Linguistic Driven Refactoring of Source Code Identifiers

Identifiers are an important source of information during program understanding and maintenance. Programmers often use identifiers to build their mental models of the software artifacts. We have performed a preliminary study to examine the relation between the terms in identifiers, their spread in entities, and fault proneness. We introduced term entropy and context-coverage to measure how scattered terms are across program entities and how unrelated are the methods and attributes containing these terms. Our results showed that methods and attributes containing terms with high entropy and context-coverage are more fault-prone. We plan to build on this study by extracting linguistic information form methods and classes. Using this information, we plan to establish traceability link from domain concepts to source code, and to propose linguistic based refactoring.