Inferring Test Models from Kate's Bug Reports Using Multi-objective Search

Models inferred from system execution logs can be used to test general system behaviour. In this paper, we infer test models from user bug reports that are written in the natural language. The inferred models can be used to derive new tests which further exercise the buggy features reported by users. Our search-based model inference approach considers three objectives: (1) to reduce the number of invalid user events generated (over approximation), (2) to reduce the number of unrecognised user events (under approximation), (3) to reduce the size of the model (readability). We apply our approach to 721 of Kate’s bug reports which contain the information required to reproduce the bugs. We compare our results to start-of-the-art KLFA tool. Our results show that our inferred models require 19 tests to reveal a bug on average, which is 98 times fewer than the models inferred by KLFA.