Method Name Prediction for Automatically Generated Unit Tests

Writing intuitively understandable method names is an important aspect of good programming practice. The method names have to summarize the codes' behavior such that software engineers would easily understand their purpose. Modern automatic testing tools are able to generate potentially unlimited number of unit tests for a project under test. However, these tests suffers from unintelligible unit test names as it is quite difficult to understand what each test triggers and checks. This inspired us to adapt the state-of-the-art method name prediction approaches for automatically generated unit tests. We have developed a graph extraction pipeline with prediction models based on Graph Neural Networks (GNNs). Extracted graphs contain information about the structure of unit tests and their called functions. The experiment results have shown that the proposed work outperforms other models with precision = 0.48, recall = 0.42 and F1 = 0.45 results. The dataset and source codes are released for wide public access.