PITTCULT: Recommender System using Trusted Human Network
Danielle Hyunsook Lee, Advised Dr, Peter L. Brusilovsky · D-Scholarship@Pitt (University of Pittsburgh) · 2008
As information on the Web continues to increase, available on more and more sites, it is increasingly important to be able to collect information in one place and filter out unnecessary information for users. PITTCULT is to share and recommend cultural event information in the Pittsburgh area. This system basically utilizes human psychology to conform to opinions of friends. When people go to a music concert or exhibition, they commonly ask their friends ' opinions, and invite them to go along. Centered on this trusted human network, users can recommend items to their friends and rate their friends ' taste about a certain genre of cultural events. As a popular recommendation technology, collaborative filtering-based recommendation (CF) is designed to find like-minded peers and is known to recommend more diverse and serendipitous recommendations. It works well in a domain where contents are not easily comparable, like music, movies, jokes and cultural events (Schafer, et al., 2007). However, CF has a data sparsity problem. If the users ' ratings are relatively fewer than the number of items, there will be too little overlap among users to make recommendation. In addition, there is cold-start problem for new users who have no or insufficient ratings (Massa & Avesani, 2007; Schafer, et al., 2007). Another problem of CF is created by ad-hoc users. Users possessing malicious intentions can harness others ' ratings to make a profit or distort the system