Automatic detection of architectural bad smells through semantic representation of code

Bad design decisions in software development can progressively affect the internal quality of a software system, causing architecture erosion. Such bad decisions are called Architectural Smells (AS) and should be detected as soon as possible, because their presence heavily hinders the maintainability and evolvability of the software. Many detection approaches rely on software analysis techniques which inspect the structure of the system under analysis and check with rules the presence of AS. However, some recent approaches leverage natural language processing techniques to recover semantic information from the system. This kind of information is useful to detect AS which violate "conceptual" design principles, such as the separation of concerns one. In this research study, I propose two detection strategies for AS detection based on code2vec, a neural model which is able to predict semantic properties of given snippets of code.