BiGCAT: A Graph-Based Representation Learning Model with LLM Embeddings for Named Entity Recognition

Md. Akram Hossain, Abdul Aziz, Muhammad Anwarul Azim, Abu Nowshed Chy, Md Zia Ullah, Mohammad Khairul Islam


Abstract
Named entity recognition from financial text is challenging because of word ambiguity, huge quantity of unknown corporation names, and word abbreviation compared to nonfinancial text. However, models often treat named entities in a linear sequence fashion, which might obscure the model’s ability to capture complex hierarchical relationships among the entities. In this paper, we proposed a novel named entity recognition model BiGCAT, which integrates large language model (LLM) embeddings with graph-based representation where the contextual information captured by the language model and graph representation learning can complement each other. The method builds a spanning graph with nodes representing word spans and edges weighted by LLM embeddings, optimized using a combination of graph neural networks, specifically a graph-convolutional network (GCN) and a graph-attention network (GAT). This approach effectively captures the hierarchical dependencies among the spans. Our proposed model outperformed the state-of-the-art by 10% and 18% on the two publicly available datasets FiNER-ORD and FIN, respectively, in terms of weighted F1 score. The code is available at: https://github.com/Akram1871/BiGCAT-RANLP-2025.
Anthology ID:
2025.ranlp-1.52
Volume:
Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing - Natural Language Processing in the Generative AI Era
Month:
September
Year:
2025
Address:
Varna, Bulgaria
Editors:
Galia Angelova, Maria Kunilovskaya, Marie Escribe, Ruslan Mitkov
Venue:
RANLP
SIG:
Publisher:
INCOMA Ltd., Shoumen, Bulgaria
Note:
Pages:
433–440
Language:
URL:
https://aclanthology.org/2025.ranlp-1.52/
DOI:
Bibkey:
Cite (ACL):
Md. Akram Hossain, Abdul Aziz, Muhammad Anwarul Azim, Abu Nowshed Chy, Md Zia Ullah, and Mohammad Khairul Islam. 2025. BiGCAT: A Graph-Based Representation Learning Model with LLM Embeddings for Named Entity Recognition. In Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing - Natural Language Processing in the Generative AI Era, pages 433–440, Varna, Bulgaria. INCOMA Ltd., Shoumen, Bulgaria.
Cite (Informal):
BiGCAT: A Graph-Based Representation Learning Model with LLM Embeddings for Named Entity Recognition (Hossain et al., RANLP 2025)
Copy Citation:
PDF:
https://aclanthology.org/2025.ranlp-1.52.pdf