Feature Detection and Classification in Financial News

The thesis concerns the detection of composite features in news articles as well as their significance for the classification's performance. The considered documents are related to the monetary policy conducted by the Federal Reserve. One principal goal of this work is to quantify embedded information in financial texts by using phrase structure grammar trees in combination with statistical measures and domain knowledge. Thereby, each document is represented as a combination of linguistic features and feature-values. Furthermore, the work examines the correlations between the determined features and an equity market index by modelling the index volatilities as functions of key announcements. A design is targeted, which should enable the temporal tracking of information alterations. Essential aspects of the thesis are the identification, the extraction, and the representation of domain-specific features and their conditional feature-values.