Research On Aspect-level Sentiment Classification Methods Based On Deep Learning

Sentiment classification is one of the basic tasks in the field of natural language processing.Sentiment classification is divided into coarse-grained and fine-grained.With the development of the times,coarse-grained sentiment classification has gradually been unable to meet the needs of human beings,so fine-grained sentiment classification has gradually become the focus of research.Fine grained sentiment classification,also known as aspect level sentiment classification,aims to judge the sentiment polarity of a specific aspect in a sentence.The core research content of this paper is to use the method based on deep learning for aspect level sentiment classification.The innovation and work of this paper are as follows.(1)Most aspect level sentiment classification methods based on traditional recurrent neural network and attention mechanism can not extract the semantic features of text completely;secondly,they ignore the interaction information between context and aspect words.To solve these problem,this paper proposes a model named Bi LSTM-IAN,which is based on Bidirectional Long Short-Term Memory and interactive attention network.Firstly,the model uses Bi LSTM to extract more complete semantic features of context and aspect words respectively.Then,two attention mechanisms are used to construct an interactive attention network to learn the interaction information between context and aspect words.The experimental results prove that Bi LSTM-IAN model can effectively improve the classification effect.(2)The attention mechanism in Bi LSTM-IAN may assign too low weight to the important sentimental words in the context.Secondly,the interactive information between context and aspect words is not sufficiently learned.To solve these problem,this paper proposes a model named LAR-WIAM,which is based on word-level interactive attention mechanism.Firstly,the model improves the traditional input method,and divides the sentence into the above containing the aspect word,the aspect word and the following containing the aspect word as input,so that the sentimental words in the above or below can be assigned enough weight.Then,the word level interactive attention mechanism is proposed to learn more fully the interaction information between the above and aspect words,and between the below and aspect words.The experimental results prove that LAR-WIAM model can further improve the classification effect.(3)Although LAR-WIAM model achieves fairly good results,like most models,it does not consider the syntactic dependency information between context words and aspect words.To solve these problem,this paper proposes a model named Bi GCN-WIAM,which is based on bidirectional graph convolution network and word-level interactive attention mechanism.This model improves the traditional undirected graph convolution network,and proposes a bidirectional graph convolution network which can sense the direction information in the syntactic dependency tree,which is used to extract the syntactic dependency information between context words and aspect words.After the Bi GCN layer,the mask layer is used to obtain the representation of specific aspect words.Finally,the word level interactive attention mechanism is used to learn the interaction information between context and aspect words sufficiently.The experimental results prove that Bi GCN-WIAM model can more further improve the classification effect.