Personalized Resources Recommendation In Social Tagging Systems

With the development of social tagging systems,the emergence of tag technology has brought new opportunities for personalized recommendations. On one hand, tags reflect users‘preferences; on the other hand, they describe resources with natural language. Tags‘rich semantic information has brought important data source for personalized recommendations, which could effectively solve the sparse problems in traditional collaborative recommendation algorithm.This thesis focused on how to recommend resources with the help of tags in social tagging systems. To overcome the problems of polysemy and synonyms, tag semantic similarity is calculated dynamically in the structure of Wikipedia categories. K-dist and percentile is used to reduce DBSCAN‘s sensitivity of threshold and the number of the core objects. This thesis introduced the algorithm of resources recommendation based on tag clustering. K nearest neighbors were calculated through users‘interest of clusters. Then, we could determine the interesting clusters, which were the features of users. Based on Bayesian classification, the thesis calculated the posterior probability of how users were interested in resources.