Active Tag Recommendation System Based On The Query Log Analysis

With the rapid development of Internet and the abroad social identity of digital resources, the advantages of digital libraries became more fully reflected and received a high degree of attention in many countries. In the past ten years, the fast development of digital libraries makes it become one of the most important approaches for people to access information. Recommendation system is one of the primary services in digital library. This paper proposes an in-depth research on improving digital library recommendation results, to effectively help users to find digital books and other resources, based on user’s query-click log.The traditional digital book online reading system is a request-response model with one-way flow of data, all start from users and the requests from users need to be precise. However, in real world, the users’ requests are often uncertain, open, changing and ambiguous. For this situation, traditional online digital reading service can’t meet user’s requirements, and users need more time to find what they want.The main contributions of this paper as follows:First, the structure of recommendation system based on Spectral Clustering algorithm and crowdsourcing technology. In this structure, users’ query log is primary input, and then use Spectral Clustering algorithm to get the results of clustering, then using crowdsourcing system, to improve the result of clustering, and along with the times of crowdsourcing, the clustering result will better and better.Second, real tag recommendation system application on CADAL digital library system. We use user’s query and click log, collected by CADAL Log Collection System, then through simple data cleaning and settle, and then use the structure above mentioned, to get clustering result. By crowdsourcing system, we improve the result of clustering many times, in the end, we recommend query as tag to users when they searching or reading in CADAL system.