AI-based document processing system and document processing method

Abstract

According to the document processing system and method based on the AI, the document processing system comprises a document recognition system and an archiving and verification system, the document recognition system can conduct text recognition on original documents uploaded by a user, conduct document classification on the original documents according to the recognition result, and conduct document classification on the original documents according to the recognition result. After the original document is split and renamed according to a classification result, the split document is sent to an archiving and cancel-after-verification system, after the archiving and cancel-after-verification system receives the split document, the archiving and cancel-after-verification system can perform archiving and cancel-after-verification on the split document according to a preset archiving and cancel-after-verification strategy, the process does not need workers to perform operation, and the efficiency is improved. In this way, automatic archiving and cancel-after-verification operation can be completed through the document recognition system and the archiving and cancel-after-verification system, and therefore a large amount of manual work is remarkably reduced, and the archiving and cancel-after-verification efficiency is effectively improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an AI-based document processing system and document processing method. Background Technology

[0002] Currently, when archiving and verifying documents, since most existing documents are complex and composite documents, this type of document usually contains multiple convergent documents of different types in the same file. Each type of document contains multiple convergent layouts, but the types of documents contained in different files, the order in which each document appears, the number of each document, and the number of pages will vary.

[0003] For example, in a 20-page PDF file, pages 1-10 are expense settlement statements (which can be further divided into 5 expense settlement statements based on the settlement statement number within the document, with page numbers 2, 3, 2, 2, and 1 respectively, meaning documents with the same settlement statement number are the same expense settlement statement), pages 11-15 are invalid documents that have no effect on the business, and pages 16-20 are income and expenditure settlement statements (which can be further divided into 5 income and expenditure settlement statements based on the settlement statement number within the document, with each of the 5 statements being 1 page, meaning the settlement statement numbers on the 5 pages are different).

[0004] In the existing technology, when a user obtains the original document file, they will generally open the file first, check page by page whether the page should be retained, and if it is retained, check whether the order number of each page is consistent. Then, the user will manually split the file, rename the split file manually, log in to the system to search for the order number, find the corresponding system record, and manually upload the file with the same order number to the system record to complete the file archiving and verification.

[0005] During this process, users need to review documents page by page, manually split the original files according to document type, and manually rename the files to record their ownership. If there are multiple order numbers, they also need to archive and verify them separately, resulting in low efficiency and high labor costs. Summary of the Invention

[0006] The purpose of this application is to at least address one of the aforementioned technical deficiencies, particularly the technical deficiency of low efficiency and high labor costs in the prior art of manually archiving and verifying documents.

[0007] This application provides an AI-based document processing system, the system comprising:

[0008] The document recognition system is used to perform text recognition on original documents uploaded by users, classify the original documents according to the recognition results, split and rename the original documents according to the classification results, and send the split documents to the archiving and verification system.

[0009] The archiving and verification system is used to receive the split documents sent by the document recognition system and to archive and verify the split documents according to a preset archiving and verification strategy.

[0010] This application provides an AI-based document processing method, applied to the document recognition system within the AI-based document processing system described in the above embodiments. The method includes:

[0011] Retrieve the original document uploaded by the user;

[0012] The original document is subjected to text recognition, and the original document is classified according to the recognition results to obtain the classification results;

[0013] After splitting and renaming the original document according to the classification results, the split document is sent to the archiving and verification system.

[0014] Optionally, the step of performing text recognition on the original document to obtain the recognition result includes:

[0015] The content of each page in the original document is detected to determine whether each page contains images;

[0016] If it is included, then the included image is identified and the text information in the image is obtained;

[0017] If not included, the text information of each page of the document will be obtained directly after parsing each page.

