周涛-复杂网络中信息过滤.pdfVIP

Information Extracting from Complex Networks: Ranking, Predicting and Recommending Tao Zhou Web Sciences Center, UESTC Department of Modern Physics, USTC Email Address: zhutouster@ Blog:/u/p Content • Basic concepts on recommender systems - Why: Motivation and Background - What: Fundamental problem on recommending - How: Main Methods • Significance of diversity and novelty • Metrics • Diversity-accuracy dilemma • Discussion and Outlook • Ranking (appendix) • Link Prediction (appendix) Motivation and Background • The exponential growth of the Internet and World Wide Web confronts people with information overload: they encounter too much data and sources to be able to find those most relevant for them. People may choose from thousands of movies, millions of books and billions of web pages. The amount of information is increasing more quickly than our processing ability. • Personalized recommender systems provide a promising way to solve the information overload problem. • Personalized recommender systems have already been successfully applied in many e-commerce web sites, such as A . • Information filtering techniques are shifting from finding out what you want to what you like, from centralized to decentralized, from population-based to personalized. Problem Description – The Simplest Version Known information: the record of interactions between users and objects, the users’ profiles, the objects’ attributes, the content, the time stamps, the user-user relationships, etc. Required information: whether a target user will like an unselected object, and if so, to what extent he/she likes it. Basically, a personalized recommender system should provide an ordered list of unselected objects to every target user. Personalized recommender systems u