Comparison collaborative filtering recommendation method based on adaptive decoupling

Abstract

The invention discloses a contrast collaborative filtering recommendation method based on adaptive decoupling, and belongs to the technical field of recommendation. According to the method, global potential intention information is modeled to enrich preference representations of users and projects, and a task of generally minimizing node similarity is completed by using adaptive weight, so that the problem of excessive smoothness is relieved. The method is mainly composed of four parts: the first part carries out graph information propagation based on a user-project interaction matrix, generates local embedding of users and projects, and captures more uniformly distributed user and project representations by using cross-layer contrast learning; in the second part, a K-means clustering algorithm is utilized to construct decoupling intention embedding of a user and a project, and final node representation is obtained by fusing information of two levels; in the third part, based on neighbor harmony scores of the nodes, the difference between the nodes is controlled by applying self-adaptive weights, and the problem of excessive smoothness is relieved; and 4, completing recommendation according to the project features, the decoupling information of the user and the interest preference. According to the method, user preferences are modeled in a finer-grained mode by decoupling intention features, node differences are controlled through self-adaptive weights, so that the problem of excessive smoothness is relieved, and recommendation accuracy is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation systems, and in particular to a comparative collaborative filtering recommendation method based on adaptive decoupling. Background Art

[0002] In the era of information explosion, the rapid development of the Internet has brought many conveniences to people, but it has also exacerbated the problem of information overload. Recommendation systems learn users' interests from historical interaction information and provide personalized recommendations, thereby helping users filter out the parts that may be of interest from massive amounts of information. Collaborative filtering is a classic recommendation algorithm that mainly predicts users' preferences for various items based on their historical behaviors (including browsing, clicking, and purchasing, etc.).

[0003] Existing collaborative filtering recommendation algorithms focus on how to effectively model users' interest preferences. With the booming development of graph neural networks, many people have tried to introduce graph neural networks into collaborative filtering, which can model interaction data as a graph (such as a user-item bipartite graph), and then iteratively propagate neighborhood information to learn effective node representations, which is called graph collaborative filtering. This technology significantly improves the model's ability to capture high-order collaborative signals. Although these methods have achieved relatively good results, they still face some problems. Most previous studies have ignored the fact that users' interactive behaviors may be affected by many potential intent factors. Recently, some methods have tried to enrich the embedding representations of users and items using decoupled intent factors. Although these methods have achieved good results, they still have many limitations, namely, they do not take into account the fact that unreasonable integration of additional modeling factors may aggravate the over-smoothing problem.

[0004] Therefore, how to better model and decouple intention factors from a global perspective and reduce the harm caused by unreasonable integration is an issue that needs to be addressed urgently. Summary of the invention

[0005] In order to solve the above problems, the present invention extracts local and global supervisory signals in a cross-view manner and adaptively controls the differences between nodes to alleviate the over-smoothing problem. A contrastive collaborative filtering recommendation method based on adaptive decoupling is designed and invented to predict the user's preference for candidate items.

[0006] The technical solution adopted by the present invention is:

[0007] A comparative collaborative filtering recommendation method based on adaptive decoupling, the method comprising the following steps:

[0008] Step 1: Based on the user-item interaction matrix in the dataset Learning local embeddings for users and items via graph convolutional network embedding propagation and neighbor aggregation u and ei In order to promote a more uniform distribution of user and item representations, cross-layer contrastive learning is used to enhance the supervision signal, and the embeddings of the first and last layers are compared based on the principle of maximizing mutual information.

[0009] Step 2: Cluster the local embeddings at the user and item levels and regard the cluster centers as intent prototypes with global context. Using local embedding u With each intent prototype c u The correlation score between them is used to calculate the final decoupled intention representation h u Finally, the final user and item node representation s is obtained by fusing the information from the two levels. u and i .

[0010] Step 3: The clustering algorithm will cause users with similar historical interaction behaviors to have basically the same decoupled information representation. At the same time, unreasonable fusion of additional modeling information may aggravate the over-smoothing phenomenon. Special nodes are identified based on the degree of harmony between the node and its complete neighbor information, and different weights are given to adaptively control the differences between nodes to alleviate the over-smoothing problem.