[0018] Optionally, the recognition result includes the text information corresponding to each page of the original document;

[0019] The step of classifying the original document based on the recognition result to obtain the classification result includes:

[0020] Based on the text information corresponding to each page of the original document, the original document is initially classified, and then grouped according to the initial classification results to obtain at least one set of documents;

[0021] For each document set, extract the index number from the document set, and perform secondary classification on the document set according to the extraction results to obtain the document types contained in the document set, the page number range corresponding to each type of document, and the index number.

[0022] Optionally, the step of performing preliminary classification of the original document based on the text information corresponding to each page of the original document to obtain a preliminary classification result includes:

[0023] Based on the text information corresponding to each page of the original document, key information is extracted from each page of the document, and semantic information corresponding to the key information of each page of the document is determined;

[0024] Obtain user requirements, and perform preliminary classification of each page of documents based on the semantic information of each page and the user requirements to obtain preliminary classification results.

[0025] Optionally, the preliminary classification results include valid documents and invalid documents, as well as the document type of valid documents;

[0026] The original documents are grouped according to the preliminary classification results to obtain at least one set of documents, including:

[0027] When the preliminary classification results contain at least one type of valid documents, the valid documents of the same type are grouped together from front to back according to the page number order of the original documents to obtain at least one set of documents.

[0028] When the preliminary classification results contain invalid documents and at least one type of valid documents, the valid documents of the same type and the invalid documents following the valid documents of that type are grouped together according to the page number order of the original documents, to obtain at least one set of documents.

[0029] Optionally, the step of extracting index numbers from the set of documents and obtaining the extraction results includes:

[0030] Identify the key information in each page of the document collection;

[0031] The index number is extracted from the key information of each page of the document using the number extraction model, and the extraction result corresponding to each page of the document set is obtained.

[0032] Optionally, the secondary classification of the document set according to the extraction results to obtain the document types contained in the document set, the page number range corresponding to each type of document, and the index number includes:

[0033] Determine if any documents without an index number exist in the extraction results;

[0034] If it exists, discard the documents in the document set that do not contain an index number, and take the documents with the same index number as the same type of documents in the remaining documents, so as to obtain the document types contained in the document set, the page number range corresponding to each type of document and the index number.

[0035] If none exists, documents with the same index number in the document set are directly treated as documents of the same type, thus obtaining the document types contained in the document set, the page number range corresponding to each type of document, and the index number.

[0036] Optionally, the step of splitting and renaming the original document according to the classification result to obtain the split document includes:

[0037] The original document is split and renamed according to the classification results using the rule-based splitting service to obtain the split document.

[0038] This application also provides an AI-based document processing method, applied to the archiving and verification system within the AI-based document processing system described in the above embodiments, the method comprising:

[0039] Receive the segmented document sent by the document recognition system;

[0040] The split documents are archived and revoked according to the preset archiving and revocation strategy.

[0041] Optionally, the step of archiving and reconciling the split documents according to a preset archiving and reconciliation strategy includes:

[0042] Parse the index numbers corresponding to the split documents;

[0043] The index number is compared with the pre-stored document records to determine whether the index number exists in the document records;

[0044] If it exists, then after parsing the document type corresponding to the split document, the split document is archived and cancelled.

[0045] If it does not exist, then determine whether the document classification or numbering of the split document is incorrect;

[0046] If the document classification and numbering of the split documents are correct, then the split documents will be supplemented with data and archived for verification.

[0047] If the document classification or numbering of the split documents is incorrect, the split documents will be corrected and archived for verification.

