Event forecasting through news sentiment analysis

Forecasting is the process of making predictions about events whose actual outcomes have not yet been observed. Accurate forecasting helps people make the right decision and thus saves time and effort. Recently, more and more effort has been put into designing sophisticated models to improve forecasting accuracy. There are time series models, qualitative models, and causal models. The techniques have advanced from simple regression to multiple regression, from moving average to trend analysis. A number of assertions have been made that more sophisticated models yield better forecasting results. In this thesis, we tried to improve our forecasting accuracy by incorporating sentiment analysis data. We showed that forecasting for sports betting and political events can be significantly improved by using sentiment analysis as a pre-processor. This is because the sentiment data is generated from widely-covered news sources and can be viewed as an aggregation of a very large group of people. As argued by The Wisdom of Crowds [26], decisions drawn from aggregation of information in groups are often better than those that could have been made by any single member of the group. We also showed that with sentiment analysis data, even simple intuitive models, like linear regression models and regression models with exponentially weighted moving average techniques, can generate better results. We also further derived profitable strategies for sports betting and accurate forecasting for political events. As a result, our experimental results also validated our sentiment analysis system em Lydia. We hope further investigation of the usage of sentiment data in forecasting can be performed in different areas, like national economical data forecasting, most popular product forecasting, and even Oscar awardee forecasting