Personalized Recommendation Algortithm In Tagging System Based On Clusting Analysis

With the rapid growth and popularity of Web2.0technologies, data on the Internet increases explosively. Recommendation system is built to help users find resources more effectively. It is crucial for the recommendation system to understand a user’s inclination in resources clearly. Generally speaking, even the user may find it hard to depict his inclination in resources, and most users would not bother to describe their interests elaborately during their interaction with the recommendation system. Tagging system is designed to get more subjective opinions from the users. Users can tag resources freely in a tagging system. These records of a user’s tagging history can help analyze his interest. Thus, recommendation algorithm has to decide how to exploit users’tagging history and rank resources based on their similarity with a specific user.The major innovative work of this thesis is as following:(1) Considering tagging systems’characteristics, an algorithm that mixed the two concepts, i.e. content-based similarity and user-based similarity is proposed. In this algorithm, tagging records are converted into a tagging cluster, which are then served as an intermediary between users and resources.(2)"Synonyms" has been never studied before in any recommendation algorithms for tagging systems. In this thesis,"community detection" algorithms are introduced in dealing with "synonym" tags, which is verified effectiveness.(3) A prototype of tagging recommendation system is developed, including the "synonyms " tagging node splitting algorithm, condensed clustering algorithm and clustering tagging recommendations algorithm. The accuracy rate, recall rate, coverage and other indexes are tested, and the results show that the proposed algorithm in the process of tag recommendation is superior to the traditional content-based, collaborative filtering, and K-means clustering recommendation algorithm.