Mining Topics and Sentiments from Yahoo! Finance Message Boards

The finance portion of Yahoo! web portal hosts message boards for many stock ticker symbols. The message traffic on each of the tickers’ message board ranges from less than 10 messages per day to thousands of messages per day. While some discussions on the message boards are relevant to a company and its stock performance, there are many messages on the message boards which are off-topic or contain very little information (similar to spam for e-mail). This thesis aims to explore a new method of filtering the messages on these message boards, by combining concepts from a previously developed SVM-based message filtering system and the Topic-Sentiment Mixture (TSM) model. By applying the TSM model to output from the message filtering system, this new system aims to automatically extract the topics and sentiments of meaningful messages, and allow the user to more easily find topics and document sentiments that he or she is interested in. The SVM system and TSM are detailed below.