Web Mining For Financial Market Prediction Based On Online Sentiments

Financial market prediction is a critically important research topic in financial data mining because of its potential commerce application and attractive profits. Previous studies in financial market prediction mainly focus on financial and economic indicators. Web information, as an information repository, has been used in customer relationship management and recommendation, but it is rarely considered to be useful in financial market prediction. In this paper, a combined web mining and sentiment analysis method is proposed to forecast financial markets using web information. In the proposed method, a spider is firstly employed to crawl tweets from Twitter. Secondly, Opinion Finder is offered to mining the online sentiments hidden in tweets. Thirdly, some new sentiment indicators are suggested and a stochastic time effective function (STEF) is introduced to integrate everyday sentiments. Fourthly, support vector regressions (SVRs) are used to model the relationship between online sentiments and financial market prices. Finally, the selective model can be serviced for financial market prediction. To validate the proposed method, Standard and Poor’s 500 Index (S&P 500) is used for evaluation. The empirical results show that our proposed forecasting method outperforms the traditional forecasting methods, and meanwhile, the proposed method can also capture individual behavior in financial market quickly and easily. These findings imply that the proposed method is a promising approach for financial market prediction.