Know thy neighbor: Community-aware recovery of content selection preferences

We design two strategies for recovering unknown content selection preferences in online communities. The techniques take advantage of the community graph and the peers' affinities, expressed through its edge weights, to optimize the computation of the missing data. The first strategy is distributed and comprises a local computation step and a message passing step that are iteratively applied at each vertex until convergence. We carry out a random walk based analysis of its operation and verify the analytical findings via numerical experiments. The second strategy is centralized and involves a sparsifying transform of the content preferences represented as a function over the community graph. We solve the related optimization problem of recovering the unknown preferences via an iterative algorithm based on variable splitting and alternating direction of multipliers. We take into account the data specifics by incorporating multiple regularization terms into the optimization. We investigate the underpinnings of the sparse reconstruction technique via simulations that reveal its characteristics and how they affect its performance. We also carry out experiments using Twitter data on which we further study the performance of our strategies and verify the modeling assumption made in the context of the decentralized one. Our experiments include a comparison to common reference methods. We show that our message passing technique outperforms the reference methods by a considerable margin. We also show that though our multi-regularized sparse reconstruction technique improves over conventional sparse recovery, it still suffers from the graph-signal smoothness assumption it implicitly considers.