Work

Universal entity linking

Year: 2025

Type: article

Abstract: Entity linking is a process of connecting mentions of entities in a document to corresponding entries in a knowledge base. Traditional models for entity linking often require specific fine-tuning to work with knowledge bases other than those they were originally pretrained on, which limits their flexibility and scalability. Building on the concept of entity profile generation, we propose a novel approach that enables entity linking across various knowledge bases without the need for such fine-tuning. Our pipeline leverages a fine-tuned Large Language Model, a generic embedding model, and a vector store to achieve high precision on the TweekiGold and Reuters-128 datasets. Additionally, it demonstrates strong retrieval rates across the TweekiGold , Reuters-128 , and ISTEX-1000 Wikidata entity linking datasets. We also illustrate the applicability of our method to other knowledge bases, using the Agrovoc knowledge base as an example. This solution offers a more versatile and scalable approach to entity linking.

Source: Engineering Applications of Artificial Intelligence

DOI: 10.1016/j.engappai.2025.112185

Authors: Adam Aron Rynkiewicz, Raul Palma, Piotr Formanowicz

Institutions: Poznan Supercomputing and Networking Center, Poznań University of Technology

Language: English

Topic: Topic Modeling

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $3,200

Funder: Politechnika Poznańska