An Embarrassingly Simple Model for Dialogue Relation Extraction

Dialogue relation extraction (RE) is to predict the relation type of two entities mentioned in a dialogue. In this paper, we model Dialogue RE as a multi-label classification task and propose a simple yet effective model named SimpleRE. SimpleRE captures the interrelations among multiple relations in a dialogue through a novel input format, BERT Relation Token Sequence (BRS). In BRS, multiple [CLS] tokens are used to capture different relations between different pairs of entities. A Relation Refinement Gate (RRG) is designed to extract relation-specific semantic representation adaptively. Experiments on DialogRE show that SimpleRE achieves the best performance with much shorter training time. SimpleRE outperforms all direct baselines on sentence-level RE without using external resources.