E-Commerce and Web Technologies
暂无分享,去创建一个
To save customers’ time and efforts in searching the goods in the Internet, a customized recommender system is required. It is very important for a recommender system to predict accurately by analyzing customer’s preferences. A recommender system utilizes in general an information filtering technique called collaborative filtering, which is based on the ratings of other customers who have similar preferences. Because a recommender system using collaborative filtering predicts customer’s preferences based only on the items without useful information on the attributes of each item, it may not give high quality recommendation consistently to the customers. In this paper we show that exploiting the attributes of each item improves prediction quality. We analyze the dataset and retrieve the preferences for the attributes because they have not been rated by customers explicitly. In the experiment the MovieLens dataset of the GroupLens Research Center has been used. The results on various experiments using several neighbor selection methods which are quite popular techniques for recommender systems show that the recommender systems using the attributes provide better prediction qualities than the systems without using the attribute information. Each of the systems using the attributes has improved the prediction quality more than 9%, compared with its counterpart. And the clustering-based recommender systems using the attributes can solve the very large-scale dataset problem without deteriorating prediction quality.
[1] Jaideep Srivastava,et al. Data Preparation for Mining World Wide Web Browsing Patterns , 1999, Knowledge and Information Systems.
[2] MAGDALINI EIRINAKI,et al. Web mining for web personalization , 2003, TOIT.