Financial News Mining:: Extracting useful Information from Continuous Streams of Text

Online financial news sources continuously publish information about actors involved in the Norwegian financial market. These are often short messages describing temporal relations. However, the amount of information is overwhelming and it requires a great effort to stay up to date on both the latest news and historical relations. Therefore it would have been advantageous to automatically analyse the information. In this report we present a framework for identifying actors and relations between them. Text mining techniques are employed to extract the relations and how they evolve over time. Techniques such as part of speech tagging, named entity identification, along with traditional information retrieval and information extraction methods are employed. Features extracted from the news articles are represented as vectors in a vector space. The framework employs the feature vectors to identify and describe relations between entities in the financial market. A qualitative evaluation of the framework shows that the approach has promising results. Our main finding is that vector representations of features have potential for detecting relations between actors, and how these relations evolve. We also found that the approach taken is dependent on an accurate identification of named entities