NLPeople at SemEval-2023 Task 2: A Staged Approach for Multilingual Named Entity Recognition

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 114–120 July 13-14, 2023 ©2023 Association for Computational Linguistics Samsung Research China - Beijing at SemEval-2023 Task 2: An AL-R Model for Multilingual Complex Named Entity Recognition Haojie Zhang�, Xiao Li, Renhua Gu Xiaoyan Qu, Xiangfeng Meng, Shuo Hu, Song Liu Samsung Research China-Beijing (SRC-B) {tayee.chang,xiao013.li,renhua.gu,s0101.liu}@samsung.com Abstract This paper describes our system for SemEval- 2023 Task 2 Multilingual Complex Named En- tity Recognition (MultiCoNER II). Our team Samsung Research China - Beijing proposes an AL-R (Adjustable Loss RoBERTa) model to boost the performance of recognizing short and complex entities with the challenges of long- tail data distribution, out of knowledge base and noise scenarios. We first employ an adjustable dice loss optimization objective to overcome the issue of long-tail data distribution, which is also proved to be noise-robusted, especially in combatting the issue of fine-grained label con- fusing. Besides, we develop our own knowl- edge enhancement tool to provide related con- texts for the short context setting and address the issue of out of knowledge base. Experi- ments have verified the validation of our ap- proaches. In the official test result, our system ranked 2nd on the English track in this task. 1 Introduction Named Entity recognition is an important task in natural language processing. In low-context data(Malmasi et al., 2022a), semantic ambiguity, complex entities, multilingual and code-mixed set- tings make this task even more difficult. Recog- nizing complex entities in low-context situations was recently outlined by Meng et al. (2021). Mul- tilingual and code-mixed settings were extended through Fetahu et al. (2021). To address these issues, SemEval-2022 task 11 Multilingual Complex Named Entity Recogni- tion(Malmasi et al., 2022b) received 34 system papers, the best of which were able to accurately recognise complex entities by incorporating exter- nal knowledge related to the data(Verlinden et al., 2021; Wang et al., 2022). However, when the entity recognition scenario is more complex, the top sys- tem (Wang et al., 2022) has drawbacks that make it difficult to cope. 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 Sorted Label Index 500 1000 1500 2000 2500 3000 3500 Number of Entities Figure 1: The long-tailed distribution of entity label for English dataset. SemEval-2023 task 2 MultiCoNER II Multilin- gual Complex Named Entity Recognition (Fetahu et al., 2023b) aims to build a system to address the following problems: 1). The entities are com- plex and long-tailed illustrated by Figure 1. 2). The entities are out of knowledge-base entities. 3). The noisy scenarios like the presence of spelling mistakes and typos. In this paper, we build an AL-R (Adjustable Loss RoBERTa ) based system for MultiCoNER II task. Our system utilizes the RoBERTa-large (Liu et al., 2019) as the pre-trained model and fine tunes it on the English dataset (Fetahu et al., 2023a). For the long-tailed dataset and noise scene, we add the adjustable dice loss (Li et al., 2019b) to overcome the issue of long …

Content truncated to 500 words