Sameness: An Experiment in Code Search

—To date, most dedicated code search engines use ranking algorithms that focus only on the relevancy between the query and the results. In practice, this means that a developer may receive search results that are all drawn from the same project, all implement the same algorithm using the same external library, or all exhibit the same complexity or size, among other possibilities that are less than ideal. In this paper, we propose that code search engines should also locate both diverse and concise (brief but complete) sets of code results. We present four novel algorithms that use relevance, diversity, and conciseness in ranking code search results. To evaluate these algorithms and the value of diversity and conciseness in code search, twenty-one professional programmers were asked to compare pairs of top ten results produced by competing algorithms. We found that two of our new algorithms produce top ten results that are strongly preferred by the programmers.

• Type: Journal
• Year: 2015
• Where: MSR
• Authors: Lee Martie, André van der Hoek
• Keywords: MSR 2015, Software Engineering