Accurate Content Push for Content-Centric Social Networks: A Big Data Support Online Learning Approach

With the rapid growth of the social network, information overload becomes a critical issue. Service providers push a lot of redundant contents and advertisements to users every day. Thus, users' interests and the probability of reading them have dropped considerably and the network load is wasted. To address this issue, accurate content push is needed, where the main challenges are proving precise descriptions of users and supporting the big data nature of users and contents. Content-centric networking (CCN) has emerged as a new network architecture to meet today's requirement for content access and delivery. By using the named content, CCN makes it possible to track users' real-time interests and motivates us studying a novel content accurate push (or called content recommendation) system. In this paper, we model this issue as a novel contextual multiarmed bandit based Monte Carlo tree search problem and propose a big data support online learning algorithm to meet the demand of content push with low cost. To avoid destroying CCN's energy efficient feature, the energy consumption is considered into our module. Then, we theoretically prove that our online learning algorithm achieves sublinear regret hound and sublinear storage, which is very efficient in the big data context and do not increase the network burden. Experiments in an offline collected dataset show that our approach significantly increases the accuracy and convergence speed against other state-of-the-art bandit algorithms and can overcome the cold start problem as well.