Title: A Text Mining System for Evaluating the Stock Market’s Response To News
Supervisor: Luis Torgo; Co-supervisor: José João, University of Minho, Portugal
MAP-i Doctoral Programme in Computer Science
Abstract
This thesis presents a text mining system which was designed to predict the direction of a share or financial market. The text mining system is a complete pipeline which: 1. scrapes new stories from the Internet, 2. extracts news text from the scraped news stories, 3. identifies relevant news stories for a specific company or financial market, 4. classifies sentences, news stories and quotes and 5. makes a trading inference using these classications.
The thesis documents advances in 1. ontology construction and maintenance, 2. fine grained event and sentiment extraction and classification at the sentence level, 3. news story classification, 4. direct speech classification and 5. information retrieval. These advances were also contributions to the fields of semi-supervised learning and ontology engineering.
The advances in the news classification at the document, sentence and direct speech level demonstrate measurable advantages in trading experiments on the FTSE 250financial markets over competing text classification strategies. The complete system, however, did not demonstrate a measurable trading advantage in experiments conducted on the shares of Apple, Google, IBM and Microsoft. The system, however, provides a blueprint for future systems.
Current Position:
Brett is currently the Head of Research at SciCrop