Strong Matchers, Weak Architecture: Towards Confidence Guided Global Consistency in LLM-based Record Linkage
Resumo
Record Linkage produces transitive inconsistencies that existing LLM-based methods address locally or by construction, but none optimizes at partition level. We investigate whether LLM confidence scores carry sufficient signal to drive global partition optimization in clean-clean multi-source settings. On MusicBrainz, scores discriminate true from false matches among predicted matches (mean 0.936 vs. 0.729). A global ILP exploiting scores as edge weights resolves all 42 inconsistent components by removing 52 edges (98.1% true false positives) fewer than greedy alternatives and with an optimality guarantee. These results motivate LLM confidence scores as input to partition-level optimization, with efficient approximations as future work.
Palavras-chave:
Record Linkage, Entity Resolution, Large Language Models, Correlation Clustering, Global Consistency
Referências
Bansal, N.; Blum, A.; Chawla, S. Correlation clustering. In: IEEE SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE, 43., 2002. Proceedings [...]. [S.l.: s.n.], 2002. p. 238–247.
Binette, O.; Steorts, R. (Almost) All of Entity Resolution. Science Advances, v. 8, 2022.
Cao, H. et al. Graph Deep Active Learning Framework for Data Deduplication. Big Data Mining and Analytics, v. 7, n. 3, p. 753–764, 2024.
Carvalho, G. H. de et al. Ice-ID: A Novel Historical Census Data Benchmark Comparing NARS against LLMs & A ML Ensemble on Longitudinal Identity Resolution. arXiv preprint, 2025.
Christen, V. et al. Graph Metrics-driven Record Cluster Repair meets LLM-based active learning. Journal of Data and Information Quality, v. 17, n. 2, 2025.
Demaine, E. D. et al. Correlation Clustering in general weighted graphs. Theoretical Computer Science, v. 361, n. 2, p. 172–187, 2006.
Draisbach, U.; Christen, P.; Naumann, F. Transforming Pairwise Duplicates to Entity Clusters for High-quality Duplicate Detection. Journal of Data and Information Quality, v. 12, n. 1, 2019.
Fu, J. et al. In-context Clustering-based Entity Resolution with Large Language Models: A Design Space Exploration. Proceedings of the ACM on Management of Data, v. 3, n. 4, 2025.
Huang, K. et al. ThriftLLM: On Cost-Effective Selection of Large Language Models for Classification Queries. Proceedings of the VLDB Endowment, v. 18, n. 11, p. 4410–4423, 2025.
Li, H. et al. On leveraging Large Language Models for Enhancing Entity Resolution. arXiv preprint, 2024.
Li, Y. et al. Deep Entity Matching with Pre-Trained Language Models. Proceedings of the VLDB Endowment, v. 14, n. 1, p. 50–60, 2020.
Nikoletos, K.; Ioannou, E.; Papadakis, G. The Five Generations of Entity Resolution on Web Data. In: STEFANIDIS, K. et al. (ed.). Web Engineering. Cham: Springer, 2024. p. 469–473.
Okayama, K.; Ito, H.; Morishima, A. Learning from Unknown-Unknowns: Inconsistency-Driven Sampling for Improving LLM Entity Matching. Information Research, 2026.
Pardo, F. de M. et al. GraLMatch: Matching Groups of Entities with Graphs and Language Models. In: SIMITSIS, A. et al. (ed.). Proceedings of the 28th International Conference on Extending Database Technology (EDBT 2025). Barcelona: OpenProceedings.org, 2025. p. 1–12.
Pardo, F. de M. et al. TransClean: Finding False Positives In Multi-Source Entity Matching Under Real-World Conditions Via Transitive Consistency. IEEE Access, v. 13, p. 195856–195870, 2025.
Peeters, R.; Bizer, C. Dual-Objective Fine-Tuning of BERT for Entity Matching.
Proceedings of the VLDB Endowment, v. 14, n. 10, p. 1913–1921, 2021.
Peeters, R.; Bizer, C. Using ChatGPT for Entity Matching. In: EUROPEAN CONFERENCE ON ADVANCES IN DATABASES AND INFORMATION SYSTEMS. [S.l.]: Springer, 2023. p. 221–230.
Saeedi, A.; Peukert, E.; Rahm, E. Comparative Evaluation of Distributed Clustering Schemes for Multi-Source Entity Resolution. In: KIRIKOVA, M.; NØRVÅG, K.; PAPADOPOULOS, G. A. (ed.). Advances in Databases and Information Systems. Cham: Springer, 2017. p. 278–293.
