Textual Analysis for Code Smell Detection

Abstract

The negative impact of smells on the quality of a software systems has been empirical investigated in several studies. This has recalled the need to have approaches for the identification and the removal of smells. While approaches to remove smells have investigated the use of both structural and conceptual information extracted from source code, approaches to identify smells are based on structural information only. In this paper, we bridge the gap analyzing to what extent conceptual information, extracted using textual analysis techniques, can be used to identify smells in source code. The proposed textual-based approach for detecting smells in source code, coined as TACO (Textual Analysis for Code smell detectiOn), has been instantiated for detecting the Long Method smell and has been evaluated on three Java open source projects. The results indicate that TACO is able to detect between 50% and 77% of the smell instances with a precision ranging between 63% and 67%. In addition, the results show that TACO identifies smells that are not identified by approaches based on solely structural information.