Deep Graph Matching and Searching for Semantic Code Retrieval

Code retrieval is to find the code snippet from a large corpus of source repositories that highly matches query natural language description. Recent work mainly uses processing techniques process both texts (i.e., human language) and snippets machine programming language), however neglecting deep structured features codes, which contain rich semantic information. In this paper, we propose an end-to-end graph matching searching (DGMS) model based on neural networks for task retrieval. To end, first represent with unified graph-structured data, then use proposed retrieve best snippet. particular, DGMS not only captures more structural information individual or but also learns fine-grained similarity between them by cross-attention operations. We evaluate two public datasets representative languages Java Python). Experiment results demonstrate significantly outperforms state-of-the-art baseline models margin datasets. Moreover, our extensive ablation studies systematically investigate illustrate impact each part DGMS.