Session: Multi-Dimensional Recommendation, Context-Awareness and Group Recommendation

Chair: Francesco Ricci
Date: Tuesday, October 25, 10:45-12:30

• Multi-criteria service recommendation based on user criteria preferences

  by Liwei Liu, Nikolay Mehandjiev, Dong-Ling Xu

  Research in recommender systems is now starting to recognise the importance of multiple selection criteria to improve the recommendation output. In this paper, we present a novel approach to multi-criteria recommendation, based on the idea of clustering users in “preference lattices” (partial orders) according to their criteria preferences. We assume that some selection criteria for an item (product or a service) will dominate the overall ranking, and that these dominant criteria will be different for different users. Following this assumption, we cluster users based on their criteria preferences, creating a “preference lattice”. The recommendation output for a user is then based on ratings by other users from the same or close clusters. Having introduced the general approach of clustering, we proceed to formulate three alternative recommendation methods instantiating the approach: (a) using the aggregation function of the criteria, (b) using the overall item ratings, and (c) combining clustering with collaborative filtering. We then evaluate the accuracy of the three methods using a set of experiments on a service ranking dataset, and compare them with a conventional collaborative filtering approach extended to cover multiple criteria. The results indicate that our third method, which combines clustering and extended collaborative filtering, produces the highest accuracy.

  Details

• The effect of context-aware recommendations on customer purchasing behavior and trust

  by Michele Gorgoglione, Umberto Panniello, Alexander Tuzhilin

  Despite the growing popularity of Context-Aware Recommender Systems (CARSs), only limited work has been done on how contextual recommendations affect the behavior of customers in real-life settings. In this paper, we study the effects of contextual recommendations on the purchasing behavior of customers and their trust in the provided recommendations. In particular, we did live controlled experiments with real customers of a major commercial Italian retailer in which we compared the customers’ purchasing behavior and measured their trust in the provided recommendations across the contextual, content-based and random recommendations. As a part of this study, we have investigated the role of accuracy and diversity of recommendations on customers’ behavior and their trust in the provided recommendations for the three types of RSes. We have demonstrated that the context-aware RS outperformed the other two RSes in terms of accuracy, trust and other economics-based performance metrics across most of our experimental settings.

  Details

• Random walk based entity ranking on graph for multidimensional recommendation

  by Sangkeun Lee, Sang-il Song, Minsuk Kahng, Dongjoo Lee, Sang-goo Lee

  In many applications, flexibility of recommendation, which is the capability of handling multiple dimensions and various recommendation types, is very important. In this paper, we focus on the flexibility of recommendation and propose a graph-based multidimensional recommendation method. We consider the problem as an entity ranking problem on the graph which is constructed using an implicit feedback dataset (e.g. music listening log), and we adapt Personalized PageRank algorithm to rank entities according to a given query that is represented as a set of entities in the graph. Our model has advantages in that not only can it support the flexibility, but also it can take advantage of exploiting indirect relationships in the graph so that it can perform competitively with the other existing recommendation methods without suffering from the sparsity problem.

  Details

• Group recommendation using feature space representing behavioral tendency and power balance among members

  by Shunichi Seko, Takashi Yagi, Manabu Motegi, Shinyo Muto

  This paper proposes an algorithm to estimate appropriate or novel content for groups of people who know each other such as friends, couples, and families. To achieve high recommendation accuracy, we focus on “Groupality”, the entity or entities that characterize groups such as the tendency of content selection and the relationships among group members. Our algorithm calculates recommendation scores using a feature space that consists of the behavioral tendency of a group and the power balance among group members based on individual preference and the behavioral history of group. After gathering the behavioral history of subject groups when watching TV, we verify that our proposed algorithm can recommend appropriate content, and find novel content. Evaluations show that our proposal achieves higher performance than existing methods.

  Details

Back to Program