A Conditional Cascade Model for Relational Triple Extraction

TL;DR

This paper introduces a novel conditional cascade model for relational triple extraction that effectively mitigates class imbalance issues, achieving state-of-the-art results on NYT and WebNLG datasets.

Contribution

It proposes a three-step extraction framework and a confidence threshold-based loss to address class imbalance in tagging-based relational triple extraction.

Findings

Achieves state-of-the-art performance on NYT and WebNLG datasets.

Effectively mitigates class imbalance issues in relational triple extraction.

Outperforms existing methods in experimental evaluations.

Abstract

Tagging based methods are one of the mainstream methods in relational triple extraction. However, most of them suffer from the class imbalance issue greatly. Here we propose a novel tagging based model that addresses this issue from following two aspects. First, at the model level, we propose a three-step extraction framework that can reduce the total number of samples greatly, which implicitly decreases the severity of the mentioned issue. Second, at the intra-model level, we propose a confidence threshold based cross entropy loss that can directly neglect some samples in the major classes. We evaluate the proposed model on NYT and WebNLG. Extensive experiments show that it can address the mentioned issue effectively and achieves state-of-the-art results on both datasets. The source code of our model is available at: https://github.com/neukg/ConCasRTE.