NetEase.AI at SemEval-2023 Task 2: Enhancing Complex Named Entities Recognition in Noisy Scenarios via Text Error Correction and External Knowledge

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 897–904 July 13-14, 2023 ©2023 Association for Computational Linguistics NetEase.AI at SemEval-2023 Task 2: Enhancing Complex Named Entities Recognition in Noisy Scenarios via Text Error Correction and External Knowledge Ruixuan Lu1†, Zihang Tang1‡, Guanglong Hu2†, Dong Liu2†, Jiacheng Li*† †NetEase Digital Sail, NetEase Inc. ‡Anhui Province Key Laboratory of Big Data Analysis and Application, University of Science and Technology of China {luruixuan,hzhuguanglong,hzliudong,lijiacheng01}@corp.netease.com zihangt@mail.ustc.edu.cn Abstract Complex named entities (NE), like the titles of creative works, are not simple nouns and pose challenges for NER systems. In the Se- mEval 2023, Task 2: MultiCoNER II was pro- posed, whose goal is to recognize complex en- tities against out of knowledge-base entities and noisy scenarios. To address the challenges posed by MultiCoNER II, our team NetEase.AI proposed an entity recognition system that in- tegrates text error correction system and exter- nal knowledge, which can recognize entities in scenes that contain entities out of knowledge base and text with noise. Upon receiving an input sentence, our systems will correct the sentence, extract the entities in the sentence as candidate set using the entity recognition model that incorporates the gazetteer information, and then use the external knowledge to classify the candidate entities to obtain entity type features. Finally, our system fused the multi-dimensional features of the candidate entities into a stacking model, which was used to select the correct en- tities from the candidate set as the final output. Our system exhibited good noise resistance and excellent entity recognition performance, re- sulting in our team’s first place victory in the Chinese track of MultiCoNER II. 1 Introduction Named Entity Recognition (NER) is a core Nat- ural Language Processing (NLP) task that aims to detect and classify entities in a sequence. In practical and open-domain settings, however, it is difficult for machines to process complex and am- biguous named entities(Ashwini and Choi, 2014), and the performance of the NER system can be significantly reduced when the text contains noise (such as spelling errors). To alleviate this issue, SemEval 2023 Task 2(Fetahu et al., 2023b,a; Mal- masi et al., 2022b,a) was formulated to focus on recognizing semantically ambiguous and complex 1:First authors,2:Second authors,*:Corresponding authors. entities against out of knowledge-base entities and noisy scenarios. In practice, noise in the text leads to deviation of the text semantics. Therefore, it is often necessary to employ some methods to cor- rect the erroneous text in order to minimize the semantic distortion and its impact on the textual understanding. When labeling complex entities, people often rely on external knowledge to aid them. Several studies have found that incorporating prior knowledge is beneficial for entity recognition (Tedeschi et al., 2021; Hu and Wei, 2020; Seyler et al., 2018). Therefore, we believe that correcting noise and leveraging external knowledge can help to recognize entities in noisy text. To effectively combine all the features, we use a stacking model (Pavlyshenko, 2018) to integrate the semantic fea- tures of the text and external …

Content truncated to 500 words