A systematic review of relation extraction task since the emergence of Transformers

TL;DR

This systematic review analyzes RE research since Transformer-based models emerged, synthesizing 34 surveys, 64 datasets, and 104 models from 2019–2024 using the SciLEx framework to ensure reproducibility and open access. It provides a fine-grained, multi-dimensional annotation of literature and benchmarks, revealing rapid transformer-driven progress, a shift toward end-to-end generation, and growing attention to knowledge graphs and Semantic Web integrations. Key findings include substantial dataset development (including encyclopedic, news, and biology domains), diverse architectural families (encoder-based vs Seq2Seq), and persistent challenges in multilingual coverage, data quality, and computational cost. The authors advocate for knowledge-grounded, ontology-driven RE, explore prompt-based and PEFT techniques, and emphasize scalable, interpretable RE in knowledge-intensive applications as future directions.

Abstract

This article presents a systematic review of relation extraction (RE) research since the advent of Transformer-based models. Using an automated framework to collect and annotate publications, we analyze 34 surveys, 64 datasets, and 104 models published between 2019 and 2024. The review highlights methodological advances, benchmark resources, and the integration of semantic web technologies. By consolidating results across multiple dimensions, the study identifies current trends, limitations, and open challenges, offering researchers and practitioners a comprehensive reference for understanding the evolution and future directions of RE.

Figures (32)

• Figure 1: Extended PRISMA flowchart: detailed process from the initial collection to the final selection and annotation
• Figure 2: Number of annotation dimensions by Fleiss Kappa agreement level
• Figure 3: Nb. of surveys published by year
• Figure 4: Publication types of surveys
• Figure 5: Referencing of surveys