Individualized extreme dominance (IndED) : a new preference- based method for multi-objective recommender systems.

Abstract

Recommender Systems (RSs) make personalized suggestions of relevant items to users. However, the concept of relevance may involve different quality aspects (objectives), such as accuracy, novelty, and diversity. In addition, users may have their own expectations regarding what characterizes a good recommendation. More specifically, individual users may wish to prioritize the multiple objectives in different proportions based on their preferences. Previous studies on Multi-Objective (MO) recommendation do not prioritize objectives according to the individual users’ preferences systematically or are biased towards a single objective as in re-ranking strategies. Moreover, traditional preference-based multi-objective solutions do not address the specificities of RSs. In this work, we pro- pose IndED (Individualized Extreme Dominance), a new preference-based method for MORSs. IndED explores the concepts of Extreme Dominance and Statistical Significance Tests in order to define a new Pareto-based dominance relation that guides the optimization search considering users’ preferences. We also consider a new decision making process that minimizes the distance to the individual user’s preferences. Experiments show that IndED outperformed competitive baselines, obtaining results closer to the users’ preferences and better balancing the objectives trade-offs. IndED is also the method that obtains the best performance regarding the most difficult objective in each considered scenario.