Discriminative Clustering for Content-Based Tag Recommendation in Social Bookmarking Systems .

We describe and evaluate a discriminative clustering approach forcontent-based tag recommendation in social bookmarking systems. Our approachuses a novel and efficient discriminative clustering method that groups postsbased on the textual contents of the posts. The method also generates a rankedlist of discriminating terms for each cluster. We apply the clustering method tobuild two clustering models – one based on the tags assigned to posts and theother based on the content terms of posts. Given a new posting, a ranked list oftags and content terms is determined from the clustering models. The final tagrecommendation is based on these ranked lists. If the poster’s tagging history isavailable then this is also utilized in the final tag recommendation. The approachis evaluated on data from BibSonomy, a social bookmarking system. Predictionresults show that the tag-based clustering model is more accurate than the term-based clustering model. Combining the predictions from both models is betterthan either model’s predictions. Significant improvement in recommendation isobtained over the baseline method of recommending the most frequent tags forall post