Study On Collaborative Filtering Based On User Interest Sub-Class

With the development of E-commerce,the magnitudes of users and web documents grow rapidly, and result in the extreme data sparseness of users and system expansibility. The traditional algorithms of collaborative recommendation can't solve the problem very well. To address the main challenge of Web Personalized Information Recommend(WPIR), This paper use the user's Interest Sub-Class as the paper's clue and consider realizing the collaborative recommendation algorithm based on user Interest Sub-Class as our goal ,we did some useful grope and research on WPIR and the main work includes such aspects as below:First,we proposed the ideology that fractionized user profile to multi Sub-Class Interest and also gave out the framework of new collaborative recommendation system which is based on user Interest Sub-Class.Then we gave out the detailed description about the function of the main part and the procedure of information recommendation.Second, aim at the problem of user profile,a new user interest model based on Sub-Class is proposed.It makes for reducing the dimension of user profile,finding the nearest neighbors and enhancing the efficiency of recommendation.Afterward, we gave out a method of classifying user interest based on ICT( Interest Category Tree) for creating the user profile automatically.With the improvement of the Rocchio feedback algorithm,we introduced the lemma weight adjusting algorithm based on VSM relevance feedback.It makes for updating user profile automatically.Third,a collaborative recommendation algorithm based on user Interest Sub-Class is porposed. By introducing the concept of sub-classes,it makes it more easy and accurate to find the similar neighbors of a user, even if their interests are very different as a whole.The algorithm can resolve the problem of extreme data sparseness and thereby makes for finding new information.Furthermore,we improved the method of calculating keywords weight from two aspects-the location of keyword and semantics in order to describe Interest Sub-Class(section 3.4). With the method of classifying enormous documents,it realizes reducing the dimension of traditional user-item matrix and makes the problem of expansibility easy to be resolved(section 5.2).Finally,with the comparison in multi aspects among our method and other...