Online Learning Framework Based on User-Centric Access Behavior

With the development of 5G, mobile edge computing (MEC) has become a new paradigm to effectively reduce backhaul bandwidth consumption by deploying servers on the edge of wireless network. Different from the wired content caching, content popularity distribution exhibits diversity due to users' spatiotemporal mobility and different content lifespans on wireless edge. Learning the diversiform content popularity effectively becomes a major challenge in adaptive caching. Under the general MEC architecture, we propose an Online Learning framework based on user-centric access behavior (OLCB). OLCB reconstructs the group preference context by user-centric access behavior and learns the contextspecified content popularity timely. As context arrives, context space is adaptively partitioned in order to efficiently estimate the content popularity in different situations. Our experiments with the existing China Mobile User Data Record (UDR) data set show that the proposed approach significantly outperforms the existing solutions.