Editorial: Preference learning and ranking

Abstract

Informally, preferences are indications of the value of a given item in a certain context. Such valuations can occur in the form of a preference function, which assigns a utility value to the object, or in the form of a preference relation, which compares two different objects. Methods for learning and predicting preferences in an automatic way are among the very recent research topics in machine learning and related areas, such as recommender systems (Ricci et al. 2011; Jannach et al. 2010) and information retrieval (Liu 2011). In fact, it is fair to say that the topic of preference learning and ranking has established itself as a new subfield of machine learning in recent years—a development that is witnessed, amongst others, by a continuously growing number of publications on this topic as well as the organization of dedicated events. Recent workshops in this area include three workshops on Preference Learning at the ECML/PKDD conferences 2008–2010, the Yahoo! Learning to Rank Challenge at ICML 2010, a workshop on Choice Models and Preference Learning at NIPS 2011, and a workshop on Preference Learning: Problems and Applications in Artificial Intelligence at ECAI 2012. Tutorials on this subject have been held at conferences like SIGIR 2008, WWW 2009, ACML 2010, ECML/PKDD 2010, Discovery Science 2011, or ECAI 2012.1 Despite being grounded in machine learning, preference learning is an interdisciplinary field with close connections to other disciplines, including statistics, social choice theory, multiple criteria decision making, decision under risk and uncertainty, and operations research (cf. Fig. 1). In all these areas, considerable progress has been made on preference representation and the automated learning of preference models; recent overviews in these