Aspect-level sentiment classification method

Abstract

The embodiment of the invention relates to an aspect-level sentiment classification method. The method comprises the following steps: extracting feature information of aspect words and context of a to-be-analyzed text by adopting a bidirectional long-short-term memory network model to respectively obtain hidden layer representation of the aspect words and context hidden layer representation of the context; for the obtained hidden layer representation and context hidden layer representation, extracting context emotion feature information in direct grammatical relation with aspect words by adopting a graph convolutional network according to a syntax dependency tree, and obtaining context information feature representation fused with grammar dependency information; based on context information feature representation, utilizing an attention mechanism to learn interaction information of aspect words and the context, extracting emotion feature information making important contribution to aspect word emotion classification in the context, and predicting emotion polarity. Compared with a conventional classical text sentiment analysis method and a current mainstream method, the method provided by the embodiment of the invention has more ideal text sentiment analysis accuracy.

Description

technical field [0001] The present disclosure relates to the technical field of artificial intelligence, in particular to an aspect-level sentiment classification method. Background technique [0002] With the rise of social media, sentiment analysis has increasingly become a research hotspot in the direction of natural language processing in the field of artificial intelligence. Its research objects are people's opinions, emotions and attitudes towards services, events, products, topics and their attributes. Analysis of speech can often quickly grasp people's attitudes or opinions on hot topics, policies, commodities, etc., which has significant research value. [0003] In related technologies, early sentiment analysis methods mainly rely on statistical methods to establish sentiment dictionaries, and use the sentiment dictionary as the basis of sentiment analysis. However, regardless of artificial construction or automatic construction of sentiment dictionaries, it is oft...

AI Technical Summary

Problems solved by technology

However, regardless of artificial construction or automatic construction of sentiment dictionaries, it is often only for a certain field, and the effect of cross-domain sentiment analysis is not ideal.

At present, the research on aspect-level sentiment classification is gradually carried out along the deep learning technology, but the existing methods often ignore the interaction information between aspect words and context, which leads to poor sentiment analysis results.

Images