Text Classification for Intelligent Agent Portfolio Management

In the application domain of stock portfolio management, software agents that evaluate the risks associated with the individual companies of a portfolio should be able to read electronic news articles that are written to give investors an indication of the nancial outlook of a company. There is a positive correlation between news reports on a company's nancial outlook and the company's attractiveness as an investment. However, because of the volume of such reports, it is impossible for nancial analysts or investors to track and read each one. Therefore, it would be very helpful to have a system that automatically categorizes news reports that re ect positively or negatively on a company's nancial outlook. To accomplish this task, we treat the unsupervised reading and understanding of news articles as an automatic text classi cation problem. In this paper, we propose a text classi cation method that we call, \domain experts" and \self-con dent" sampling technique, and compare it with naive Bayes with expectation maximization (EM). We evaluate these learning techniques in terms of how well they improve with unlabeled data after being initially trained on a small number of human-labeled articles and how well they classify the latest nancial news articles. The signi cance of this work lies in the new classi cation method that we propose and in the sampling technique we used for improving classi cation accuracy.