentioned decoupled attribute domain representation module, the target behavior guided attention mechanism module and the decoupled graph attention network module. In the process of training the recommendation model, in order to learn the correlation between the embedding representing the user's target behavior and the auxiliary behavior, a contrastive learning method is used to make the embedding of the user's target behavior and the auxiliary behavior more consistent. Since the embedding of the auxiliary behavior contains a large amount of noise data, such as click behavior caused by the popularity of the item, these noise data will cause the noise information in the embedding of the auxiliary behavior to be used as a positive sample, reducing the ability to distinguish between positive and negative samples in contrastive learning.

[0123] In this embodiment, a denoising contrastive learning method is provided to reduce the noise information in the positive samples, thereby maximizing the effectiveness of contrastive learning in the recommendation model. Through the embedding learning module based on the graph attention network, the user embedding of the auxiliary behavior can be obtained by effectively capturing the user-item interaction of the target behavior preference, which can reduce more noise data in the auxiliary behavior compared with the graph convolutional network-based method commonly used in the prior art.

[0124] like Figure 5 As shown in Figure 1, a multi-attribute linearized attention mechanism is designed before contrastive learning, and the user goal embedding of the target behavior is regarded as Q, and the user goal embedding of the auxiliary behavior is regarded as K and V. First, the user goal embedding is projected as follows:

[0125]

[0126]

[0127]

[0128] in, The user embedding matrix representing the target behavior, The user embedding matrix representing the k-th auxiliary behavior of attribute a, Represents the parameter matrix. The number of behaviors is K, the first K-1 behaviors represent auxiliary behaviors, and the Kth behavior represents the target behavior.

[0129] Then, a linearized attention mechanism is used for the user target embedding of the kth auxiliary behavior of attribute a, and the attention weight is calculated using the following function:

[0130]

[0131] Among them, the feature map φ(.) is applied row by row to matrices Q and K. Specifically, φ(x)=elu(x)+1, elu(.) represents the exponential linear unit activation function, Is The linearized attention mechanism used in this embodiment can effectively reduce the complexity of the model.

[0132] In the above formula, when x is negative, the user goal embedding matrix of auxiliary behavior k is It can be calculated by adding a slice of the matrix with the attribute dimension as follows:

[0133]

[0134] Then, the user goal embedding matrix of the auxiliary behavior can be obtained And the user goal embedding matrix of the target behavior is

[0135] Contrastive learning is used to capture the fine-grained consistency between the user's target behavior and auxiliary behavior, achieving a deeper and more precise understanding of the user's target behavior preferences, and obtaining:

[0136]

[0137] In the above formula, τ is the temperature hyperparameter in softmax, and φ(.) represents the inner product of two vectors. The final contrast loss function is as follows:

[0138]

[0139] In this embodiment, a denoising contrastive learning method is designed to adjust the user preference between the target behavior and the auxiliary behavior. By denoising the user's preference for different item attributes in the auxiliary behavior, a higher weight is assigned to the user's features in the attribute domain that is more similar to the target behavior, thereby improving the purity of the positive sample and achieving more effective contrastive learning.

[0140] In order to optimize the current module, in some embodiments, a pairwise Bayesian personalized ranking (BPR) loss is used for calculation, which makes the similarity between associated nodes. The BPR loss function is as follows:

[0141]

[0142] Among them, O is used to represent the training sample set, σ is used to represent the sigmoid function, the superscript T is used to represent the transposition operation, and e u The embedding used to represent the user node u, e i The embedding used to represent item node i, e j The embedding used to characterize item node j.

[0143] In some embodiments, the loss function for training the recommendation model is as follows:

[0144]

[0145] Among them, α, β, and μ are hyperparameters that control the contrast loss, dependency loss, and recommendation task training loss and L2 regularization ratio, respectively.

[0146] In this embodiment, the contrastive loss setting can better utilize contrastive learning to enhance the consistency between the target user behavior and the auxiliary user behavior. In this embodiment, the importance weights of the user's preferences for different attribute domains under specific auxiliary behaviors relative to the target user behavior are also learned. By emphasizing the embedding of user behaviors in important attribute domains, the noise data in the auxiliary behavior embedding can be effectively reduced, thereby achieving fine-grained contrastive learning to reduce the impact of noise data on positive samples.

