Comparing Attract Functions for the Reflexion Analysis Regarding the Usage of Dependencies and Words

In this paper, we propose a new attraction function to be used to make mapping suggestions in the context of the reflexion analysis. The reflexion analysis allows developers to reconstruct or verify their architecture with regards to its implementation. To be able to do that, the reflexion analysis requires a mapping of source-code elements onto architecture components. This mapping can be a labor-intensive manual task. The attraction functions may ease this task by recommending mappings. Our method draws on existing methods and combines de-pendency information with term similarity aggregated by way of dependencies. We describe an evaluation that compares our new method against three other existing methods, namely, Count Attract and NB-Attract. All these attraction functions are applied within the framework of the HugMe method. The computed suggestions for all methods are compared against oracle mappings provided by other researchers. As measures for the comparison, we use recall and precision relative to an oracle. Our results show that our new method can be applied successfully and performs similar to NB-Attract. Furthermore, whether the usage of dependencies or terms is more suitable for high-quality mappings seems rather system dependent, even though the attract functions using terms are performing better more often. Our method, as a combination of dependency and term similarity, could be particularly well suited in situations in which one does not want to rely on only dependencies or only term similarity.