Variable-Strength Conditional Preferences for Ranking Objects in Ontologies

Abstract

Abstract. We introduce conditional preference bases as a means for ranking objects in ontologies. Conditional preference bases consist of a description logic knowledge base and a finite set of variable-strength conditional preferences. They are inspired by Goldszmidt and Pearl's approach to default reasoning from conditional knowledge bases in System Z + . We define a notion of consistency for conditional preference bases, and show how consistent conditional preference bases can be used for ranking objects in ontologies. We also provide algorithms for computing the rankings. To give evidence of the usefulness of this approach in practice, we describe an application in the area of literature search.

Variable-Strength Conditional Preferences for Ranking Objects in Ontologies

“…A conditional preference base PB = (T, A, P ) is flat iff its z-partition is given by (P ) and thus consists only of one component. Algorithm flatten in a companion paper [30] transforms a consistent conditional preference base PB = (T, A, P ) into an equivalent flat conditional preference base, denoted PB = (T, A, P ). We are now ready to define the object rankings κ sum and κ lex .…”

“…There are several computational tasks related to conditional preference bases PB = (T, A, P ). First, deciding the consistency of PB is done by algorithm consistency in [30] (which returns the z-partition of PB , if PB is consistent, and nil, otherwise), which generalizes an algorithm for deciding ε-consistency in default reasoning [21]. The extended algorithm is essentially based on O(|P | 2 ) tests whether a description logic knowledge base is satisfiable.…”

“…The extended algorithm is essentially based on O(|P | 2 ) tests whether a description logic knowledge base is satisfiable. Second, rewriting PB to an equivalent flat conditional preference base PB is done by algorithm flatten in a companion paper [30], which is similar to a rewriting algorithm in fuzzy default reasoning [16], and which requires O(|P | 2 ) description logic satisfiability tests. Finally, computing the ranking functions κ sum and κ lex is done by algorithms sumand lex-ranking in Figs.…”

“…We present a formalism for ranking objects in description logics that allows for expressing such conditional preferences in models of instances. In a companion paper [30], we present a generalization of this formalism for matchmaking in description logics.…”

On various forms of bipolarity in flexible querying

“…Preference and flexible queries have attracted a considerable interest in different circles and at different epochs among databases researchers [37,32,35,6,7,21,15,31,14,34,40,41,36]. There may exist slightly different motivations for using flexible queries.…”