Reinforcement Learning for Neural Collaborative Filtering

Artificial Intelligence (AI) has become an integral part of many modern technologies, with significant advances in real-world applications. With the rise of deep learning methods in the past decade, systems that utilize artificial neural networks have produced remarkable results in a variety of fields, such as computer vision, natural language processing and voice recognition, with performance exceeding that of humans in many cases. Examples of practical applications include self-driving cars, state-of-the-art text translators and generators, and robust object detection algorithms. The field of Recommender Systems has also taken advantage of this progress, with a plethora of novel neural networks being proposed that achieve significant improvements in providing automatic recommendations regarding the preferences of users. Aiming to further explore this area and the capabilities of different deep neural networks, we train top-performing neural collaborative filtering recommender systems under a reinforcement learning setting, which has been largely unexplored in favor of supervised learning for these models. Experimental evaluation on the MovieLens-1m dataset showcases the behavior of different neural architectures under this setting, and how the introduction of sophisticated components contributes to improved performance.