Event Detection and Stock Prediction: A Knowledge-intensive Approach

Abstract

This thesis explores knowledge-intensive event detection using articles from Dagens Næringsliv for the purpose of predicting stocks. The event detection task is solved by using LogicLDA to detect events from segments of the articles. Events and sentiment from articles and technical and fundamental analysis relating to the ten largest companies of the Oslo Stock Exchange energy sector are used as input for various machine-learning algorithms to predict stock prices. Event detection, stock prediction and a trading simulation all achieve encouraging results.