Research Of Fine-Grained Sentiment Analysis Based On Recurrent Neural Network

The rapid development of e-commerce has brought about a wealth of consumer comments,which are of high value of both economic and social aspects.While a single comment may express a user’s feelings about a product,it may also present different comments on different aspects of the entity.As a result,the overall sentiment analysis of a single comment is sometimes inaccurate.Fine-grained sentiment analysis can accurately and comprehensively extract users ’ comments on different aspects of a product,not only helping potential consumers to make purchase analysis,but also providing suppliers with advantages and disadvantages of the product so as to further improve the product or service.Therefore,fine-grained sentiment analysis has gradually become a hot topic in the field of sentiment analysis.Based on the above research background,on the basis of in-depth study of domestic and foreign development status of emotion analysis and existing models,this thesis use deep learning technology to carry out specific research on aspect level emotion analysis in finegrained emotion analysis.The main research contents are as follows:(1)Aiming at the problem that the traditional word vector technology is insufficient to express the lexical semantics in a specific domain and the existing models are overloaded with information,this thesis proposes an aspect term extraction method based on the doubleembedded Bi GRU model based on attention.In this model,the double embedding mechanism is used to convert the comment text into word vector,and then the Bi GRU network is used for feature extraction to further capture the long-term dependency relationship between words.Meanwhile,the multi-head attention mechanism is introduced to focus on the information more critical to the current task,so that the model can capture more accurate feature information.Finally,Softmax is used to extract aspect terms.This model shows good performance in the task of aspect term extraction.(2)Aiming at the problems of application limitation caused by single task research,limited semantic information contained in the existing model and insufficient learning of text features,this thesis proposes a joint extraction model based on two-channel fusion,extracts aspect terms and corresponding emotional polarity,and verifies its effectiveness through experiments.In the feature extraction layer,the model adopts the form of two channels.Bi GRU network and CNN are respectively used to extract the global semantic information and local semantic information of the text,so as to obtain more comprehensive text features.Meanwhile,multi-head attention mechanism is introduced to distinguish the importance of different words,and then the feature information on the two channels is splicing and fusion.Finally complete the task of joint extraction.