Research On Book Recommendation System Based On Tag Topic Model

We are now in an era when the information booms. People are overwhelmed by the information coming through the Internet. The developments of computer technology and web technology have made the access to knowledge become more and more convenient. The thriving of digital library is the best example. Over the past ten years, due to the great social benefit in dissemination of knowledge, digital library have drawn the attention of many governments. While the number of resources in the digital library increases, many users feel bewildered when facing so many books. How can we make it easier for users to find the books they might be interested in? By using the tag information associated with books in CADAL Digtial Library, we proposed a recommender system based on the topic model of tags.The main contribution of this dissertation is as follows:First, we proposed a method based on LDA to generate topic model of the tags and users. In order to mine the latent topic in the tag information, we use Gibbs sampling method to approximate the latent factor of the topic-word model. Then we utilize the topic-word model through LDA-inference algorithm to gain the user-topic model. We can find a given user’s neighbor by compute the user-user similarity in terms of user-topic model.Second, we researched an ensemble based method to capture the transition of users’interests. We split the user behavior data stream using time windows. Based on each time window, we build a user-topic model to describe user’s interest during that time. Then we use a conception drift detection method to determine whether a user’s interest have changed over time. By using an ensemble of user-topic model which are learned from various time window and can represent user’s different interest, we can agilely catch up with user’s latest read trend without radically discard the knowledge learnt in the past.