||The Internet has brought about an information revolution. There are huge amounts of useful information that a user can retrieve from the World Wide Web. Some of the most popular web pages are electronic newspapers. These web pages are written by financial reporters describing the overall situation on major stock markets around the world. Such newspaper articles provide clues to analyze stock markets. In this thesis, we aim to discover implicit knowledge contained in these textual newspaper articles and predict closing values of the Hang Seng Index (HSI), the most widely used stock market index in Hong Kong. The techniques to do the forecasting are roughly as follows. Keyword pairs, triples, and quadruples are chosen from the articles by domain experts. The number of occurrences of these keywords on the articles are used to calculate weights. Four different weighting mechanisms are investigated. Probabilistic datalog rules representing the implicit knowledge are constructed based on the weighting of these keywords. The rules are used to produce three mutually exclusive binary decisions: HSI up, down, and steady. If more than one, or none of the binary decision is true, a conflict resolution algorithm is applied. Finally, an expected numerical closing value of the HSI is calculated. To test the effectiveness of our model, extensive experiments are conducted. We first locate the best performing weighting mechanism. Then, cross validation is used to confirm these findings. The experimental results reveal that our system can potentially serve as a decision support tool helping Hong Kong's stock market analysts.