Tag Ranking Algorithms And Implementations

Tag is a popular element in WEB 2.0. Tags are used in many famous document or image sharing sites. People can associate some tags with web pages they are interested in. Tags contain user's understanding about the content. Consequently, they contain very important information. They are exploited in applications including document indexing and retrieval, tag or document recommendation, user's interest discovering, etc. However, the rich information in tags is usually difficult to use because tag lists on web pages may contain noise and the simple ranking by frequency does not precisely correspond to the semantic importance of the tags. To better exploit the knowledge in tags, we need a better algorithm to sort tags.In this paper, we analyzed the situation of tag using, then use graph ranking or other data mining algorithms to rank tags. We give three tag ranking algorithms for document or non-document sites. The first algorithm is based on interest propagation model, we construct a "user-tag-document" graph, then use Manifold Ranking algorithm to rank tags. The second algorithm named EigenTag is based on the mutual reinforcement relationship between users and tags. The third algorithm is based on random walk model on a tag graph.We use manual ranking as our standard ranking, and use NDCG to measure the ranking algorithms. Experiment shows that compared with the original web ranking, the algorithms presented in this paper have a big improvement.