Retrieving News Articles Relevant to Stock Market Fluctuations

This paper presents a novel approach for finding the corellation between news articles and stock prices from an information retrieval perspective. We accomplish this by ranking news articles in a very large collection based on their relevance to market price changes. The news collection is composed of six years of news and the corresponding daily stock prices. Each article in the collection is labelled as being relevant or not with respect to a significant change in stock price over a specific time window. We evaluated the performance of logistic regression, maximum entropy, rsj and compared it to a baseline. Our results show that for lower recall our methods perform significantly better than the baseline.