Design Of Personalized Information Delivery In Multimedia System

A personalized information delivery system is designed according to the characteristics of multimedia system, which has various kinds of resources and has large amounts of data. In order to help users find resources they are interested in as soon as possible, personalized recommender system is satisfied and efficient. The author designs the system from method of organization of resource, building of user profile, interest extension to interaction between user and system. In this thesis, concept-based network is given to organize system resource, human memory-based user model is designed to adapt the evolvement of user behavior and trust-enhanced collaborative filtering method is designed to extend user interest with user trustiness. The idea of Random Walk is adapted to generate personalized ranking. Content-based recommender method and collaborative filtering method are combined to delivery personalized information. Some strategies are given to decrease the influence of their shortcomings:(1) Build concept based network model via user access behavior analysis to find out relationship between different concepts. Concept based network model benefits the extension of user interests when system generates personalized information to users.(2) Build user model that contains long-term interest model and short-term interest model based on memory model, which make system adapt concurrently when change of user interest happens.(3) Trust enhanced collaborative filtering method is applied to extend user interesting model by using access history of similar neighbors. It can help system to recommend new information to user who has never accessed.(4) Bi-section clustering algorithm is designed in order to avoid the influence of data sparsity, system will cluster user models into several sub-classes to make user with similar taste allocate into a same class.(5) Design a personalized ranking algorithm based on Random Walk, such algorithm can promote the personalization ability of the recommender system.Empirical simulation is given to prove that algorithms designed in this thesis are efficient and the system can generate personalized results.