[0011] Step 4: Complete the recommendation based on the item features, user’s decoupled information and interest preferences. For each candidate item, the final recommendation probability is based on its feature s i and the current user's embeds u , process it through the softmax function to obtain the final recommendation probability

[0012] The comparative collaborative filtering recommendation method based on adaptive decoupling provided by the present invention has the following beneficial effects:

[0013] (1) The present invention decouples user-item interactions and more accurately captures the user's true behavioral intentions during the interactive behavior, thereby achieving better recommendation effects.

[0014] (2) The present invention adaptively controls the differences between nodes by adjusting the degree of harmony between the node and its complete neighbor information, thereby reducing the impact of over-smoothing on model representation learning.

[0015] (3) The present invention effectively extracts local and global supervisory signals from two perspectives. When the recommendation scene data is relatively sparse, the problem of data sparsity is solved by supplementing additional modeling factors, thereby improving the recommendation performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0017] Figure 1 A schematic diagram of the overall structure of a model of a comparative collaborative filtering recommendation method based on adaptive decoupling provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0019] Given a user-item interaction matrix If user u has purchased item i before, then otherwise According to the interaction matrix To predict the possibility of candidate user u purchasing item i Collaborative filtering algorithms assume that if two users behave similarly, they are more likely to have similar interests. Figure 1 As shown, the following is a detailed introduction to the system process:

[0020] (1) Local collaborative relationship encoding

[0021] First, the user-item interaction matrix It is constructed as a user-item bipartite graph. Following the LightGCN method, the nonlinear activation and feature transformation parts are discarded during the propagation process to obtain the user and item feature embedding vectors for each layer. and

[0022]

[0023] in and represents the neighbor set of the target user u and item i, which can offset the impact of node degree imbalance. In order to improve the feature embedding vector, we use multi-layer embedding propagation based on graph neural network and add the node embeddings of all layers to obtain the final user and item local embedding e u and e i :

[0024]

[0025] In order to promote a more uniform distribution of user and item representations, we use cross-layer contrastive learning to enhance the supervision signal. We select the embeddings of the first and last layers for comparison, which represent the results before and after aggregating neighborhood information, respectively. Finally, we construct a contrastive loss function based on the mutual information maximization principle:

[0026]

[0027] Where s(·) represents the cosine similarity function, which is used to measure the similarity between two embeddings; τ1 is the temperature in softmax, which is used to control the smoothness of the softmax curve.

[0028] (2) Global Decoupled Multi-Intent Encoding

[0029] The interaction between users and items is often affected by latent intent factors. We assume that users with similar intents tend to exhibit similar behaviors, so we use the K-means clustering algorithm to learn the intent prototype representation of users and items, and regard the center of the cluster as the intent prototype with global context:

[0030]

[0031] Among them C k represents the kth cluster obtained by clustering, Represents the kth intention prototype from the user’s perspective. Considering that a single user’s behavior may be affected by multiple intentions, we use the local embedding e u With each intent prototype c u The correlation score between them is used to calculate the final intent representation:

[0032]

[0033] Each user u has two embeddings from different perspectives, namely, e with local dependencies. u and h with global dependencies u , the two are combined by combining the weight α to obtain the final representation of the node:

[0034] s u =(1-α)e u +αh u

[0035] (3) Adaptive Weight Difference Module

[0036] In order to prevent the over-smoothing of the supervisory signal due to unreasonable decoupled representation combinations, the model proposes a method to dynamically adjust the learning of embedded representations, that is, by comparing the consistency between the embedded information and its neighboring information, adaptively determining which nodes are more likely to cause over-smoothing. Our goal is to minimize the following loss function:

[0037]

[0038] The loss consists of two parts: user-node similarity and item-node similarity. Where g(·) represents the similarity function between each node and all other nodes. The specific calculation formula is as follows:

[0039]

[0040] Among them, the hyperparameter τ2 is used to adjust the neighborhood harmony score, and ω u Controls the node similarity weight of user u, which is obtained according to the harmony score between the node and the neighbor information:

[0041]

[0042] Among them, 0<ε<1 is a hyperparameter, s a is the harmony score between the original embedding of node a and its neighbor node information, specifically:

[0043]

[0044] Among them, e a , e a′ denotes the initialized user and item embeddings, and sig(·) denotes the sigmoid activation function. We explore the neighborhood harmony score of the central node a by summing up the k-order neighbor information of the node. The higher the neighborhood harmony score, the stronger the structural consistency between the node and its neighboring nodes.

