Master Thesis in Explainable Software

1. Automated generation of commit messages with an encoder-decoder architecture
The objective is to automatically generate the commit message related to a commit pushed in a version control system. This should be done by only considering the code changes associated with the commit (i.e., the patch). We intend to beat the state-of-the-art approaches by building a novel architecture for generating descriptions.

2. Mis-information classification for French text using pre-trained language models
The objective is to asses the performance of language models that have been pre-trained on French corpus (e.g., CAMEMBERT, FLAUBERT) on the classification problem of detecting fake news automatically.