CN-122024262-A - Text structuring processing method and related device based on text recognition large model

CN122024262ACN 122024262 ACN122024262 ACN 122024262ACN-122024262-A

Abstract

The application relates to the technical field of data processing, and provides a text structuring processing method based on a text recognition large model and a related device, wherein the method comprises the steps of obtaining to-be-processed power document data; the method comprises the steps of extracting document data of the to-be-processed electric power document data according to text areas to obtain k area document data, obtaining key information of the k area document data to obtain k document key information, constructing a logic association relation scene graph according to the k document key information and the corresponding area document data to obtain a target logic association relation scene graph, carrying out structuring processing on the to-be-processed electric power document data according to the target logic association relation scene graph by adopting a preset character recognition model to obtain a structured document, carrying out structuring processing on the basis of the logic association relation scene graph and the preset character recognition model to obtain the structured document, and improving accuracy of structured document determination.

Inventors

ZHANG YING
BAO SHUQIN
LIN SHUGUANG
YANG YONG
GAO YU
ZHANG ZEQUAN
ZHANG JIAO
LI MEI
SHAO YING
WANG XUE
BAI XIN
CAI LIJUN
WANG CAI

Assignees

云南电网有限责任公司文山供电局

Dates

Publication Date: 20260512
Application Date: 20251218

Claims (10)

1. A character structuring processing method based on a character recognition large model is characterized by comprising the following steps: acquiring to-be-processed power document data; Extracting document data of the to-be-processed power document data according to the text region to obtain k region document data; acquiring key information of k regional document data to obtain k document key information; constructing a logic association relation scene graph according to k document key information and corresponding regional document data to obtain a target logic association relation scene graph; and carrying out structuring treatment on the to-be-treated power document data by adopting a preset character recognition model according to the target logic association relation scene graph to obtain a structuring document.
2. The text structuring processing method based on the text recognition big model according to claim 1, wherein the extracting the document data of the to-be-processed power document data according to text regions to obtain k region document data comprises: Acquiring document attribute information of the power document data to be processed; determining text region layout information of the power document data to be processed according to the document attribute information; determining k text region information corresponding to the to-be-processed power document data according to the text region layout information; And extracting document data of the power document data to be processed according to the k text region information to obtain k region document data.
3. The text structured processing method based on the text recognition big model according to claim 2, wherein the extracting document data of the power document data to be processed according to k pieces of text area information to obtain k pieces of area document data includes: Performing region labeling on the power document data to be processed according to the k text region information to obtain labeled power document data; And extracting the data of the marked power document data according to the marked areas to obtain k area document data.
4. The word structuring processing method based on the word recognition large model according to any one of claims 1 to 3, wherein constructing a logical association relation scene graph according to k pieces of document key information and corresponding area document data to obtain a target logical association relation scene graph includes: Carrying out relevance analysis on k pieces of document key information respectively to obtain target relevance information between every two pieces of document key information; constructing logic association description information between two corresponding document key information according to the target association information; and constructing a logic association relation scene graph according to the logic association description information between every two pieces of document key information and k pieces of document key information to obtain a target logic association relation scene graph.
5. The word structuring processing method based on the word recognition large model according to any one of claims 1 to 3, wherein constructing a logical association relation scene graph according to k pieces of document key information and corresponding area document data to obtain a target logical association relation scene graph includes: extracting features of k document key information to obtain k key feature information; randomly extracting target key feature information from k key feature information; Extracting relevance scores between k pieces of key feature information and target key feature information to obtain k relevance score values; determining scene relation distances according to the k association degree scoring values to obtain k scene relation distance values; extracting k pieces of association description extraction between the key feature information and the target key feature information to obtain k pieces of association description information; And constructing a logic association relation scene graph according to the k association degree description information, the k scene relation distance values, the k document key information and the corresponding region document data, and obtaining a target logic association relation scene graph.
6. A word structured processing device based on a word recognition large model, the device comprising: a first acquisition unit configured to acquire power document data to be processed; the extraction unit is used for extracting the document data of the to-be-processed power document data according to the text region to obtain k region document data; The second acquisition unit is used for acquiring key information of the k area document data to obtain k document key information; The construction unit is used for constructing a logic association relation scene graph according to k document key information and corresponding regional document data to obtain a target logic association relation scene graph; and the processing unit is used for carrying out structuring processing on the to-be-processed power document data by adopting a preset character recognition model according to the target logic association relation scene graph to obtain a structured document.
7. The word structuring processing device based on the word recognition big model according to claim 6, wherein the extracting unit is specifically configured to: Acquiring document attribute information of the power document data to be processed; determining text region layout information of the power document data to be processed according to the document attribute information; determining k text region information corresponding to the to-be-processed power document data according to the text region layout information; And extracting document data of the power document data to be processed according to the k text region information to obtain k region document data.
8. The text structured processing device based on a large text recognition model according to claim 7, wherein the extracting unit is specifically configured to, in terms of extracting document data of the power document data to be processed according to k pieces of the text region information, obtain k pieces of region document data: Performing region labeling on the power document data to be processed according to the k text region information to obtain labeled power document data; And extracting the data of the marked power document data according to the marked areas to obtain k area document data.
9. A terminal comprising a processor, an input device, an output device, and a memory, the processor, the input device, the output device, and the memory being interconnected, wherein the memory is configured to store a computer program comprising program instructions, the processor being configured to invoke the program instructions to perform the word structured processing method based on a word recognition big model as claimed in any of claims 1-5.
10. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the word structuring processing method based on a word recognition large model as claimed in any one of claims 1-5.

