Jianming He and Wesley W. Chu Computer Science Department University of California, Los Angeles, CA 90095 firstname.lastname@example.org, email@example.com
Abstr act. Social influence plays an important role in product marketing. However, it has rarely been considered in traditional recommender systems. In this paper we present a new paradigm of recommender systems which can utilize information in social networks, including user preferences, item's general acceptance, and influence from social friends. A probabilistic model is developed to make personalized recommendations from such information. We extract data from a real online social network, and our analysis of this large dataset reveals that friends have a tendency to select the same items and give similar ratings. Experimental results on this dataset show that our proposed system not only improves the prediction accuracy of recommender systems but also remedies the data sparsity and coldstart issues inherent in collaborative filtering. Furthermore, we propose to improve the performance of our system by applying semantic filtering of social networks, and validate its improvement via a class project experiment. In this experiment we demonstrate how relevant friends can be selected for inference based on the semantics of friend relationships and finer-grained user ratings. Such technologies can be deployed by most content providers.
1 Intr oduction
In order to overcome information overload, recommender systems have become a key tool for providing users with personalized recommendations on items such as movies, music, books, news, and web pages. Intrigued by many practical applications, researchers have developed algorithms and systems over the last decade. Some of them have been commercialized by online venders such as Amazon.com, Netflix.com, and IMDb.com. These systems predict user preferences (often represented as numeric ratings) for new items based on the user's past ratings on other items. There are typically two types of algorithms for recommender systems -- content-based methods and collaborative filtering. Content-based methods measure the similarity of the recommended item (target item) to the ones that a target user (i.e., user who receives recommendations) likes or dislikes [25, 22, 30] based on item attributes. On the other hand, collaborative filtering finds users with tastes that are similar to the target user’s based on their past ratings. Collaborative filtering will then make recommendations to the target user based on the opinions of those similar users [3, 5, 27]. Despite all of these efforts, recommender systems still face many challenging problems. First, there are demands for further improvements on the prediction ac-
curacy of recommender systems. In October 2006, Netflix announced an open competition with the grand prize of $1,000,000 for the best algorithm that predicts user ratings for films (http://www.netflixprize.com). The improvement in the prediction accuracy can increase user satisfaction, which in turn leads to higher profits for those e-commerce websites. Second, algorithms for recommender systems suffer from many issues. For example, in order to measure item similarity, content-based methods rely on explicit item descriptions. However, such descriptions may be difficult to obtain for items like ideas or opinions. Collaborative filtering has the data sparsity problem and the cold-start problem . In contrast to the huge number of items in recommender systems, each user normally only rates a few. Therefore, the user/item rating matrix is typically very sparse. It is difficult for recommender systems to accurately measure user similarities from those limited number of reviews. A related problem is the cold-start problem. Even for a system that is not particularly sparse, when a user initially joins, the system has none or perhaps only a few reviews from this user. Therefore, the system...