Mining Relevant Text Features for Retrieving Web Information

It is a big challenge to develop effective methods that can discover high quality and useful features in text documents. Most existing information retrieval and text mining methods focuses on term-based approach that often suffers from the problems of term variation and noise. This paper illustrates an innovative approach that discovers relevant knowledge to precisely describe text features for retrieving web information. In particular, it extracts precise text patterns by considering both relevant and irrelevant documents. Then, the discovered patterns are used to find accurate relevant features in a training set. The proposed approach has been evaluated through the implementation of a novel information filtering model and a comparative evaluation is conducted by invoking state-of-the-art models. The experimental results obtained based on the Reuters Corpus Volume 1 and TREC topics show that the proposed approach significantly outperforms the best baseline method.