Enhancing Dialogue-based Relation Extraction by Speaker and Trigger Words Prediction

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4580–4585 August 1–6, 2021. ©2021 Association for Computational Linguistics 4580 Enhancing Dialogue-based Relation Extraction by Speaker and Trigger Words Prediction Tianyang Zhao♦§∗, Zhao Yan§, Yunbo Cao§, Zhoujun Li♦† ♦State Key Lab of Software Development Environment, Beihang University, Beijing, China §Tencent, Beijing, China ♦{tyzhao,lizj}@buaa.edu.cn; §{zhaoyan,yunbocao}@tencent.com Abstract Identifying relations from dialogues is more challenging than traditional sentence-level re- lation extraction (RE), since the difficulties of speaker information representation and the long-range semantic reasoning. Despite the successful efforts, existing methods do not fully consider the particularity of dialogues, making them difficult to truly understand the semantics between conversational arguments. In this paper, we propose two beneficial tasks, speaker prediction and trigger words predic- tion, to enhance the extraction of dialogue- based relations. Specifically, speaker predic- tion captures the characteristics of speaker- related entities, and the trigger words predic- tion provides supportive contexts for relations between arguments. Extensive experiments on the DialogRE dataset show noticeable im- provements compared to the baseline models, which achieves a new state-of-the-art perfor- mance with a 65.5% of F1 score and a 60.5% of F1c score, respectively. 1 Introduction The task of relation extraction is to identify the relation facts between two arguments from plain text, which is the fundamental step of many natural language processing applications. Recent years have seen increasing efforts on sentence-level RE, e.g., relations only hold within a single sentence (Fu et al., 2019; Luan et al., 2019; Zhao et al., 2020; Wang and Lu, 2020; Wei et al., 2020). To adapt to complex scenarios, some current works have moved forward to the document-level RE, e.g., relations can exist across multiple sentences (Yao et al., 2019; Wang et al., 2020; Nan et al., 2020; Jain et al., 2020; Zhou et al., 2021). ∗ Work done during an internship at Tencent Cloud Xi- aowei. †Corresponding Author. S1: Yeah you see umm, well, I’m an actor. Right? …… S1: Mom! S2: Sweetie! So this is where you work? …… Dialogue 1 Dialogue 2 S1: Hello, Mr. Bing. S2: Loved your Stevie Wonder last night. S3: Thanks. Listen, about the weekly numbers, I‘m gonna need them on my desk by nine o'clock. S1: Sure. S2: No problem. Dialogue 3 Trigger Words: - (S1, per:boss, S3) (S2, per:boss, S3) Relation: Relation: (S1, per:title, actor) Trigger Words: - Relation: (S1, per:parents, S2) Trigger Words: mom Figure 1: Examples from the DialogRE dataset. sn denotes the speaker of each utterance. The underlined text indicates the relation between the argument pairs. A more challenging yet practical extension is the dialogue-based relation extraction. The dialogues contain multi-turn conversations among a group of speakers. The relations not only exist between the entities in the dialogue text but also the speakers of each utterance. Additionally, most of relations appear in multi-turn conversation, which require cross-sentence extraction. Considering the com- plexities, we divide the dialogue-based RE into three categories. In the first category, the relation can be directly inferred from the current utterance, as shown in …

Content truncated to 500 words