Journal Article 改版履歴の分析に基づく変更支援手法における時間的近接性の考慮と同一作業コミットの統合による影響
An Empirical Study of the Effects of Recency-aware History Analysis and Commits Aggregation on Change Guide

森, 達也  ,  アンダース, ハグワード  ,  小林, 隆志

58 ( 4 )  , pp.807 - 817 , 2017-04-15
Many studies on change guide, which suggest necessary code changes with co-change rules extracted from a change history, have been performed so far. Recommendations by existing approaches are adequately accurate, however, those approaches often fail to detect candidates of overlooked changes. In this study, we focus on two characteristics to recommend more overlooked changes. One is recency. Some of software dependencies used for change guide become obsolete along with long-term evolution. We use only recent commits for extracting co-change rules to avoid incorrect suggestions stemming from such obsolete dependencies. The other is commits aggregation. The granularity of commits depends on the nature of developers and projects. We aggregate commits for the same task to capture actual co-change relations, which expects to improve the quality of co-change rules. We investigate the effects of the two characteristics on the quality of co-change rules. Empirical results using typical OSS show that co-change rules vary over time and we can detect more overlooked changes by focusing on recency. Particularly, in the Eclipse case, the maximum Recall improved up to 0.28 by recency-aware history analysis whereas the one of the baseline is 0.11. The results also show that commits aggregation for the same task can improve the recommendation performance.

Number of accesses :  

Other information