Supporting Code Search with Context-Aware, Analytics-Driven, Effective Query Reformulation

Abstract

Software developers often experience difficulties in preparing appropriate queries for code search. Recent finding has suggested that developers fail to choose the right search keywords from an issue report for 88% of times. Thus, despite a number of earlier studies, automatic reformulation of queries for the code search is an open problem which warrants further investigations. In this dissertation work, we hypothesize that code search could be improved by adopting appropriate term weighting, context-awareness and data-analytics in query reformulation. We ask three research questions to evaluate the hypothesis, and then conduct six studies to answer these questions. Our proposed approaches improve code search by incorporating (1) novel, appropriate keyword selection algorithms, (2) context-awareness, (3) crowdsourced knowledge from Stack Overflow, and (4) large-scale data analytics into the query reformulation process.