Maximizing recommender's influence in a social network: An information theoretic perspective

We study a social network in which individuals make decisions influenced by a recommender as well as the previous actions taken by themselves or other users. The recommender aims to tailor its suggestions to maximize the benefit from utilizing social interactions. We refer to this benefit as the recommender's influence which, in essence, measures the value of controlling the specific suggestions offered to the individuals. We show that this influence can be quantified by the directed information between the suggestions and people's actions. Accordingly, we identify the precise relationship between the social network-based recommendation system and a finite state communication channel whose capacity analysis provides the solution for the influence maximization problem for the recommender. Our results demonstrate that a recommender that tailors its suggestions based on the social dynamics of its customer base can have a significantly greater influence.