[0147] Optionally, in some embodiments, the item is a commodity, the target behavior is purchase, and the auxiliary behavior includes at least one of click, collection, and comment;

[0148] The interaction between the user and the item under the multiple behaviors is used to characterize the situation in which the user performs the target behavior and the auxiliary behavior on the commodity, and the attribute domain is used to characterize the attributes of the commodity.

[0149] The recommendation method based on decoupled learning and target behavior-oriented learning provided in the embodiment of the present invention can be used in the scenario of personalized product recommendation for users. The item can be the product targeted by the user's behavior, and the behavior performed by the user can be purchase, click, favorite, comment, etc. In this embodiment, purchase is determined as the target behavior, and other behaviors except purchase are determined as auxiliary behaviors. Through historical data, data on various operations performed by users on multiple products (i.e., the interaction between users and items under multiple behaviors) can be obtained.

[0150] In the scenario of personalized product recommendation, it is usually necessary to make personalized product recommendations to users based on their historical behaviors. However, the fact that users click, comment, and collect products does not necessarily mean that they have a purchase preference for the product. The commonly used recommendation methods in the prior art only learn coarse-grained user preferences and fail to take into account the subtle attributes of products that users pay attention to in different behaviors. Therefore, when personalized product recommendations are made to users with purchase as the target behavior, auxiliary behaviors such as clicks, comments, and collections in the user's historical data may have a negative impact on the recommendation results.

[0151] Through the method provided in this embodiment, the user preference between the target behavior and the auxiliary behavior can be adjusted, and the product attributes that the user is interested in the auxiliary behavior can be effectively extracted, while reducing the influence of irrelevant product attributes, thereby capturing the fine-grained semantic interaction between users and products in different attribute spaces, and improving the accuracy of the recommended results.

[0152] See also Figure 6 The embodiment of the present invention further provides a recommendation device 600 based on decoupled learning and target behavior oriented learning, comprising:

[0153] A construction module 601 is used to construct a user-item interaction graph under each of the behaviors based on the interaction between the user and the item under the multiple behaviors, wherein the user-item interaction graph includes a user node for representing the user and an item node for representing the item, and the multiple behaviors include a target behavior and an auxiliary behavior;

[0154] A decoupled representation module 602 is used to perform decoupled attribute domain representation processing on the user-item interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains;

[0155] An attention calculation module 603 is used to calculate, for each of the plurality of attribute domains corresponding to the user-item interaction graph, an attention score between the user node and its adjacent item nodes under each of the attribute domains based on the item decomposition embedding and the user decomposition embedding;

[0156] Aggregation processing module 604, used to perform aggregation processing on the user node and the item node based on the attention score to obtain a user goal embedding and an item goal embedding under each of the behaviors;

[0157] The analysis and processing module 605 is used to analyze and process the user goal embedding and the item goal embedding under the multiple behaviors to obtain a recommendation result.

[0158] Optionally, the decoupling characterization module 602 includes:

[0159] A first processing unit is used to construct a plurality of decoupled attribute domains, and obtain a user initial embedding corresponding to the user node and a project initial embedding corresponding to the project node;

[0160] A decomposition unit is used to decompose the item initial embedding in each of the attribute domains to obtain the item decomposed embedding in the attribute domain, and determine the user decomposed embedding in each of the attribute domains based on the user initial embedding.

[0161] Optionally, the decomposition unit is specifically used for:

[0162] In each of the attribute domains, the initial embedding of the item is projected into the embedding of the attribute domain using the pre-learned projection matrix to obtain the item decomposition embedding in the attribute domain. The item decomposition embedding in the attribute domain is:

[0163]

[0164] in, The item decomposition embedding e is used to represent item node i in attribute domain a. i The initial embedding of the item used to characterize item node i, Used to characterize the projection matrix, which is a learnable parameter matrix, and ||||2 is used to characterize the L2 norm.

