Reinforcement learning based recommender systems for web applications: scenarios of radio and game aggregators

• Autor(a) principal: Batista, André
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10284/12538
• topic: Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

With the progression of time and the continuous evolution of digital entertainment services such as YouTube, Netflix, Spotify, and online gaming platforms, recommendation systems have become an essential daily tool for users. These systems save users time by analyzing various content, facilitating searches, and suggesting relevant content in a personalized manner. However, the same level of personalization is not consistently found across all media domains, particularly within the radio streaming and gaming sectors. For radio streaming, users must currently search explicitly for a specific internet radio station’s name either through a search engine or a radio aggregator like myTuner. This process can lead to significant time consumption and potential loss of user interest, especially if users are unsure of the type of radio they wish to listen to. A similar challenge is faced in the gaming industry, where an overwhelming array of choices can lead to difficulty in discovery and decision-making for players. Furthermore, even within aggregators that offer some form of recommendation, a convergence is often observed where the most popular items dominate the top spots. This dynamic makes it challenging to discover lesser-known radio stations or games, resulting in a homogenized user experience. In response to these challenges, this thesis presents the design, implementation, and empirical evaluation of a recommendation system, specialized in the aforementioned domains. Utilizing machine learning and emphasizing deep reinforcement learning techniques, the system optimizes content suggestions, considering variables such as language, region, and user history, fostering personalized recommendations. The system has been deployed in two distinct production scenarios, demonstrating promising preliminary results. It exhibits consistent improvement and adaptability over time, reinforcing its practical applicability and effectiveness.