An Empirical Study on Ranking Change Recommendations Retrieved Using Code Similarity

摘要

Providing change suggestions for a particular code snippet on the basis of how similar code snippets were changed in the past has been investigated by a number of studies. These studies rank change recommendations emphasizing their frequency of occurrence during the prior evolution. In our study, we investigate the ranking of change recommendations on the basis of their recency of occurrence in the past and compare this technique with the frequency based technique. According to our experimental results on thousands of revisions of six subject systems we observe that while ranking on the basis of frequency performs better than recency based ranking, a combination of these two techniques performs significantly better than the discrete techniques. We find that the combined technique provides better overall rankings for 16.58% more cases when compared with the frequency ranking technique, and 57.48% more cases when compared with the recency ranking technique.