Research On Short Text Sentiment Analysis Based On Deep Learning

The influence of online short text is mainly spread through word-of-mouth.It has a direct impact on customers' purchase intention and loyalty.These online short texts have been regarded as an important medium affecting their online reputation by various industries.Therefore,it is of great significance to analyze the sentimental changes of these short texts for the development of various industries.However,because of its concise language expression and less carrying features,the short text can use very few features when expressing short text,so that the effect of short text sentimental analysis is not ideal.Moreover,in the text representation,the traditional word embedding model often ignores the sentimental information of the words themselves,and only directly determines the word embedding representation through the context.At the same time,in practical application,few people analyze the multi-dimensional emotional real-time changes of short text.By analyzing the sentimental real-time changes of short text dimension,we can help enterprises better understand customer needs in different time periods.To solve these problems,this dissertation uses deep learning technology to mine valuable information contained in short text.The main research work of this paper is as follows:(1)This dissertation aims at the problems of short text,such as concise language expression,less carrying features and lack of sentimental information.This dissertation proposes a short text sentiment analysis method based on multi feature fusion attention mechanism.The features considered include semantic features,emotional features and location features.The feature representation of short text is constructed by considering various information,so as to improve the classification accuracy.At the same time,we also design attention mechanism to capture the relationship between pairs of words in the text.(2)Customers who are satisfied with the service can provide good reputation for the enterprise,and its effect will be better than the traditional advertising.Therefore,if an enterprise wants to maintain a good competitiveness in its peers,it must master the emotional changes of various dimensions of customer service.Therefore,we propose a short text dimension affective analysis method based on coherent topic model.This method digs out the dimensions of short text through the theme model and word embedding,and judges the sentimental polarity of the theme word by combining the sentimental dictionary and dependency syntax analysis.Finally,it analyzes the sentimental changes of each dimension of short text over time in different time periods.All of the above models are validated by using the customer text reviews of Sanya's deluxe,upscale and comfortable resort hotels as experimental data.The results show the feasibility of this method.