TDEER: An Efficient Translating Decoding Schema for Joint Extraction of Entities and Relations

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8055–8064 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 8055 TDEER: An Efficient Translating Decoding Schema for Joint Extraction of Entities and Relations Xianming Li, Xiaotian Luo, Chenghao Dong ∗, Daichuan Yang, Beidi Luan, Zhen He Ant Group, Shanghai {niming.lxm,lxt267638,dch166451,beidi.lbd}@antgroup.com {daichuan.ydc,fuyi.hz}@antgroup.com Abstract Joint extraction of entities and relations from unstructured texts to form factual triples is a fundamental task of constructing a Knowl- edge Base (KB). A common method is to de- code triples by predicting entity pairs to ob- tain the corresponding relation. However, it is still challenging to handle this task effi- ciently, especially for the overlapping triple problem. To address such a problem, this pa- per proposes a novel efficient entities and rela- tions extraction model called TDEER, which stands for Translating Decoding Schema for Joint Extraction of Entities and Relations. Un- like the common approaches, the proposed translating decoding schema regards the rela- tion as a translating operation from subject to objects, i.e., TDEER decodes triples as subject + relation → objects. TDEER can naturally handle the overlapping triple prob- lem, because the translating decoding schema can recognize all possible triples, including overlapping and non-overlapping triples. To enhance model robustness, we introduce neg- ative samples to alleviate error accumulation at different stages. Extensive experiments on public datasets demonstrate that TDEER produces competitive results compared with the state-of-the-art (SOTA) baselines. Fur- thermore, the computation complexity anal- ysis indicates that TDEER is more efficient than powerful baselines. Especially, the pro- posed TDEER is 2 times faster than the re- cent SOTA models. The code is available at https://github.com/4AI/TDEER. 1 Introduction Extraction of entities and relations from unstruc- tured texts is one of the most essential informa- tion extraction tasks. It aims to extract entities and their corresponding semantic relations from unstructured texts, which are usually presented in a ∗Corresponding author. Normal Shanghai is located in China. EPO Jackie Chan starred in his directed film Police Story. SEO Jackie R. Brown was born in Washinton, the capital of the United States. Shanghai China country Jackie Chan Police Story star in direct movie Jackie R. Brown Washinton birth place capital of the United States Figure 1: Cases of Normal, EntityPairOverlap (EPO), and SingleEntityOverlap (SEO) overlapping triples. triple form of (subject, relation, object), e.g., (Mi- crosoft, co-founder, Bill Gates). It is also a crucial step in building a large-scale KB and exerts an important role in the development of web search (Szumlanski and Gomez, 2010), question answer- ing (Fader et al., 2014), biomedical text mining (Huang and Lu, 2016), etc. Traditional approaches (Zelenko et al., 2003; Chan and Roth, 2011; Rink and Harabagiu, 2010) handle this task in a pipeline manner, i.e., extract- ing the entities first and then identifying their re- lations. The pipeline framework simplifies the ex- traction task, but it ignores the relevance between entity identification and relation prediction. To ad- dress this problem, several joint learning models have been proposed and can be categorized …

Content truncated to 500 words