Towards a More Generalized Approach in Open Relation Extraction
Qing Wang, Yuepei Li, Qiao Qiao, Kang Zhou, Qi Li
TL;DR
This work addresses OpenRE under a generalized setting where unlabeled data mix contains both known and novel relations. It introduces MixORE, a two-phase framework combining novel relation detection with open-world semi-supervised continual learning to jointly classify known relations and cluster novel ones. The approach leverages a relation encoder, Semantic Autoencoder for NRD, Gaussian mixture clustering for novel groups, and a contrastive-leaning, exemplar-based joint objective to refine representations. Empirical results on FewRel, TACRED, and Re-TACRED show that MixORE improves known-relation classification while achieving competitive or superior novel-relation clustering, highlighting its practical applicability to real-world OpenRE tasks. The work advances OpenRE by relaxing restrictive assumptions and providing a scalable, adaptable pipeline for mixed unlabeled data.
Abstract
Open Relation Extraction (OpenRE) seeks to identify and extract novel relational facts between named entities from unlabeled data without pre-defined relation schemas. Traditional OpenRE methods typically assume that the unlabeled data consists solely of novel relations or is pre-divided into known and novel instances. However, in real-world scenarios, novel relations are arbitrarily distributed. In this paper, we propose a generalized OpenRE setting that considers unlabeled data as a mixture of both known and novel instances. To address this, we propose MixORE, a two-phase framework that integrates relation classification and clustering to jointly learn known and novel relations. Experiments on three benchmark datasets demonstrate that MixORE consistently outperforms competitive baselines in known relation classification and novel relation clustering. Our findings contribute to the advancement of generalized OpenRE research and real-world applications.