[0165] Optionally, the attention calculation module 603 includes:

[0166] A first determining unit, configured to determine an average value of item decomposition embeddings of all adjacent item nodes corresponding to the user node in the attribute domain as a query vector corresponding to the user node;

[0167] A calculation unit is used to calculate an attention score between the user node and its adjacent item nodes in the attribute domain using the query vector corresponding to the user node.

[0168] Optionally, the attention score between the user node and its adjacent item node is:

[0169]

[0170] in, It is used to represent the query vector corresponding to the user node u in the attribute domain a under the target behavior. are pre-learned parameters, [;] are used to represent the concatenation of vectors, Item decomposition embedding used to characterize item node i under attribute domain a.

[0171] Optionally, the aggregation processing module 604 includes:

[0172] an aggregation unit, configured to aggregate the user node and its adjacent project nodes based on the attention score to obtain a user aggregate embedding, and to aggregate the project node and its adjacent user nodes to obtain a project aggregate embedding;

[0173] An adding unit is used to add the user aggregate embeddings in all the attribute domains under the same user-project interaction graph to obtain the user target embedding under the behavior, and to add the project aggregate embeddings in all the attribute domains under the same user-project interaction graph to obtain the project target embedding under the behavior.

[0174] Optionally, the analysis and processing module 605 includes:

[0175] A second determining unit, configured to determine a user final embedding based on the user target embedding under all the behaviors, and to determine a project final embedding based on the project target embedding under all the behaviors;

[0176] An analysis and processing unit is used to perform analysis and processing based on the user final embedding and the item final embedding to obtain the recommendation result.

[0177] Optionally, in some embodiments, the item is a commodity, the target behavior is purchase, and the auxiliary behavior includes at least one of click, collection, and comment;

[0178] The interaction between the user and the item under the multiple behaviors is used to characterize the situation in which the user performs the target behavior and the auxiliary behavior on the commodity, and the attribute domain is used to characterize the attributes of the commodity.

[0179] The recommendation device 600 based on decoupled learning and target behavior-oriented learning provided in the embodiment of the present application can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated in this embodiment.

[0180] It should be noted that the division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0181] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0182] like Figure 7 As shown, an embodiment of the present application provides an electronic device 700, including: a memory 702, a processor 701, and a program stored in the memory 702 and executable on the processor 701; the processor 701 is used to read the program in the memory 702 to implement the steps in the recommendation method based on decoupled learning and target behavior-oriented learning as described above.

[0183] An embodiment of the present application also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, the various processes of the above-mentioned recommendation method embodiment based on decoupled learning and target behavior-oriented learning are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here. Among them, the readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical storage (such as compact disk (CD), digital video disc (DVD), Blu-ray Disc (BD), high-definition versatile disc (HVD), etc.), and semiconductor memory (such as read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read only memory (EEPROM), non-volatile memory (NAND FLASH), solid-state drive (Solid State Disk or Solid State Drive, SSD)), etc.

[0184] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0185] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0186] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A recommendation method based on decoupled learning and target behavior oriented learning, characterized in that: include: Based on the interaction between the user and the project under the multiple behaviors, construct a user-project interaction graph under each of the behaviors, wherein the user-project interaction graph includes a user node for representing the user and a project node for representing the project, and the multiple behaviors include a target behavior and an auxiliary behavior; Performing decoupled attribute domain representation processing on the user-project interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains; For each of the plurality of attribute domains corresponding to the user-item interaction graph, based on the item decomposition embedding and the user decomposition embedding, calculating an attention score between the user node and its adjacent item nodes under each of the attribute domains; Aggregate the user node and the item node based on the attention score to obtain a user goal embedding and an item goal embedding under each of the behaviors; The user goal embedding and the item goal embedding under the multiple behaviors are combined for analysis and processing to obtain a recommendation result.

2. The method according to claim 1, characterized in that: The step of performing decoupled attribute domain representation processing on the user-project interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains includes: Constructing a plurality of decoupled attribute domains, and obtaining the user initial embedding corresponding to the user node and the project initial embedding corresponding to the project node; The item initial embedding is decomposed in each of the attribute domains to obtain the item decomposed embedding in the attribute domain, and the user decomposed embedding in each of the attribute domains is determined based on the user initial embedding.

