Harvesting the Tagging Power: An Approach to Social Recommendation of Information Products
31 Pages Posted: 19 Jan 2012
Date Written: January 18, 2012
Social tagging, as a flexible and effective information management platform, has proliferated rapidly on the Internet and intranets in the past few years. Social recommendation, which automates the process of recommending information products of potential interest to users based on tags, can improve the user experience of information discovery and promote the consumption of information products on various kinds of content delivery sites. Although a number of tag-based social recommenders have been developed, several problems resulting from the special nature of tagging data, including tagging bias, vocabulary problem, and data sparsity, have not been adequately addressed and largely limit the performance of existing recommenders. In response to these problems, this paper proposes a tag-based semantic analysis model to capture the semantics emerging in tagging data and reformulates the recommendation problem as a constrained optimization problem on both labeled (selected) and unlabeled items. An iterative learning framework developed to solve this optimization problem combines the self-organization theory in cognitive science, the probabilistic latent semantic analysis framework in information retrieval and the positive unlabeled learning methodology in machine learning. Empirical studies using social tagging datasets from three different domains demonstrate the efficacy of the proposed method.
Keywords: social tagging, collaborative filtering, social recommendation, information discovery, knowledge management, sparsity, positive unlabeled learning
Suggested Citation: Suggested Citation