Research On Sentiment Analysis Algorith Based On Topic Model And Deep Learning

Various factors have forced our lives to shift from offline to online in the recent years.The change of lifestyle promotes the rapid development of internet industry.There is a huge amount of textual data that carries people's views,positions,emotions and other subjective information,which can help us gain experience and make decisions.How to mine and analyze these data is the focus in the field of natural language processing.The thesis analyzes the shortcomings of the existing research,aiming at the lack of topic and sentence concerns in the existing affective analysis methods,the following four points are examined:1.The thesis aims at the lack of attention to the theme of the text in the existing affective analysis methods to extends and improves the traditional sentiment analysis model.This thesis puts forward a method of integrating topic information into deep neural network,and constructs the emotion classification model of integrating topic information(TB-LSTM).Experiments show that this method of integrating topic information into deep neural network is effective.2.In this thesis,a double-layer attention neural network model with topic information(TBAM)is proposed to deal with the lack of attention to sentences in text.The model follows the hierarchical structure of the document,and uses attention mechanism at the lexical level and the sentence level respectively,so that it can pay attention to the more important and the less important contents respectively when constructing the document representation.At the lexical level,the method of embedding text topic information into neural network proposed in the third chapter makes up for the neglect of topic information in classification model,and increases the generalization ability of the model.According to the experimental results,the efficiency of the model is significantly improved when considering the information and document hierarchy.3.TBAM model is applied to multi-dimensional sentiment classification.In this thesis,TBAM model is applied to an internet public data set containing twelve human emotions.Then,according to the literature,this data set is transformed and integrated into a seven-dimensional sentiment classification data set.Finally,TBAM model is used to classify emotions again.According to the experimental results,compared with other models,TBAM model has a good effect in multi-dimensional sentiment classification.4.In order to verify that the proposed model algorithm can be applied to real life,this thesis designs and implements a multidimensional emotion classification prototype system.