arXiv Science⌕ Search

arXiv subjects

Andrei Olar

Publications and source records attributed to Andrei Olar.

2 recordsLinked to original sources

Beyond Symmetry: Asymmetric Matching with Transformers for Collective Entity Resolution

A core assumption of entity resolution (ER) is its symmetric nature, which rests on a principle so self-evident it borders on tautology: if a refers to the same object as b, then b must also refer to the same object as a. However, this global property of symmetry does not necessarily extend to each of the complex decision processes within ER systems. In cases involving vague or ambiguous information, enforcing symmetry can be detrimental to performance, an issue exemplified by the ``bad triplet problem''. This paper investigates the effects of input-order asymmetry within a collective ER framework that integrates distinct matching and clustering stages. The input order-aware matching stage utilizes pre-trained transformers to explicitly model and exploit asymmetry, making it possible to evaluate its effect on clustering performance across multiple benchmark datasets. Our results demonstrate that asymmetry is not only measurable but also introduces a significant skew in the final entity clusters, highlighting the need to re-evaluate supervised training methodologies for matchers, particularly the development of direction-aware training paradigms that can preserve, rather than neutralize, the inherent asymmetry in matching tasks.

cs.DB↗

Resolvi: A Reference Architecture for Extensible, Scalable and Interoperable Entity Resolution

Context: Entity resolution (ER) plays a pivotal role in data management by determining whether multiple records correspond to the same real-world entity. Because of its critical importance across domains such as healthcare, finance, and machine learning and its long research history designing and implementing ER systems remains challenging in practice due to the wide array of methodologies and tools available. This diversity results in a paradox of choice for practitioners, which is further compounded by the various ER variants (record linkage, entity alignment, merge/purge, a.s.o). Objective: This paper introduces Resolvi, a reference architecture for facilitating the design of ER systems. The goal is to facilitate creating extensible, interoperable and scalable ER systems and to reduce architectural decision-making duration. Methods: Software design techniques such as the 4+1 view model or visual communication tools such as UML are used to present the reference architecture in a structured way. Source code analysis and literature review are used to derive the main elements of the reference architecture. Results: This paper identifies generic requirements and architectural qualities of ER systems. It provides design guidelines, patterns, and recommendations for creating extensible, scalable, and interoperable ER systems. Furthermore, it highlights implementation best practices and deployment strategies based on insights from existing systems. Conclusion: The proposed reference architecture offers a foundational blueprint for researchers and practitioners in developing extensible, interoperable, and scalable ER systems. Resolvi provides clear abstractions and design recommendations which simplify architecture decision making, whether designing new ER systems or improving existing designs.

cs.DB↗