Domain adaptation using stock market prices to refine sentiment dictionaries

As part of a larger project where we are examining the relationship and influence of news and social media on stock price, here we investigate the potential links between the sentiment of news articles about companies and stock price change of those companies. We describe a method to adapt sentiment word lists based on news articles about specific companies, in our case downloaded from the Guardian. Our novel approach here is to adapt word lists in sentiment classifiers for news articles based on the relevant stock price change of a company at the time of web publication of the articles. This adaptable word list approach is compared against the financial lexicon from Loughran and McDonald (2011) as well as the more general MPQA word list (Wilson et al., 2005). Our experiments investigate the need for domain specific word lists and demonstrate how general word lists miss indicators of sentiment by not creating or adapting lists that come directly from news about the company. The companies in our experiments are BP, Royal Dutch Shell and Volkswagen