Dependency Parsing of Financial News to Improve Sentiment Analysis for Predicting Market Prices

Abstract

Online information is available for investors to track the financial status of target companies. Since financial markets could be affected by many factors, it's challenging to predict their prices. Given huge amount of financial news articles, it would be useful if their potential influences on market prices could be effectively inferred from news sentiments. In this paper, we propose an automatic approach to market price prediction by dependency parsing for improving sentiment classification. First, financial news related to target regions are collected by customized crawlers. Grammatical relations between sentiment words and negation modifiers are extracted from news articles by dependency parsing. Then, for all target- related news, sentiment scores are aggregated by exponential smoothing as an estimation of market prices. Finally, the correlation between news sentiments and market prices are analyzed. From the experimental results of major financial news articles on Taiwan stock index, with the help of dependency parsing, news sentiments can be effectively classified with an accuracy of 87.24%. When considering the time lag of two days after news release, it showed the highest positive correlation of 0.614 with market index in the corresponding region. This shows the potential of our proposed approach to market prices prediction with dependency parsing for improving sentiment analysis of financial news articles. Further investigation is needed to evaluate the performance of the proposed approach in different regions and time periods.