An Accurate Identifier Renaming Prediction and Suggestion Approach

Identifiers play an important role in helping developers analyze and comprehend source code. However, there exist many identifiers that are inconsistent with the corresponding code conventions or semantic functions, leading to flawed identifiers. Hence, need be regularly renamed. Even though researchers have proposed several approaches identify renaming further suggest correct for them, these only focus on a single limited number of granularities without universally considering all series sub-tokens composing completely generating new In this paper, we propose novel identifier prediction suggestion approach. Specifically, given set training code, first extract multiple granularities. Then, design five groups features from capture inherent properties themselves, relationships between as well other related entities, enclosing files, change history. By parsing history identifiers, can figure out whether specific been renamed not. These their used train Random Forest classifier, which predict needs Subsequently, renaming, entities Based intuition co-evolved relevant similar patterns sequences, could recommend those We conduct extensive experiments validate our approach both Java projects Android projects. Experimental results demonstrate average F-measure more than 89%, outperforms state-of-the-art by 8.30% 21.38% addition, achieves Hit@10 48.58% 40.97% suggesting 29.62% 15.75% respectively.