User-interested-keywords set discovery using rough sets for intelligent information agents

With the exponential growth of the World Wide Web, it becomes difficult and time-consuming for users to find what they are interested in. This leads to a clear demand for automated methods, commonly referred to as intelligent information agents. User interest mining, which is a key technology for implementation of intelligent information agents, has attracted much more attention recently. In content-based mining, due to the sheer amount of words in the Web pages, selecting keywords that describe users' interests is critical for the performance of intelligent information agents. Because rough sets theory shows its powerful ability in attribute selection from large data set, it has been attracting much attention since it was proposed by Z. Pawlak (1982). In this paper, we present a user-interested-keywords set discovery algorithm employing rough sets. We evaluate the classification precision of the discovered keyword sets by using the KNN classifier. The remarkable character of our algorithm is that users are not required to provide negative examples of no interest. Since users are not likely to save or spend time on anything they're not interested in, our algorithm is more reasonable and practical at this aspect. We test our algorithm on some Web pages that a user has browsed. Experiments demonstrate that our algorithm can create satisfactory user-interested-keywords set.