Cross-domain sentiment analysis method based on GCN under lifelong learning framework

Abstract

The invention provides a cross-domain sentiment analysis method based on GCN under a lifelong learning framework, relates to the field of cross-domain sentiment classification in natural language processing, and is realized by the following process: proposing a cross-domain sentiment classification algorithm based on a graph convolutional neural network, namely CDSGCN, on the basis of proposing the CDSGCN, proposing a cross-domain sentiment classification algorithm based on a graph convolutional neural network under a lifelong learning framework, namely LLCDSGCN in combination with a lifelonglearning thought, wherein the lifelong learning is different from transfer learning or multi-task learning and other related learning tasks due to the characteristics, the limitation of isolated learning is broken through, the influence caused by time-consuming and labor-consuming manual data annotation is relieved, and the characteristics are in accordance with the initial intention of cross-domain emotion classification tasks.

Description

technical field [0001] The invention relates to the field of cross-domain sentiment classification in natural language processing, in particular to a GCN-based cross-domain sentiment analysis method under a lifelong learning framework. Background technique [0002] The popularization of the Internet has brought a larger user group to merchants, and also brought more user feedback information. Users can easily express their views and opinions on the products they use or receive services on the Internet, such as a large number of product reviews. Can be easily obtained on Douban (movie reviews), Amazon (book reviews), Meituan (restaurant reviews), etc. Document-level sentiment classification refers to classifying a given review according to the sentiment expressed by the review author. The task is very important. First of all, for merchants, consumers’ evaluations of products reflect their likes and dislikes for product quality or services. Secondly, consumers who have not pur...

AI Technical Summary

Problems solved by technology

[0003] Problems: Traditional supervised learning algorithms have been proven to be effective in dealing with sentiment classification problems, and are widely used in the task of predicting the sentiment polarity of reviews given a domain

But the graph structure is different from the matrix structure, and the situation near each node may be different from each other, so it does not have translation invariance

Images