The Study Of Methods Of Content-Based And Collaborative Filtering In Science Literature

Text Filtering is of great value and used widely with.the increasing online readable text information because it can help users get information which they are interested in and realize personalized information service.There are two kinds of text filtering: Content-based Filtering and Collaborative Filtering. Content-based Filtering mainly adopts some technologies such as natural language processing, artificial intelligence and probability statistic to analyze text content, then calculate degree of similarity between content vector and user profiles vector and select high correlative text to registered users. Collaborative Filtering mainly makes use of users' opinions who have similar interest to predict and recommend. Now it has been used in personalized recommendation system. But with the system scale enlarging, its efficiency gradually declines and some problems such as Sparsity, Scalability and Early rater will appear.First, two kinds of text filtering approach are described, then collaborative filtering technologies are deeply studied. Aiming at some problems of collaborative filtering technologies, we have explored item-based collaborative filtering algorithm, which solves effectively Sparsity and Scalability problems. Second, a new text filtering approach that combines content-based filtering with collaborative filtering is proposed, which makes full use of the advantages of content-based filtering and collaborative filtering and solves effectively Early rater problem and improves system performance. Lastly, the construction approach for User Profile is described, which includes explicit feedback learning and implicit feedback learning. Three foundations of the experiment system for updating User Profiles are also described at the end of the thesis.In order to evaluate our new collaborative filtering algorithm and combined approach, we have developed a Prototype System for Chinese computer science literature automatic filtering. The results of experiment prove that improved filtering algorithm is better than user-based filtering algorithms and combined filtering approach has better system performance.