DARES: An Asynchronous Distributed Recommender System Using Deep Reinforcement Learning

Traditional Recommender Systems (RS) use central servers to collect user data, compute profiles and train global recommendation models. Central computation of RS models has great results in performance because the are trained using all available information full profiles. However, centralised require users share their whole interaction history with server general not scalable as number interactions increases. RSs also have a point attack respect privacy, stored centrally. In this work we propose DARES, an distributed recommender system algorithm that uses reinforcement learning is based on asynchronous advantage actor-critic model (A3C). DARES developed combining approaches A3C federated (FL) allows keep data locally own devices. The architecture consists (i) local devices (ii) updates computed We evaluate proposed well-known datasets compare its against state art algorithms. show although being asynchronous, it can achieve comparable many cases better than current state-of-the-art