A Machine Learning Approach for Source Code Similarity via Graph-Focused Features

Source code similarity aims at recognizing common characteristics between two different codes by means of their components. It plays a significant role in many activities regarding software development and analysis which have the potential of assisting software teams working on large codebases. Existing approaches aim at computing similarity between two codes by suitable representation of them which captures syntactic and semantic properties. However, they lack explainability and generalization for multiple languages comparison. Here, we present a preliminary result that attempts at providing a graph-focused representation of code by means of which clustering and classification of programs is possible while exposing explainability and generalizability characteristics.