Code stylometry, a metric learning approach

Authorship attribution, also recognized as code stylometry, has always been a milestone in obtaining important information for what concerns plagiarism and de-anonymization tasks, assessing the author in several different ways through the years. The proposed work revolves around the whole problem, starting with the mining of a new dataset which faces data scarcity and domain bias problems that afflicted the former works. Diving then into a new machine learning model design, derived from the former state-of-the-art techniques, which tries to gain advantages from Natural language process practices adopted by the newest language models. The problem is then tackled by moving through a metric learning technique, dealing for the first time with the stylometry problem as a querying snippet mechanism that allows a zero-shot inference over authors not present in the training set.