Sélection de la langue

Gouvernement du Canada / Government of Canada

Recherche

Better entity matching with transformers through ensembles

DOITrouver le DOI : https://doi.org/10.1016/j.knosys.2024.111678
AuteurRechercher : Identifiant ORCID : https://orcid.org/0000-0001-6618-7443; Rechercher : Identifiant ORCID : https://orcid.org/0000-0001-8423-2906; Rechercher : 1Identifiant ORCID : https://orcid.org/0000-0002-3460-6946
Affiliation
  1. Conseil national de recherches Canada. Technologies numériques
Bailleur de fondsRechercher : Natural Sciences and Engineering Research Council of Canada; Rechercher : Canada Research Chairs Program
FormatTexte, Article
Sujetentity resolution; entity matching; deduplication; neural networks; deep learning; transformers
Résumé
In this paper, we introduce AttendEM, a framework for entity matching (EM), i.e., pairwise identification of duplicates across databases. Eschewing the prevalent focus on text cleaning and training data augmentation of other transformers-based EM solutions, AttendEM leverages intra-transformer ensembling of distinctively rearranged text, additional aggregator tokens, and extra self-attention to enhance the base transformer architecture. Against state-of-the-art (SOTA) solutions on the ER-Magellan benchmark datasets, AttendEM achieved higher F1 scores in most cases. These SOTA solutions are Ditto (mean improvement of 0.21% with Ditto’s own reported results, 3.93% with DAEM’s Ditto replication, 2.99% with HierGAT’s Ditto replication), DAEM (0.53%), and HierGAT (0.54%). AttendEM’s improvements are comparable to solutions that claimed to have outperformed Ditto, HierGAT (Yao et al., 2022) (2.46% compared to AttendEM’s 2.99%) and DAEM (Huang et al., 2022) (3.42% compared to AttendEM’s 3.93%), when calculated using results from their respective Ditto replications.
Date de publication
Maison d’éditionElsevier
Dans
  • Knowledge-Based Systems 293, nº C, 111678 (7 juin 2024).
Langueanglais
Publications évaluées par des pairsOui
IdentificateurS0950705124003137
Exporter la noticeExporter en format RIS
Signaler une correctionSignaler une correction (s'ouvre dans un nouvel onglet)
Identificateur de l’enregistrementb1bbe090-0613-4be7-bd90-2bb4c2f1bbd5
Enregistrement créé2024-04-08
Enregistrement modifié2024-04-08

Détails de la page

Par :

Date de modification :