Analyzing a Semantics-Aware Bug Seeding Tool's Efficacy: A qualitative study with the SemSeed tool.

Software developers can benefit from machine learning solutions to predict bugs. Machine learning solutions usually require a lot of data to train a model in order to achieve reliable results. In this context, developers use bug-seeding approaches to generate synthetic bugs, which should be similar to human-made bugs. A recent state-of-the-art tool, called SemSeed, uses a semantics-aware bug seeding approach in order to hopefully achieve more realistic bugs. In this study, we report on the investigation of SemSeed’s efficacy. We create a survey that shows developers a bug and asks whether it is a Real or Synthetic bug. We collected and analyzed the answers from 47 developers, and we show that SemSeed can be very accurate in seeding realistic bugs.