Semantic Grounding of Tag Relatedness in Social Bookmarking Systems
Someone from Tagora spoke about the relationships between different tags that get applied to the same document, and the tags that different people apply to the same documents. The looked at the ternary relation (user, resource, tag) and tried to infer measures of similarity. What tags are “related” to others, or to other pages? What measures capture useful notions of relatedness? One simple example is co-occurrence—how often two tags were assigned to the same resource. But there are more sophisticated measures that are better. Someone else tried using a page-rank style algorithm where you start with high weight on one tag and see what other tags receive high weight from it (page-rank in an undirected graph is kind of overkill; mathematically it reduces to just looking at the degree of nodes). They look at more context—two tage are similar if they appear with the same sets of tags. They also look at grounding tags into wordnet to find semantic matches—measuring relatedness by distance in wordnet, for example. Different relatedness measures can discover different relationships: specific versus general, or synonyms. You can check stuff out at bibsonomy.org