Abstract. Recommender systems are widely used to help deal with the problem of information overload. However, recommenders raise serious privacy and security issues. The personal information collected by recommenders raises the risk of unwanted exposure of that information. Also, malicious users can bias or sabotage the recommendations that are provided to other users. This paper raises important research questions in three topics relating to exposure and bias in recommender systems: the value and risks of the preference information shared with a recommender, the effectiveness of shilling attacks designed to bias a recommender, and the issues involved in distributed or peer-to-peer recommenders. The goal of the paper is to bring these questions to the attention of the information and communication security community, to invite their expertise in addressing them. 1

Studies have shown that users have great difficulty specifying their security and privacy policies in a variety of application domains. While machine learning techniques have successfully been used to refine models of user preferences, such as in recommender systems, they are generally configured as “black boxes ” that take control over the entire policy and severely restrict the ways in which the user can manipulate it. This article presents an alternative approach, referred to as user-controllable policy learning. It involves the incremental manipulation of policies in a context where system and user refine a common policy model. The user regularly provides feedback on decisions made based on the current policy. This feedback is used to identify (learn) incremental policy improvements which are presented as suggestions to the user. The user, in turn, can review these suggestions and decide which, if any, to accept. The incremental nature of the suggestions enhances usability, and because the user and the system manipulate a common policy representation, the user retains control and can still make policy modifications by hand. Results obtained using a neighborhood search implementation of this approach are presented in the context of data derived from the deployment of a friend finder application, where users can share their locations with others, subject to privacy policies they refine over time. We present results showing policy accuracy, which averages 60 % upon initial definition by our users, climbing as high as 90 % using our technique.

The Stock Tracker is an adaptive recommendation system for trading stocks that automatically acquires content-based models of user preferences to tailor its buy and sell advice. The system incorporates an e#cient algorithm that exploits the fixed structure of user models and relies on unobtrusive data-gathering techniques. In this paper, we describe our approach to personalized recommendation and its implementation in this domain. We also discuss experiments that evaluate the system's behavior on both human subjects and synthetic users. The results suggest that the Stock Tracker can rapidly adapt its advice to di#erent types of users.

Personalized service is important on the Internet, especially in Web-based learning. Generally, most personalized systems consider learner preferences, interests, and browsing behaviors in providing personalized services. However, learner ability usually is neglected as an important factor in implementing personalization mechanisms. Besides, too many hyperlink structures in Web-based learning systems place a large information burden on learners. Consequently, in Web-based learning, disorientation (losing in hyperspace), cognitive overload, lack of an adaptive mechanism, and information overload are the main research issues. This study proposes a personalized e-learning system based on Item Response Theory (PEL-IRT) which considers both course material difficulty and learner ability to provide individual learning paths for learners. The item characteristic function proposed by Rasch with a single difficulty parameter is used to model the course materials. To obtain more precise estimation of learner ability, the maximum likelihood estimation (MLE) is applied to estimate learner ability based on explicit learner feedback. Moreover, to determine an appropriate level of difficulty parameter for the course materials, this study also proposes a collaborative voting approach for adjusting course material difficulty. Experiment results show that applying Item Response Theory to Web-based learning can achieve personalized learning and help learners to learn more effectively and efficiently.

Predicting the interests of a user in information is an important process in personalized information systems. In this paper, we present a way to create prediction engines that allow prediction techniques to be easily combined into prediction strategies. Prediction strategies choose one or a combination of prediction techniques at the moment a prediction is required, taking into account the most up-to-date knowledge about the current user, other users, the information and the system itself. Results of two experiments show that prediction strategies improve both the accuracy and stability of prediction engines. One of these experiments involves a TV recommender system. This paper describes the method of prediction strategies, how they have been applied in the TV recommender system and results of the experiment in detail.

Abstract. A new approach is outlined in which group modeling techniques are used to model an individual user. This helps to reduce cold-start problems, and allows aggregating multiple criteria. 1 Background Interactive television offers the possibility of personalized viewing experiences. Different domains have been identified in which this personalization would have a great impact, such as education, news, advertising, and electronic program guides [1,2]. As watching television tends to be a social activity, adaptation needs to be to groups rather than individuals. In [3,4] we have explored this issue of group adaptation, in particular group modeling. We have empirically explored how humans select a sequence of items for a group to watch based on data about the individuals’ preferences, and compared their behavior with twelve strategies (inspired by Social Choice Theory). The results show that humans care about fairness and avoiding individual misery. In a second experiment, we have investigated how satisfied people believe they would be with sequences chosen by the different strategies, and how

Abstract. In this paper, we describe our on-going work on applying web mining to guide focused collaborative filtering for paper recommendations in a web-based learning system. In particular, we propose to first apply a data clustering technique on web usage data to form clusters (groups) of users with similar browsing patterns, which can be viewed as filtering based on implicit ratings (browsing sequences) according to [21]. Then, collaborative filtering techniques would be adopted on each cluster, instead of on the whole pool of users for recommendations as in other clustering-based collaborative filtering approaches. By using our two-layered collaborative filtering approach, we will not only maintain the diversity of users, but also focus on groups of users with similar browsing patterns. Therefore, our proposed approach could not only make personalized but also ‘grouplized ’ recommendations, thus overcoming previous claims that data clustering will only produce ‘less-personal recommendations ’ [33]. In addition, both explicit and implicit ratings are taken into consideration, which can reinforce and complement each other to make more accurate recommendations. 1.

In this paper, natural language (nl) dialogue is suggested as interaction technique for personalized EPGs to handle a variety of well-known problems. nl interaction addresses the new-user cold-start problem since the necessary user model can be gathered gracefully, without the high initial user e#ort prominent in traditional recommendation systems. It is argued that nl interaction enhances the user's freedom of expressing her preferences to increase recommendation accuracy. The conversational interaction style also enhances ease- and enjoyment-ofuse.

We examine the problems with automated recommendation systems when information about user preferences is limited. We equate the problem to one of content similarity measurement and apply techniques from Natural Language Processing to the domain of movie recommendation. We describe two algorithms, a nave word-space approach and a more sophisticated approach using topic signatures, and evaluate their performance compared to baseline, gold standard, and commercial systems.

Abstract not found