The data that has few terms decreases retrieve performance.
It will be a solution method to extend terms before searching
[4]. The key point is to extend one term to more interrelated
terms. That makes a bigger search range and a better result.
The pseudo relevance feedback is a general method to extend
terms [5-7]. This method takes keywords from top relevant
documents and extends them. However, every term in the
document owns different meaning. The paper calculates the weight of terms extracted from documents to substitute query
extension by the terms of top relevant documents.
When the system extracts the term, some terms are
unsuitable to represent documents. It must analyze inner terms
of documents with language structure and term statistics to find
out keywords. This research is called Automatic Term
Recognition (ATR) traditionally [8, 9].