A Small Survey on Preference Languages and Preference Mining Techniques

The research literature on preference modeling, reasoning and eliciting is extensive. Different specification approaches have been proposed to model user preferences such as graphs, logical formalisms and SQL-like constructors. The approach of CP-Nets [2] uses a very simple graphical model which captures users qualitative conditional preference over tuples, under a ceteris paribus semantics. The approach of TCP-Nets [3] generalizes the CP-Nets by introducing the ability of expressing absolute and relative importance of object attributes in the graphical model. In [19] a simple logical framework is introduced for expressing conditional preference statements (cp-rules). It generalizes both the CP-Net and TCP-Net approaches, by relaxing the ceteris paribus semantics and allowing more expressiveness. This formalism is used in [7] in order to extend the SQL query language with preference support. In [5] a simple logical framework is proposed for expressing preferences by means of preference formulae. These formulae are incorporated into the Relational Algebra and into SQL through the operator winnow parameterized by a preference formula. This approach is very expressive and generalizes the CP-Nets, TCP-Nets and the cp-rules of [19]. From a more operational standpoint, [14] proposes to inductively construct complex preferences by means of various basic and intuitive preference construc-tors. These preference constructors are used to extend the SQL query language with preference support [15]. Unlike the previous approaches, the formalism of [14] does not allow to express contextual preferences. In the present paper, we follow the approach of [1] for expressing contextual preference rules. In this work, the formalism i + ≻ i − | X for specifying contextual preference rules has been introduced and a method for incorporating the preference rules into the query-answering mechanism of a database system has been proposed. The preference rules are supposed to be available and the problem of how the rules are obtained is out of the scope of that paper. For a comprehensive survey on preference modeling and query languages with preference support see [18]. In what follows we focus on the literature related to the topic of preference elicitation. Preference learning can be divided into two distinct problems: label ranking and object ranking. Label ranking aims at predicting an order over a set of labels for a given object [8, 11]. One can view the objects as tuples containing information about users: age, education, economic status, city, profession, etc. And the labels represents objects to rank. The problem of label ranking …