Integrazione di algoritmi di raccomandazione session-aware in un framework di valutazione e analisi delle prestazioni

The exponential growth of e-commerce, social networks and media streaming website sizes has made the recommender system a very popular type of tool to filter the large amounts of data that the user is exposed while surfing on the net. The purpose of the recommender system is to provide recommendations to the user about choices, via forecasts of what (s)he is supposed to like. Among the existing recommender systems approaches, in this thesis we focus on sessionaware recommendation, that also uses interactions from previous sessions (long-term preferences) rather than only relying on the interaction from the current session (short-term preferences). The latter type of algorithms are denoted as session-based RS. As the research on session-aware is still scarce, the performance of proposed session-aware models is usually compared with that of session-based models. This is the reason why the state-of-the-art in this field is hard to understand. The goal of the research I have collaborated to is an extensive comparison between session-aware algorithms as well as among session-aware and sessionbased ones, in order to clarify the state-of-the-art in this area. Specifically, the contribution of this thesis is the integration of three algorithms within an evaluation framework, and the evaluation with a systematic approach after hyperparameter optimization. The algorithms integrated and compared are NSAR and SHAN. The comparison revealed that simple techniques based on nearest neighbors outperform all recent neural techniques for the examined datasets, and recent session-aware models were not better than session-based ones.