Subjectively Related Association Term Discovery towards Personalized Web Information Retrieval

In this paper, we propose a new semi-supervised clustering methodology to extract topically coherent contents from given Web pages, according to a user´s topic interests. It is an effort to resolve low information retrieval performance, caused by one fact that even a single Web page often contains multi-topic related contents. Our evaluation results showed some advantages of our semi-supervised clustering methodology: it reduces the needs of term classification knowledge between the given Web pages and a user´s topic interests. It also gets better clustering performances than those which can be achieved with the well-known supervised feature-term selection method chi² statistics.