Rearch And Implementation Of Sentiment Analytic System For Stock Reviews

With the rapid development of the stock market,an increasing number of investors prefer to comment on the financial websites,through which investors can express their views on the individual share and large-capitalization stocks,therefore,there is an increasingly abundant stock comments on the Internet,as well as an growing demand of reviews that can be referred by investors.Users' emotional tendency,which is expressed by their own stock information on the internet,influences the tendency of the individual share and large-capitalization stocks to some extent,in order to assist the stock investors to make individual stock decisions,we need to further carry out a sentiment analysis of the stock reviews.However,it is very hard to analyze the emotional tendency of the stock reviews by ourselves,to better solve this problem,we need to design a set of automatic emotional elements analysis technologies and tools based on the stock reviews.Consequently,this paper mainly carries out the following several aspects of researchs.(1)The method which a sentiment tendency analysis based on sentiment lexicons for stock reviews is studied.On one hand,by referring to the idea of corpus study from machine learning,we introduce the emotional words of the stock field,on the other hand,we take the degree the influence of degree adverb and negative on the sentiment classification into consideration,and build the negative words and degree adverbs auxiliary lexicons,according to whose matching result to accomplish relative sentiment analysis,The experiment results show that,compared with traditional methods which are based on the sentiment lexicons,this method is preferable both in accuracy rate and recall rate.(2)The method which a sentiment tendency analysis based on Long Short-Term Memory(LSTM)for reviews is studied.A deep learning model for sentiment classification of the stock reviews texts is built based on LSTM network,word embedding,generated by word2vec,is used to input,then the sentence or chapter vectors,which are next used to input,and expressed by LSTM network,at last,the sentiment classification of the comments are finished through the Softmax activation function,which not only increases the accuracy rate of the method proposed in(1),but also make better use of the semantic information and words order information within sentences and chapters to sentiment classification.(3)Research and design a sentiment analysis system for stock reviews.In this paper,the method proposed in(2)was applied to the practical data,and an sentiment analysis system for stock reviews are founded to help the stock investors to make individual stock decisions,meanwhile,the feasibility of the methods mentioned above have been proved accordingly in the following texts.