A search log mining based query expansion technique to improve effectiveness in code search

The effectiveness of a code search engine is reduced when query terms do not represent the information needs properly or terms are ambiguous. As a result, many irrelevant code snippets and software artifacts are retrieved that hinder the developers reusing existing source code. In this paper, a technique named QExpandator is proposed that improves the effectiveness in code search by expanding query terms with search topic and content specific keywords. It extracts user queries and clicked code fragments from the previous search history and represents each query in a vector document. Jaccard similarity score is calculated for each term in the document vector and a posting list of conceptually similar words is created based on the similarity score. Finally, to expand a user query, top scored terms are retrieved for each query term and appended to the original query. To evaluate the technique, 22 user queries were selected and an existing approach was employed. QExpandator shows 48.6% more effectiveness in terms of precision at 10 (P@10) than the existing one. Moreover, for each query, it increases P@10 from 60.9% to 90% on an average due to using search topic and context specific keywords.