Measuring Similarity in Description Logics Using Refinement Operators

Similarity measures over refinement graphs

“…In particular, we will present refinement operators and a refinement graph for feature terms and use them to define similarity measures. However, thanks to the idea of the refinement graph the similarity measures introduced in this paper are, in principle, applicable to any other formalism given a refinement graph that satisfies several requirements that we will identify (a preliminary validation of the ideas presented in this paper to description logics can be found in our previous work Sánchez- Ruiz et al 2011).…”

Refinement-Based Similarity Measure over DL Conjunctive Queries

“…The work presented in this paper extends our previous work on similarity on Description Logics [20], where we studied how to assess similarity between individuals by transforming them to concepts, and then assessing the similarity of these concepts. The approach presented in this paper is more general (since the language DL queries is common to all DL), more efficient, and more accurate (since we might lose information when converting individuals to concepts).…”

“…Like in our previous work on similarity assessment [20], we selected this dataset since it is available in many representation formalisms (Horn clauses, feature terms and description logic), and therefore, we can compare our similarity measure with existing similarity measures in the literature. The dataset consists of 10 trains, 5 of them labelled as "West", and 5 of them labelled as "East."…”

“…The dataset consists of 10 trains, 5 of them labelled as "West", and 5 of them labelled as "East." We compared our similarity measure against 7 others: S DLρ [20], a similarity measure for the EL description logic; González et al [10], a similarity measure for acyclic concepts in description logic; RIBL [9], which is a Horn clause similarity measure; SHAUD [1], which is a similarity measure for feature terms; and S λ , S π , and S wπ [18], which are similarity measures for feature terms but also based on the idea of refinement operators. For RIBL, we used the original version of the trains dataset, for SHAUD, S λ , S π , and S wπ , we used the feature term version Queries for Gp1 and Gp2 Qp1({x1}, {}) = M argherita(x1) Qp2({x2}, {y1, y2}) = P izza(x2) ∧ hasT opping(x2, y1) ∧ Chicken(y1)∧ hasT opping(x2, y2) ∧ V egetable(y2)…”

“…Specifically, we focus in the problem of measuring similarity between individuals. The similarity measure presented in this paper, S Q , works as follows: 1) given two individuals, we convert them into DL Conjunctive Queries, 2) the similarity between the two queries is measured using a refinement-operator-based similarity measure [17,20]. The approach presented in this paper differs from previous work [8,20] in that the S Q similarity is defined over the space of DL Conjunctive Queries, rather than in the space of DL concepts.…”