Targeted Filter Bubbles Mitigating via Edges Insertion

The emergence of filter bubbles leads to various harms. To mitigate filter bubbles, some recent works select the seeds for different viewpoints to minimize the formation of bubbles under the influence propagation model. Different from these works where the diffusion networks remain unchanged, in this paper, we conduct the first attempt to mitigate filter bubbles via edge insertion. Besides, to be more generalized, we focus on mitigating filter bubbles for the given target node set since the audiences can be different for different scenarios. Specifically, we propose the concept of openness score for each target node, which serves as a metric to assess the likelihood of this node being influenced by multiple viewpoints simultaneously. Given a directed graph G, two seed sets, a positive integer k and a target node set, we aim to find k edges incident to the given seeds such that the total openness score is maximized. We prove the NP-hardness of problem studied. A baseline method is first presented by extending the greedy framework. To handle large graphs efficiently, we develop a sampling-based strategy. A data-dependent approximation method is developed with theoretical guarantees. Experiments over real social networks are conducted to demonstrate the advantages of proposed techniques.