IITD at SemEval-2023 Task 2: A Multi-Stage Information Retrieval Approach for Fine-Grained Named Entity Recognition

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 800–806 July 13-14, 2023 ©2023 Association for Computational Linguistics IITD at SemEval-2023 Task 2: A Multi-Stage Information Retrieval Approach for Fine-Grained Named Entity Recognition Shivani Choudhary, Niladri Chatterjee, Subir Kumar Saha {shivani@sire, niladri@maths, saha@mech}.iitd.ac.in Indian Institute of Technology Delhi Hauz Khas, Delhi-110016, India Abstract MultiCoNER-II is a fine-grained Named Entity Recognition (NER) task that aims to identify ambiguous and complex named entities in mul- tiple languages, with a small amount of contex- tual information available. To address this task, we propose a multi-stage information retrieval (IR) pipeline that improves the performance of language models for fine-grained NER. Our ap- proach involves leveraging a combination of a BM25-based IR model and a language model to retrieve relevant passages from a corpus. These passages are then used to train a model that utilizes a weighted average of losses. The pre- diction is generated by a decoder stack that includes a projection layer and conditional ran- dom field. To demonstrate the effectiveness of our approach, we participated in the English track of the MultiCoNER-II competition. Our approach yielded promising results, which we validated through detailed analysis. 1 Introduction Named entity recognition (NER) is a crucial task in natural language processing (NLP), where the goal is to identify and classify named entities in text, such as people, organizations, locations, and products. NER has a wide range of applications, in- cluding information extraction, question answering, and sentiment analysis, among others. Recently, there has been a growing interest in the development of NER models that can handle ambiguous, complex, and emerging named entities with shorter context. In response to this, the Multi- CoNER challenge was introduced, which aims to evaluate NER models on a diverse range of named entity classes and contexts. The first iteration of the MultiCoNER challenge, MultiCoNER-I, focused on classification into six classes: PER, LOC, GRP, CORP, PROD, CW (Mal- masi et al., 2022a). However, the performance of NER systems on complex entities such as creative works, groups, and products was found to be lesser than on entities such as Person (Malmasi et al., 2022b). In response to this, the second iteration of the MultiCoNER challenge, MultiCoNER-II, was introduced, which aimed to address this issue by providing a fine-grained NER task with 36 differ- ent classes(Fetahu et al., 2023a). To make the task more realistic, the organizers added noise to the test dataset. Additionally, a new coarse-grained class, Medical, was intro- duced in MultiCoNER-II, and the CORP class was subsumed into GRP. The coarse-grained classes were further subdivided into fine-grained entities. MultiCoNER-II consists of 13 different tracks, in- cluding 11 monolingual tracks for languages such as English, Spanish, Dutch, Russian, Turkish, Ko- rean, Farsi, German, Chinese, Hindi, and Bangla, as well as a code-mixed track and a multilingual track(Fetahu et al., 2023b). To tackle the MultiCoNER-II challenge, we pro- pose an NER system based on the concepts of a single-stage information retrieval (IR) framework. Our information retrieval pipeline uses two docu- …

Content truncated to 500 words