Use of Context for Recommending Code: an Approach Based on Frequent Pattern Mining

Abstract

During creating a class, several times a developer must be conscious of regularities that should be complied in order to satisfy an intended architectural design. This paper presents an approach for code recommendation, which apply concepts of frequent pattern mining to take advantage of the use of naming conventions and the organization of source code in software development. The proposal has been evaluated in projects of the organizations Apache and Eclipse. The results have shown that significant terms and regularities, in form of frequent relationships, could be mined. The approach is particular in the sense that it enables using naming patterns and package hierarchies as context for recommending code.