Improving recommendation diversity using tensor decomposition and clustering approaches

Diversity and novelty of items in recommendation, and coverage of overall recommended items are emerging recommendation quality measures for user and system respectively. These measures tend to alleviate the problem of user satisfaction in terms of redundancy in recommendations caused by accuracy-oriented approaches. This work proposes solution to provide diversity in different modes when we are dealing with multi-mode data. To provide diverse suggestion list of communities to join, the proposed framework uses Tensor Decomposition to reveal latent topics among multi-mode data including communities, users and social tags. It exploits co-clustering approaches on decomposed components to extract clusters of communities based on user similarity and tag similarity. Afterwards, clusters' information is used to develop and apply re-ranking algorithms, which leads to improvement in diversity and coverage of recommended lists.