More: A Metric Learning Based Framework for Open-Domain Relation Extraction

Abstract

Open relation extraction (OpenRE) is the task of extracting relation schemes from open-domain corpora. Most existing OpenRE methods either do not fully benefit from high-quality labeled corpora or can not learn semantic representation directly, affecting downstream clustering efficiency. To address these problems, in this work, we propose a novel learning framework named MORE (Metric learning-based Open Relation Extraction). The framework utilizes deep metric learning to obtain rich supervision signals from labeled data and drive the neural model to learn semantic relational representation directly. Experiments result in two real-world datasets show that our method outperforms other state-of-theart baselines. Our source code is available on Github 1 .

Improving Unsupervised Relation Extraction by Augmenting Diverse Sentence Pairs

“…The aforementioned performance gains show that the proposed relation representation learning (Hu et al, 2020) 38.5±2.2 44.7±1.8 41.4±1.9 37.8±2.4 43.3±1.9 40.4±1.7 35.0±2.0 Parameters of SDA MORE (Wang et al, 2021) 43.8±1.9 40.3±2.0 42.0±2.2 40.8±2.2 43.1±2.4 41.9±2.1 35.6±2.1 None OHRE 32.7±1.8 60.7±2.3 42.5±1.9 34.8±2.1 53.9±2.5 42.3±1.8 33.6±1.8 None EIURE 48.4±1.9 38.8±1.8 43.1±1.8 37.7±1.5 49.2±1.6 42.7±1.6 34.5±1.4 WikiData, T5, WebNLG HiURE w. K-Means (Liu et al, 2022) It is worth mentioning that despite utilizing K-Means as the clustering algorithm, our model's performance surpasses that of both the SelfORE and OHRE, which employ more advanced clustering techniques. We believe that improving the clustering algorithm would yield advantages to the overall URE performance, and we leave it for future exploration.…”

“…However, the frequently re-initialized linear classification layers can hinder the process of representation learning. Wang et al (2021) are the first to utilize deep metric learning in the OpenRE task and this scheme demonstrates good capabilities in representation learning.…”

“…2) SelfORE (Hu et al, 2020) leverages weak, selfsupervised signals from BERT for adaptive clustering and bootstraps these signals by improving contextualized features in relation classification. 3) MORE (Wang et al, 2021) utilizes deep metric learning to extract rich supervision signals from labeled data and drive the neural model to learn relational representations directly. 4) OHRE effectively integrates hierarchy information into relation representations via a dynamic hierarchical triplet objective and hierarchical curriculum training paradigm.…”