Evaluating similarity measures for emergent semantics of social tagging

摘要: Social bookmarking systems are becoming increasingly important data sources for bootstrapping and maintaining Semantic Web applications. Their emergent information structures have become known as folksonomies. A key question harvesting semantics from these is how to extend adapt traditional notions of similarity folksonomies, which measures best suited applications such community detection, navigation support, semantic search, user profiling ontology learning. Here we build an evaluation framework compare various general folksonomy-based measures, derived several established information-theoretic, statistical, practical measures. Our deals generally symmetrically with users, tags, resources. For purposes focus on between tags resources consider different methods aggregate annotations across users. After comparing the ability tag predict user-created relations, provide external grounding by user-validated proxies based WordNet Open Directory Project. We also investigate issue scalability. find that mutual distributional micro-aggregation users yields highest accuracy, but not scalable; per-user projection collaborative aggregation provides scalable approach via incremental computations. The results consistent resource similarity.