Relevance Feature Discovery is an innovative model that classifies terms into distinct categories and effectively updates term weights and distribution in patterns, hence boosting text mining performance.The terms that appear more frequently in relevant papers are regarded as positive specific terms. The terms that appear more frequently in irrelevant papers are classified as negative specific terms. The goal of Relevance Feature Discovery is to extract high-quality features that accurately represent the user's demands. This system outperforms term and pattern-based techniques.