A confidence-based recommender with adaptive diversity

This paper proposes a new confidence-based Collaborative Filtering (CF) technique, and studies the impact of incorporating an adaptive diversity technique for recommending composite products and services. We are demonstrating the need and value of incorporating multi-criteria ranking in new generation recommender systems to extend their capabilities and provide better quality results. First, a light yet efficient CF technique is presented to learn the preference of the user from history rating data, and then estimate similarity among users based on confidence measure. Second, An adaptive diversity algorithm is introduced. The algorithm is randomized, and iteratively relaxes the selection by the Greedy algorithm, with an exponential probability distribution. Third, we conducted extensive experimental studies on the efficacy of the proposed CF method proposed to compare precision of our ranked recommendations with broadly used CF techniques, we achieved a precision of 90% on average, in addition, the adaptive diversity technique consistently converges to find an optimal or near-optimal solutions on a dataset of 10 million ratings from Movielens.