Instance-based retrieval by analogy

Abstract

This work presents a method for retrieval in knowledge bases expressed in Description Logics, founded in the instancebased learning. The procedure implements the disjunctive version space approach exploiting a notion of semantic difference. The method can be employed both to answer to class-membership queries, even though the answers are not logically entailed by the knowledge base, e.g. there are some inconsistent assertions due to heterogeneous sources. In addition, it may also predict/suggest new assertions The method has been implemented and tested in an experimentation, where we show that it is sound and effective.

Evolutionary Conceptual Clustering of Semantically Annotated Resources

“…To this purpose we are currently investigating the application of a randomized search procedure based on simulated annealing or tabu search. Another promising research line, for extensions to matchmaking, retrieval and classification, is retrieval by analogy [6]: a search query may be issued by means of prototypical resources; answers may be retrieved based on local models (intensional concept descriptions) for the prototype constructed (on the fly) based on the most similar resources (w.r.t. some (dis)similarity measure).…”

“…From preliminary experiments where the measure was employed at solving classification/retrieval problems (as in [6]), we could obtain good results by using the very set of both primitive and defined concepts found in the ontology. However, while redundancy may be beneficial (as discussed also in [17]) using too many features increases the computational effort required for the measure.…”

“…The automation is likely to be assisted by extensions of supervised or unsupervised learning methods that are capable to the specific representations of the field [7,4,6,9,13].…”