A Mix-model based Deep Learning for Text Sentiment Analysis

Abstract

In recent years, with the rapid development of the Internet, a large number of emotional color data on commodities, events and services contain great value of mining. Traditional text sentiment analysis has been unable to effectively deal with the increasingly rich semantic information and it relies much on feature engineering. Manual features extraction consumes a lot of effort and it’s hard to cover all the information. Deep learning has strong ability to fit nonlinear functions, it can dig deeper features of language, describe the rich inner information, which has good applicability and powerful expression. Aim at the problems that single neural network model can’t fully extract features, the paper construct a bidirectional dynamic mixed memory model of text sentiment analysis based on deep learning. Firstly, using word embedding to represent the text.Secondly, building the combination of convolutional neural network and recurrent neural network, through the model fusion and the cascade configuration of memory cells, extracting features from different dimensions. Finally, adding the attention mechanism to weighted mean for output, capture the key information of text. The results show that the model can better make full use of the context of text, extract deeper features, and improve the accuracy of text analysis.