It could be a huge challenge to ensure the standard of discovered connection options in text documents for describing user preferences due to giant scale terms and information patterns. Most existing in style text mining and classi?cation strategies has adopted term-based approaches. However, they need all suffered from the issues of ambiguity and synonymousness. Over the years, there has been usually command the hypothesis that pattern-based strategies ought to perform higher than term-based ones in describing user preferences; nonetheless, a way to effectively use giant scale patterns remains a tough downside in text mining. to create a breakthrough during this difficult issue, this paper presents associate innovative model for connection feature discovery. It discovers each positive and negative patterns in text documents as higher level options and deploys them over lowlevel options (terms). It conjointly classi?es terms into classes and updates term weights supported their speci?city and their distributions in patterns. Substantial experiments victimization this model on RCV1, TREC topics and Reuters-21578 show that the projected model signi?cantly outperforms each the progressive term-based strategies and therefore the pattern primarily based strategies.