WPS 11-10: Evaluating Sentiment in Financial News Articles
Evaluating Sentiment in Financial News Articles
Robert P. Schumaker1,
and Hsinchun Chen3
1 Computer and Information Systems Dept
Cleveland State University, Cleveland, Ohio 44115, USA
2 Computer Information Systems, The W. A. Franke College of Business,
Northern Arizona University, Flagstaff, Arizona 86011, USA
3 Artificial Intelligence Lab, Department of Management Information Systems
The University of Arizona, Tucson, Arizona 85721, USA
We investigate the pairing of a financial news article prediction system, AZFinText, with sentiment analysis techniques. From our comparisons we found that news articles of a subjective nature were easier to predict in both price direction (59.0% vs 50.4% without sentiment) and through a simple trading engine (3.30% return vs 2.41% without sentiment). Looking into sentiment further, we found that news articles of a negative sentiment were easiest to predict in both price direction (50.9% vs 50.4% without sentiment) and our simple trading engine (3.04% return vs 2.41% without sentiment). Investigating the negative sentiment further, we found that AZFinText was best able to predict price decreases in articles of a positive sentiment (53.5%) and price increases in articles of a negative or neutral sentiment (52.4% and 49.5% respectively).
Key words: Knowledge management, prediction from textual documents, sentiment analysis
11-10 October 2011
PDF of working paper HERE.
Categories: working paper series 2010-2011 2010-2011 working paper series robert p. schumaker yulei zhang chun-neng huang hsinchun chen evaluating sentiment in financial news articles knowledge management prediction from textual documents sentiment analysis