Wang, T. et al. Match, Compare, or Select? An Investigation of Large Language Models for Entity Matching. In: Proceedings of the 31st International Conference on Computational Linguistics. 2025. p. 96–109.
ZEAKIS, A. et al. AvengER: Ensembling and Fine-Tuning LLMs for SELECT prompts in Entity Resolution. In: Extended Semantic Web Conference. 2025.
Binette, O.; Steorts, R. (Almost) All of Entity Resolution. Science Advances, v. 8, 2022.
Cao, H. et al. Graph Deep Active Learning Framework for Data Deduplication. Big Data Mining and Analytics, v. 7, n. 3, p. 753–764, 2024.
Carvalho, G. H. de et al. Ice-ID: A Novel Historical Census Data Benchmark Comparing NARS against LLMs & A ML Ensemble on Longitudinal Identity Resolution. arXiv preprint, 2025.
Christen, V. et al. Graph Metrics-driven Record Cluster Repair meets LLM-based active learning. Journal of Data and Information Quality, v. 17, n. 2, 2025.
Demaine, E. D. et al. Correlation Clustering in general weighted graphs. Theoretical Computer Science, v. 361, n. 2, p. 172–187, 2006.
Draisbach, U.; Christen, P.; Naumann, F. Transforming Pairwise Duplicates to Entity Clusters for High-quality Duplicate Detection. Journal of Data and Information Quality, v. 12, n. 1, 2019.
Fu, J. et al. In-context Clustering-based Entity Resolution with Large Language Models: A Design Space Exploration. Proceedings of the ACM on Management of Data, v. 3, n. 4, 2025.
Huang, K. et al. ThriftLLM: On Cost-Effective Selection of Large Language Models for Classification Queries. Proceedings of the VLDB Endowment, v. 18, n. 11, p. 4410–4423, 2025.
Li, H. et al. On leveraging Large Language Models for Enhancing Entity Resolution. arXiv preprint, 2024.
Li, Y. et al. Deep Entity Matching with Pre-Trained Language Models. Proceedings of the VLDB Endowment, v. 14, n. 1, p. 50–60, 2020.
Nikoletos, K.; Ioannou, E.; Papadakis, G. The Five Generations of Entity Resolution on Web Data. In: STEFANIDIS, K. et al. (ed.). Web Engineering. Cham: Springer, 2024. p. 469–473.
Okayama, K.; Ito, H.; Morishima, A. Learning from Unknown-Unknowns: Inconsistency-Driven Sampling for Improving LLM Entity Matching. Information Research, 2026.
Pardo, F. de M. et al. GraLMatch: Matching Groups of Entities with Graphs and Language Models. In: SIMITSIS, A. et al. (ed.). Proceedings of the 28th International Conference on Extending Database Technology (EDBT 2025). Barcelona: OpenProceedings.org, 2025. p. 1–12.
Pardo, F. de M. et al. TransClean: Finding False Positives In Multi-Source Entity Matching Under Real-World Conditions Via Transitive Consistency. IEEE Access, v. 13, p. 195856–195870, 2025.
Peeters, R.; Bizer, C. Dual-Objective Fine-Tuning of BERT for Entity Matching.
Proceedings of the VLDB Endowment, v. 14, n. 10, p. 1913–1921, 2021.
Peeters, R.; Bizer, C. Using ChatGPT for Entity Matching. In: EUROPEAN CONFERENCE ON ADVANCES IN DATABASES AND INFORMATION SYSTEMS. [S.l.]: Springer, 2023. p. 221–230.
Saeedi, A.; Peukert, E.; Rahm, E. Comparative Evaluation of Distributed Clustering Schemes for Multi-Source Entity Resolution. In: KIRIKOVA, M.; NØRVÅG, K.; PAPADOPOULOS, G. A. (ed.). Advances in Databases and Information Systems. Cham: Springer, 2017. p. 278–293.
Wang, T. et al. Match, Compare, or Select? An Investigation of Large Language Models for Entity Matching. In: Proceedings of the 31st International Conference on Computational Linguistics. 2025. p. 96–109.
ZEAKIS, A. et al. AvengER: Ensembling and Fine-Tuning LLMs for SELECT prompts in Entity Resolution. In: Extended Semantic Web Conference. 2025.
Publicado
08/09/2026
Como Citar
PIRES, Luma C. R.; A. JUNIOR, Jorge R..
Strong Matchers, Weak Architecture: Towards Confidence Guided Global Consistency in LLM-based Record Linkage. In: SIMPÓSIO BRASILEIRO DE BANCO DE DADOS (SBBD), 41. , 2026, São Carlos/SP.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 917-923.
ISSN 2763-8979.
DOI: https://doi.org/10.5753/sbbd.2026.249478.
