Optimizing Developer Assignment with Language Model and Knowledge Graph

Automatic bug triage is an important task of software maintenance. It takes a lot of human resources to choose the adequate developer for a given bug, which can be saved by automatic developer assignment techniques. There are various approaches to solving this problem, including information retrieval and machine learning-based methods. In this paper, we create a model for the automatic bug triage problem for a smaller dataset and we search for the optimal feature set for the huBERT model. In the literature, the results are usually tested on large datasets of big companies' repositories but it is not investigated how well these techniques perform on smaller repositories. In this work, we evaluate our method on a smaller repository that is actively used and we search for adequate features. Furthermore, we create a Knowledge Graph(KG), which is used for developer profiling. This KG includes all necessary information about the project for the bug assignment task and developer profiling, it also contains the bug-tossing graph as a subset of our knowledge graph. Based on our experiments different features should be used for smaller repositories than larger ones.