Graph neural collaborative filtering method based on attention mechanism

Abstract

The invention discloses a graph neural collaborative filtering method based on an attention mechanism. The method comprises the following steps: firstly, carrying out matrix decomposition of historical interaction data of a user and an article to obtain two matrixes which represent initial feature vector matrixes of the user and the article respectively; secondly, obtaining a correlation coefficient matrix between adjacent nodes in a graph according to initial feature vectors, and adding a social graph to construct a Laplacian matrix; then performing graph convolution on the initial feature vectors of the user and the article and the constructed Laplacian matrix to obtain multi-layer neighbor node information; and finally, performing nonlinear fusion on obtained multi-layer potential feature vectors to improve the recommendation effect of a model by improving vectors. According to the model, the problem of no differentiation in the initial feature vector randomization and graph convolution process is solved, the learning rate of experimental model is improved, and the model has interpretability. Experiments were carried out on two public data sets, and the result showed that the model achieves a better effect.

Description

technical field [0001] The invention belongs to the field of recommendation systems, in particular to a graph neural collaborative filtering model method based on an attention mechanism. Background technique [0002] With the exponential growth of network data, personalized recommendation can help people quickly obtain the information they are interested in. The core idea is to effectively model the user's historical behavior data and recommend items that the user may be interested in. Collaborative filtering is the most widely used algorithm in the recommendation system. It assumes that similar users have similar interest preferences for the same items. The most common way is to parameterize the user's historical behavior information, and based on these parameters. predict. As machine learning and deep learning have made great progress in the fields of image recognition and natural language processing, more and more scholars combine deep learning and collaborative filterin...

AI Technical Summary

Problems solved by technology

However, in the previous research on inference algorithms based on graph convolutional neural networks, they did not consider the different preference relationships between different neighbor nodes on the interaction graph.

Images