E-commerce recommendation method based on self-attention mechanism and graph neural network

Abstract

The invention provides an e-commerce recommendation method based on a self-attention mechanism and a graph neural network. The method comprises the following steps: firstly, preprocessing e-commerce transaction data, extracting sessions meeting requirements, and generating sequences and tags to form a data set for an experiment; forming a session graph according to a data set obtained by preprocessing, carrying out weight normalization processing, and inputting the session graph into a graph neural network to obtain node vector representation in the graph; and finally, extracting a local interest vector representation and a global vector representation from the vector representations of the nodes in the graph, then respectively using a self-attention mechanism on the local interest vector representation and the global vector representation to obtain a corresponding local self-attention vector and a corresponding global self-attention vector, and aggregating the vectors to obtain vector representation of the mixed interests used for recommending favorite high-score articles to the user. According to the e-commerce recommendation method, related information clicked by the user in the past is fully considered, and the recommendation method with a better effect is provided.

Description

technical field [0001] The invention belongs to the technical field of personalized recommendation in data mining applications, and in particular relates to a self-attention mechanism and a graph neural network e-commerce recommendation method. Background technique [0002] At present, in the era of massive data, accompanied by the problem of information cocoons for users, how to provide effective personalized recommendation results based on the user's historical data is an important problem to be solved. Using scientific and effective methods to mine data to extract user interests and generate a suitable personalized recommendation system is the main means to solve this problem. [0003] The difference from the above is that the historical user behavior data in actual work is often too long, but in the face of massive data, low latency, and limited computing resources, the recommendation algorithm has to be established in a short session, but At the same time, the extracti...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to overcome the shortcomings of not making full use of the user portrait and the user's past information, and the inaccurate extraction of the conversion relationship between the item and the item in the conversation graph. Accurate capture, better access to the conversion relationship between items, and effectively extract the user's interest preferences

Images