DomRec: Investigating Domain-centric Recommendation and Analysis of Entity Linking Methods

ACL ARR 2024 December Submission1342 Authors

Abstract:

Detecting textual mentions and linking them to corresponding entities in a knowledge base is an essential task performed by a variety of existing entity linking approaches.
This paper investigates the relationship between domains and system performance for 12 state-of-the-art annotators using 6 common datasets, arguing performance based on domain using learned topic vectors and machine learning models.
By analysing domain-specific characteristics across domains and methods, we demonstrate that no single technique excels across all domains, and that performance can be significantly enhanced by selecting the most suitable system for each context.
Our findings underline the importance of domain awareness in the development and deployment of text-processing systems, providing a pathway for more adaptable and robust methodologies. We release and open source all generated data, code and findings on our repository and on Zenodo.

Paper Type: Long

Research Area: Information Retrieval and Text Mining

Research Area Keywords: entity linking, topic modeling, analysis

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Publicly available software and/or pre-trained models, Data resources, Data analysis

Languages Studied: English

Submission Number: 1342