Research Article
A Transformer-Based Model for Named Entity Recognition in Winning Bid Text
@INPROCEEDINGS{10.4108/eai.18-12-2025.2365295, author={Yalan Ling and Zhuangye Luo and Feng Zeng and Xiaowei Xie}, title={A Transformer-Based Model for Named Entity Recognition in Winning Bid Text}, proceedings={Proceedings of the 13th International Conference on Identification, Information and Knowledge in the Internet of Things, IIKI 2025, 18-21 December 2025, Chengdu, China}, publisher={EAI}, proceedings_a={IIKI}, year={2026}, month={6}, keywords={Named Entity Recognition (NER) Bidirectional Encoder Representations from Transformers (BERT) winning bid text mining}, doi={10.4108/eai.18-12-2025.2365295} }- Yalan Ling
Zhuangye Luo
Feng Zeng
Xiaowei Xie
Year: 2026
A Transformer-Based Model for Named Entity Recognition in Winning Bid Text
IIKI
EAI
DOI: 10.4108/eai.18-12-2025.2365295
Abstract
There are quite a few business information in the winning bid document, which is relatively important for the business transactions in the bidding market. The mining of winning bid text belongs to the relevant matters of the Named Entity Recognition (NER). This paper constructs TransBERT, which uses channel attention and bidirectional encoding representation of character-level features to explore the information in text. The model has deep semantic representation, can capture sequence dependence, and enforce label consistency. In order to solve the problem of entity boundary segmentation errors, the Channel Attention Mechanism (DSENet) is constructed, and the Character-level ConvNet (CharCNN) is introduced to capture character-level semantic information. These methods focus on entity content and localization, and also suppress irrelevant features to enhance recognition. The experimental evaluation shows that the performance of TransBERT is better than the current most advanced baseline model. Its precision has increased by 17.37%, the recall has increased by 7.54%, and the F1-score has increased by 12.84%.
