Study And Application On Fine-grained Sentiment Analysis Of Financial Microblog

Text sentiment analysis has become a research hotspot in Natural Language Processing,however,researches focus on financial domain much less than which focus on movie rating,product reviews and public opinions.In this dissertation,we introduce a fine-grained sentiment analysis algorithm for financial microblog and apply the result of sentiment analysis to a stock prediction model.We improve the prediction accuracy of the stock prediction model and provide suggestions for investors to make their decisions by investigating the correlation between sentiment of investors and stock price.The study is constructed by several parts.First,in order to insure the accurancy of the object we study on,we design and implement a web crawler to collected more than 910,000 pieces of stock-market-related microblog data from February 27,2017 to July 31,2017.Then,we classify the sentiment into 5 categories: LIKE,DISGUST,ANGER,FEAR and OBJECTIVE and introduce an automatic domain specific sentiment word dictionary construction algorithm base on average cosine similarity,which improves accuracy of domain specific sentiment dictionary by reduce words which have high vector cosine but with inverse semantics;After that,we introduce a sentiment analysis algorithm combining rules with machine learning,which combines sentiment features of prior knowledge with statistical regularity of corpus that make the macro F-measure value of fine-grained sentiment analysis achieve 70.2%,which is 15% better than using the algorithm base on sentiment dictionary and regular analysis.Finally,we apply the above research results to build a stock forecast model on Shanghai Composite Index.The result shows that SVR model which considering the closing values and ANGER get the best prediction.We find an accuracy of 80.95% in predicting the daily up and down changes in the closing values of the Shanghai Composite Index and the Mean Average Percentage Error is only 0.51%,which is 33.34% better than the model only considering the closing values in predicting the daily up and down changes.The results of dissertation show that the sentiment analysis in the financial domain has practical significance.The stock prediction model base on fine-grained sentiment analysis of financial microblog has certain reference value.