The invention discloses a
document processing method, device and product based on a
large model, and relates to the technical field of
document processing, and the method comprises the steps: inputting a current page text and a previous page text into a preset
large model, and outputting a
classification result and classification confidence of the current page text; determining a last page text of the latest document sleeve based on a
classification result, inputting the current page text and the last page text into a preset
large model, and outputting a continuity
score of the current page; and if the classification confidence coefficient and the continuity
score meet a preset suite condition, updating the current page as a new last page of the newest document suite. The current text page and the last page of text serve as model input, the accuracy of the
classification result output when the model faces
paging receipts is improved, based on the accurate classification result, the current page of text and the last page of text of the newest
receipt sleeve serve as input, and the classification accuracy is improved. And multi-
dimensional analysis is carried out on the structure, the entity and the
business logic by using context
semantics again, so that the sorting accuracy of the model when facing complex and cross-page receipts is improved.