Abstract: With the acceleration of digitalization in public resource transactions, data such as tender announcements, bid-winning notices, and bidding records have shown an exponential growth trend. These data contain core information including transaction entities, project attributes, and related entities. How to achieve accurate and generalized entity recognition in such data remains an urgent challenge in this field. This paper proposes a named entity recognition (NER) method for public transaction data based on the LBC (LIFL-BERT-CRF) model. The LIFL module addressed the limitation of insufficient semantic representation in individual characters, BERT provided pre-trained contextual embedding, and the CRF layer optimized global sequence dependencies by considering contextual relationships across the entire sequence. This integrated approach achieved globally optimized decision-making for NER in public resource transactions. The experimental results demonstrate that the proposed model significantly outperforms existing baseline methods in both accuracy and effectiveness.