Description

Text structuring processing method and related device based on text recognition large model Technical Field The application relates to the technical field of data processing, in particular to a text structuring processing method based on a text recognition large model and a related device. Background When processing such data, the characteristic extraction is usually carried out by manually setting a keyword rule, and then the data of the type is structured based on the extracted characteristic. When the method is used for processing, the limitation of keyword extraction is faced, and the problems of inaccurate keyword generation, keyword redundancy and the like exist, so that the accuracy of the data in structuring processing is reduced. Disclosure of Invention The embodiment of the application provides a character structuring processing method and a related device based on a character recognition large model, which can construct a logic association relation scene graph of electric document data to be processed, and carry out structuring processing based on the logic association relation scene graph and a preset character recognition model to obtain a structuring document, thereby improving the accuracy of the structuring document in determination. A first aspect of an embodiment of the present application provides a text structuring processing method based on a text recognition large model, where the method includes: acquiring to-be-processed power document data; Extracting document data of the to-be-processed power document data according to the text region to obtain k region document data; acquiring key information of k regional document data to obtain k document key information; constructing a logic association relation scene graph according to k document key information and corresponding regional document data to obtain a target logic association relation scene graph; and carrying out structuring treatment on the to-be-treated power document data by adopting a preset character recognition model according to the target logic association relation scene graph to obtain a structuring document. In one possible implementation manner, the extracting document data from the to-be-processed power document data according to text regions to obtain k region document data includes: Acquiring document attribute information of the power document data to be processed; determining text region layout information of the power document data to be processed according to the document attribute information; determining k text region information corresponding to the to-be-processed power document data according to the text region layout information; And extracting document data of the power document data to be processed according to the k text region information to obtain k region document data. In one possible implementation manner, the extracting document data of the power document data to be processed according to k pieces of text region information to obtain k pieces of region document data includes: Performing region labeling on the power document data to be processed according to the k text region information to obtain labeled power document data; And extracting the data of the marked power document data according to the marked areas to obtain k area document data. In one possible implementation manner, the constructing a logic association relationship scene graph according to k document key information and corresponding region document data to obtain a target logic association relationship scene graph includes: Carrying out relevance analysis on k pieces of document key information respectively to obtain target relevance information between every two pieces of document key information; constructing logic association description information between two corresponding document key information according to the target association information; and constructing a logic association relation scene graph according to the logic association description information between every two pieces of document key information and k pieces of document key information to obtain a target logic association relation scene graph. In one possible implementation manner, the constructing a logic association relationship scene graph according to k document key information and corresponding region document data to obtain a target logic association relationship scene graph includes: extracting features of k document key information to obtain k key feature information; randomly extracting target key feature information from k key feature information; Extracting relevance scores between k pieces of key feature information and target key feature information to obtain k relevance score values; determining scene relation distances according to the k association degree scoring values to obtain k scene relation distance values; extracting k pieces of association description extraction between the key feature information and the target key feature information