The Application Of Sentiment Analysis In Financial Corpus

There are a lot of financial information in the Internet.Many investors like to share their viewpoints about the stock market on the web.Researches on the public opinion in the financial market is an effective complementary to behavioral finance.Using the valuable information mined from the stock data can help financial institutions or individual investors on making decisions on investment.This paper studies the sentiment analysis task of short texts in the financial field.First,we collected a corpus of 7514 financial news from some popular financial websites.Each news text was manually labelled with sentiment polarity by financial professionals.Second,based on the above financial corpus,we used a few algorithms,including TFIDF(term frequency–inverse document frequency),IG(Information Gain),and chi,to extract feature words.A preliminary emotion dictionary was built based on the extracted feature words.This dictionary can be further improved by allowing a manual modification of weight of sentiment words if they are very significant.Third,the weights of word vectors are studied based on the derogatory meaning and the relative importance of emotional words.Last,sentinel analysis Models were trained using several classification methods including SVM(Support Vector Machine),RF(Random Forest),RNN(Recurrent Neural Networks),CNN(Convolutional Neural Network),and gcForest(MultiGrained Cascade forest).The performance of these learned models were compared and analyzed.To sum up,aiming at financial short text,the paper studied the sentiment analysis task based different feature representation methods and different classification algorithms.