3. The method according to claim 2, characterized in that: Decomposing the initial embedding of the item in each of the attribute domains to obtain the decomposed embedding of the item in the attribute domain includes: In each of the attribute domains, the initial embedding of the item is projected into the embedding of the attribute domain using the pre-learned projection matrix to obtain the item decomposition embedding in the attribute domain. The item decomposition embedding in the attribute domain is: in, The item decomposition embedding e is used to represent item node i in attribute domain a. i The initial embedding of the item used to characterize item node i, Used to characterize the projection matrix, the projection matrix is ​​a learnable parameter matrix, and ||||2 is used to characterize the L2 norm.

4. The method according to claim 1, characterized in that: The calculating, based on the item decomposition embedding and the user decomposition embedding, an attention score between the user node and its adjacent item nodes in each attribute domain comprises: Determine an average of the item decomposition embeddings of all adjacent item nodes corresponding to the user node in the attribute domain as a query vector corresponding to the user node; The query vector corresponding to the user node is used to calculate the attention score between the user node and its adjacent item nodes in the attribute domain.

5. The method according to claim 4, characterized in that: The attention score between the user node and its adjacent item node is: in, It is used to represent the query vector corresponding to the user node u in the attribute domain a under the target behavior. are pre-learned parameters, [;] are used to represent the concatenation of vectors, Item decomposition embedding used to characterize item node i under attribute domain a.

6. The method according to claim 1, characterized in that: The aggregating the user node and the item node based on the attention score to obtain a user goal embedding and an item goal embedding under each of the behaviors includes: Based on the attention score, the user node and its adjacent project nodes are aggregated to obtain a user aggregate embedding, and the project node and its adjacent user nodes are aggregated to obtain a project aggregate embedding; The user aggregate embeddings in all the attribute domains under the same user-item interaction graph are added together to obtain the user target embedding under the behavior, and the project aggregate embeddings in all the attribute domains under the same user-item interaction graph are added together to obtain the project target embedding under the behavior.

7. The method according to claim 1, characterized in that: The item is a commodity, the target behavior is purchase, and the auxiliary behavior includes at least one of click, collection and comment; The interaction between the user and the item under the multiple behaviors is used to characterize the situation in which the user performs the target behavior and the auxiliary behavior on the commodity, and the attribute domain is used to characterize the attributes of the commodity.

8. A recommendation device based on decoupled learning and target behavior oriented learning, characterized in that: include: A construction module, configured to construct a user-item interaction graph under each of the behaviors based on the interaction between the user and the item under the multiple behaviors, wherein the user-item interaction graph includes a user node for representing the user and an item node for representing the item, and the multiple behaviors include a target behavior and an auxiliary behavior; A decoupled representation module, used to perform decoupled attribute domain representation processing on the user-project interaction graph to obtain project decomposition embeddings corresponding to the project nodes and user decomposition embeddings corresponding to the user nodes in multiple decoupled attribute domains; an attention calculation module, configured to calculate, for each of the plurality of attribute domains corresponding to the user-item interaction graph, an attention score between the user node and its adjacent item nodes under each of the attribute domains based on the item decomposition embedding and the user decomposition embedding; An aggregation processing module, used for performing aggregation processing on the user node and the project node based on the attention score to obtain a user goal embedding and a project goal embedding under each of the behaviors; The analysis and processing module is used to analyze and process the user goal embedding and the project goal embedding under multiple behaviors to obtain recommendation results.

9. An electronic device, comprising: A memory, a processor, and a program stored in the memory and executable on the processor; wherein the processor is used to read the program in the memory to implement the steps in the recommendation method based on decoupled learning and target behavior-oriented learning as described in any one of claims 1 to 7.

10. A readable storage medium for storing a program, characterized in that: When the program is executed by a processor, the steps of the recommendation method based on decoupled learning and target behavior-oriented learning are implemented as described in any one of claims 1 to 7.

Images