A Topic-level Sentiment Classification Model Based on Deep Learning

Abstract

The key point of topic-level sentiment classification is how to construct a contextual sentence representation related to the topic word according to the given topic word and context sentence. The method based on attention and recurrent neural network can be calculated end-to-end according to the topic word representation. This type of method has achieved excellent performance on the relevance of the given subject words and the parts of the context sentence. This paper improves the mainstream neural network work based on the attention mechanism, and combines the AEN (Attention Encoder Networks) model and the AOA (Attention Over Attention) model to propose a new AEN-AOA model, which is based on the BDCI2018-Automotive Industry User View Theme and Good results have been achieved on emotion recognition tasks. This model can effectively mine the emotional tendency based on the topic level, and has a good application prospect and use value.