Communication Dans Un Congrès Année : 2025

Abstracting Entity Matching for Analysing and Explaining Identity and Difference Decision and Indecision

Résumé

Knowledge graphs (KGs) are crucial for applications like search engines, conversational agents, and recommender systems because they provide structured, semantically rich representations of knowledge, enabling more accurate information retrieval, contextual understanding, and intelligent decision-making. Entity matching approaches aim to find identity relations between different entity descriptions. In KGs, owl:sameAs is the standard predicate that represents the identity relation and its expresses the strict identity as defined by Libniz Identity Relation [2]. However, in many cases, a strict identity relation between entities in a knowledge graph is impractical, as some entities have more complex relationships that do not conform to a rigid one-to-one correspondence with real-world objects. In this work we investigate the notions of weak identity and weak difference while ensuring an explanation of the decision that can be taken. The developed approach is able to assign each individual pair to a range in the decision space that we define in this paper. Thanks to the computation of a holistic abstraction of the entity matching process, we ensure providing an explanation for each assignment in terms of sets of incomplete and/or conflicting properties. A first experimental evaluation on a benchmark of DBepdia and YAGO entities has provided promising results.

Fichier principal
Vignette du fichier
KES2025(5).pdf (412.04 Ko) Télécharger le fichier

Dates et versions

hal-05324375 , version 1 (21-10-2025)

Licence

Identifiants

Citer

Olivier Inizan, Fatiha Saïs, Anne Goelzer, Joe Raad, Danai Symeonidou. Abstracting Entity Matching for Analysing and Explaining Identity and Difference Decision and Indecision. KES, Sep 2025, Osaka, Japan. ⟨10.1016/j.procs.2025.09.333⟩. ⟨hal-05324375⟩
2390 Consultations
218 Téléchargements

Altmetric

Partager

  • More