[0045] (4) Model training

[0046] During the training phase, the interaction probability between users and items is calculated based on the final embeddings of users and items, and the softmax function is applied to obtain the final recommendation probability of each candidate item:

[0047]

[0048] The larger the interaction preference score, the greater the possibility that user u will interact with item i. At the same time, paired marginal loss is used to optimize the recommendation task, specifically:

[0049]

[0050] in, represents all items that user u has interacted with before, and i′ represents negative samples that have not been interacted with.

[0051] The above contents are only preferred embodiments of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principles of the present invention. These improvements and modifications should all be included in the protection scope of the present invention.

Claims

1. A comparative collaborative filtering recommendation method based on adaptive decoupling, characterized in that: The method comprises the following steps: In step 1, based on the user-item interaction matrix in the dataset, local embeddings of users and items are learned through graph convolutional network embedding propagation and neighbor aggregation, and cross-layer contrastive learning is used to capture more evenly distributed representations. In step 2, the local embeddings of users and items are clustered to obtain the decoupled intent embedding at the global level, and the final user and item node representation is obtained by fusing the information from the two levels. Step 3: Identify special nodes based on the degree of harmony between the nodes and their complete neighbor information, and adaptively control the differences between nodes through different weights to alleviate the over-smoothing problem. Step 4: Complete the recommendation based on the item features, user’s decoupled information and interest preferences. For each candidate item, the final recommendation probability is based on its features and current user preferences, and is processed by the softmax function to obtain the final recommendation probability.

2. The method for comparative collaborative filtering recommendation based on adaptive decoupling according to claim 1, characterized in that: In step 1, based on the user-item interaction matrix We follow the LightGCN approach, discarding nonlinear activation and feature transformation during propagation to obtain user and item feature embedding vectors for each layer: in and represents the neighbor set of the target user u and item i, which can offset the impact of node degree imbalance. In order to improve the feature embedding vector, we use multi-layer embedding propagation and add the node embeddings of all layers to obtain the user and item local embedding e u and e i . In order to promote a more even distribution of user and item representations, we use cross-layer contrastive learning to enhance the supervision signal, select the embeddings of the first and last layers for comparison, and construct a contrastive loss function based on the principle of maximizing mutual information: Where s(·) represents the cosine similarity function, which is used to measure the similarity between two embeddings; τ1 is the temperature in softmax, which is used to control the smoothness of the softmax curve.

3. The method for contrastive collaborative filtering recommendation based on adaptive decoupling according to claim 1, characterized in that: In step 2, the interaction between users and items is often affected by potential intention factors. Therefore, the K-means clustering algorithm is used to learn the intention prototype representation of users and items, and the center of the cluster is regarded as the intention prototype with global context: Among them C k represents the kth cluster obtained by clustering, represents the kth intention prototype from the user’s perspective. We use the local embedding e u With each intent prototype c u The correlation score between them is used to calculate the final intent representation: Finally, the local embedding and the global decoupled embedding are fused by combining the weight α to obtain the final representation of the node: s u =(1-a)e u +ah u 4. The method for contrastive collaborative filtering recommendation based on adaptive decoupling according to claim 1, characterized in that: In step 3, the clustering algorithm will cause similar users to have basically the same decoupled information, and unreasonable fusion of additional modeling information may aggravate the over-smoothing phenomenon. Therefore, the harmony score between the embedded information and the neighbor information is used to adaptively control the differences between nodes. The loss function is specifically: Among them, g(·) represents the similarity function between each node and all other nodes, ω u represents the adaptive weight of user u, which is obtained according to the harmony score between the node and the neighbor information: Among them, 0<ε<1 is a hyperparameter, s a is the harmony score between the original embedding of node a and its neighbor node information, specifically: Among them, e a ,e a′ denotes the initialized user and item embeddings, and sig(·) denotes the sigmoid activation function. We explore the neighborhood harmony score of the central node a by summing up the k-order neighbor information of the node. The higher the neighborhood harmony score, the stronger the structural consistency between the node and its neighboring nodes.

5. The method for contrastive collaborative filtering recommendation based on adaptive decoupling according to claim 1, characterized in that: In step 4, the interaction probability between the user and the item is calculated based on the final embedding of the user and the item, and the softmax function is applied to obtain the final recommendation probability of each candidate item: We use pairwise margin loss to optimize the recommendation task, specifically: in, represents all items that user u has interacted with before, and i′ represents negative samples that have not been interacted with.

Images