Diversification in tag recommendation system using binomial framework

Abstract

Diversity has been recently identified to be one of the major contributors for improving performance of a recommendation system in terms of user satisfaction. In social tagging-based systems, tag recommendation is used to suggest tags to a user for a resource. Diversity in tag recommendations has been overlooked in traditional tag recommendation techniques so far. In this paper, we propose a novel tag recommendation system that recommends a diverse yet relevant set of tags to the user. Our system utilizes a simple greedy-based algorithm that optimizes an objective function, defined using recently proposed binomial framework for diversification that considers high coverage and penalizes redundancy. To incorporate user preference to the recommendations, tags are suggested based on user’s affinity with tags along with diversity. We experimented with the MovieLens 10M dataset. Effectiveness of the system has been evaluated off-line with respect to both relevance and diversity.