Abstract:With the rapid development of the electronic bidding business of State Grid Corporation of China, suppliers need to timely and accurately obtain key bidding information related to their own business from a large number of bidding documents, such as bidding projects, bid opening time, qualification conditions, etc. This article aims to study a bidding document parsing technology that is suitable for the characteristics of State Grid bidding documents, allowing data to be structured and visualized, thereby helping suppliers lock in bidding opportunities and support business decisions. Firstly, the bidding document text is analyzed through text analysis, table detection, text error correction, etc. to obtain valid text data. Input data into different analytical algorithm models, annotate the data to determine the effectiveness of the model, and obtain 5 suitable algorithm technology options. Then, using sample data from the bidding documents of the State Grid, further research and testing were conducted through model customization and tuning. By collecting training corpus data and constructing test datasets, a bidding document parsing model integrating bidirectional long short memory Bi-LSTM, machine learning CRF, and deep learning BERT was constructed. Finally, 823 sample data from the bidding documents of the State Grid were used for practical training and comparative testing of the model. The final conclusion indicates that the various indicators of the bidirectional long short memory fusion model are higher than those of the BERT+Bi LSTM model, and the machine learning CRF model layer can ensure the effectiveness of the final prediction results by adding some constraints, which can be automatically learned during training data, and can more accurately and effectively improve the parsing effect of bidding documents.