[Paper Review] More Data, More Relations, More Context and More Openness: A Review and Outlook for Relation Extraction

Xu Han, Tianyu Gao+8 more|arXiv (Cornell University)|Apr 7, 2020

Topic Modeling129 references41 citations

TL;DR

A comprehensive review of relation extraction (RE) methods, outlining four directions for advancement—more data, more efficient learning, more complex context, and more openness—along with challenges and future research directions.

ABSTRACT

Relational facts are an important component of human knowledge, which are hidden in vast amounts of text. In order to extract these facts from text, people have been working on relation extraction (RE) for years. From early pattern matching to current neural networks, existing RE methods have achieved significant progress. Yet with explosion of Web text and emergence of new relations, human knowledge is increasing drastically, and we thus require "more" from RE: a more powerful RE system that can robustly utilize more data, efficiently learn more relations, easily handle more complicated context, and flexibly generalize to more open domains. In this paper, we look back at existing RE methods, analyze key challenges we are facing nowadays, and show promising directions towards more powerful RE. We hope our view can advance this field and inspire more efforts in the community.

Motivation & Objective

• Survey the evolution of relation extraction from pattern-based and statistical methods to neural models.
• Identify core challenges in real-world RE such as data quality, long-tail relations, contextual complexity, and open-domain growth.
• Propose four future directions to enhance RE: more data, efficient learning, complex context handling, and open-domain openness.

Proposed method

• Review existing RE literature across pattern-based, statistical, and neural approaches.
• Summarize datasets, benchmarks, and evaluation trends in supervised, distant supervision, few-shot, and document-level RE.
• Analyze four forward-looking directions with supporting experiments and dataset analyses.
• Highlight two key challenges with empirical evidence and open-source toolkits for replication (OpenNRE).

Experimental results

Research questions

• RQ1 What are the main limitations of current RE methods in real-world scenarios?
• RQ2 How can RE leverage more data, efficient learning, complex context, and open domains to improve robustness and scalability?
• RQ3 What challenges arise when distantly supervised or few-shot or document-level RE are scaled to open domains?
• RQ4 What public datasets and tools best illuminate these directions and challenges?

Key findings

• Neural RE models dramatically improved performance over earlier methods, but still struggle with real-world complexity.
• Distant supervision introduces label noise, and mitigating this noise is crucial for robust extraction.
• Few-shot and long-tail relations require efficient learning and domain transfer capabilities beyond standard supervised RE.
• Document-level context and cross-sentence reasoning are essential for many facts but remain underdeveloped.
• Open information extraction and relation discovery offer pathways for open-domain RE, yet canonicalization and NOTA handling are unresolved.
• Entity names and text both contribute significantly to RE performance, with some cases where names alone outperform text in current models.