Context-Sensitive Ranking of Dependencies for Software Navigation

Software navigation during change tasks requires developers to perform numerous search cycles to discover new elements related to their task. In a typical search cycle a developer selects an element of interest, triggers a search for its dependencies, inspects the list of results, and selects a result that appears to be relevant. We are interested in improving the efficiency of developers engaged in program navigation by automatically searching for all dependencies to or from elements already discovered, and continually providing the list of dependencies, ranked in decreasing order of likelihood to be related to the task. In this paper, we present a framework for the systematic evaluation of ranking algorithms based on multi-element contexts, and report on a preliminary experiment to assess the value of three ranking strategies for dependencies: using the topology of the dependency graph, textual similarity, or combination of both. Our results show that all three strategies significantly improve the probability of recommending relevant elements in many situations, but that no ranking strategy appears to be universally optimal.