An exploratory study on assessing feature location techniques

Abstract

This paper presents an exploratory study of ten feature location techniques that use various combinations of textual, dynamic, and static analyses. Unlike previous studies, the approaches are evaluated in terms of finding multiple relevant methods, not just a single starting point of a feature's implementation. Additionally, a new way of applying textual analysis is introduced by which queries are automatically composed of the identifiers of a method known to be relevant to a feature. Our results show that this new type of query is just as effective as a query formulated by a human. We also provide insights into situations when certain feature location approaches work well and fall short. Our results and observations can be used to guide future research on feature location.