[0048] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0049] This application provides an AI-based document processing system and method. The document processing system includes a document recognition system and an archiving and verification system. The document recognition system can perform text recognition on the original documents uploaded by users, classify the original documents according to the recognition results, and split and rename the original documents according to the classification results. The split documents are then sent to the archiving and verification system. After receiving the split documents, the archiving and verification system can archive and verify the split documents according to a preset archiving and verification strategy. This process does not require manual operation and can be automated through the document recognition system and the archiving and verification system, thereby significantly reducing a large amount of manual work and effectively improving the efficiency of archiving and verification. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A system architecture diagram of an AI-based document processing system provided for embodiments of this application;

[0052] Figure 2 A flowchart illustrating an AI-based document processing method provided in this application embodiment;

[0053] Figure 3 This is a schematic diagram illustrating the process of grouping documents according to an embodiment of this application;

[0054] Figure 4 This is a flowchart illustrating another AI-based document processing method provided in an embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] In the existing technology, when a user obtains the original document file, they will generally open the file first, check page by page whether the page should be retained, and if it is retained, check whether the order number of each page is consistent. Then, the user will manually split the file, rename the split file manually, log in to the system to search for the order number, find the corresponding system record, and manually upload the file with the same order number to the system record to complete the file archiving and verification.

[0057] During this process, users need to review documents page by page, manually split the original files according to document type, and manually rename the files to record their ownership. If there are multiple order numbers, they also need to archive and verify them separately, resulting in low efficiency and high labor costs.

[0058] Based on this, this application proposes the following technical solution:

[0059] In one embodiment, such as Figure 1 As shown, Figure 1 A system architecture diagram of an AI-based document processing system provided for embodiments of this application; this application provides an AI-based document processing system, which may include:

[0060] The document recognition system is used to perform text recognition on original documents uploaded by users, classify the original documents according to the recognition results, split and rename the original documents according to the classification results, and send the split documents to the archiving and verification system.

[0061] The archiving and verification system is used to receive the split documents sent by the document recognition system and to archive and verify the split documents according to a preset archiving and verification strategy.

[0062] In this embodiment, the document processing system may include not only a document recognition system but also an archiving and verification system. This allows the document recognition system to perform text recognition, splitting, and renaming operations on the original documents uploaded by users. The split documents can also be sent to the archiving and verification system as a file stream. The archiving and verification system can then archive and verify the split documents according to a preset archiving and verification strategy, thereby completing the classification, splitting, archiving, and verification operations of documents.

[0063] The document recognition system of this application can use multiple recognition methods to recognize the original document. For example, when the original document contains images, image recognition methods can be used to recognize the text in the images. When the original document contains other types of text, the corresponding recognition method can be determined by the corresponding text type, and the recognition method can be used to recognize the text information in the original document.

[0064] It is understandable that the original document can be a single page or multiple pages of a single original document, or a single page or multiple pages of multiple original documents. Therefore, when performing text recognition on the original document, this application can recognize the text on each page of the original document, thereby obtaining the text information contained in each page of the original document.

[0065] In addition, the original documents uploaded by users may contain documents of the same type, documents of different types, or multiple documents of the same type. For example, in a 20-page PDF file, pages 1-10 are expense settlement statements (which can be further divided into 5 expense settlement statements based on the settlement statement number within the document, with page numbers 2, 3, 2, 2, and 1 respectively, meaning documents with the same settlement statement number are the same expense settlement statement), pages 11-15 are invalid documents that are not relevant to the business, and pages 16-20 are income and expenditure settlement statements (which can be further divided into 5 income and expenditure settlement statements based on the settlement statement number within the document, with each of the 5 statements being 1 page, meaning the settlement statement numbers on the 5 pages are different).

[0066] Based on this, in order to improve the accuracy of archiving and reconciliation, after performing text recognition on the original document, the original document can be classified according to the recognition results, thus obtaining multiple different types of documents. Then, the original document can be split according to the classification results, and the split documents can be renamed before being sent to the archiving and reconciliation system. After receiving the split documents, the archiving and reconciliation system can call the pre-set archiving and reconciliation strategy and perform archiving and reconciliation on the split documents according to the archiving and reconciliation strategy.

[0067] It is understood that, in order to further improve the accuracy of archiving and reconciliation, this application can pre-set archiving and reconciliation strategies in the archiving and reconciliation system. These strategies can be rules set for the archiving and reconciliation process, as well as solutions for problems that may be encountered during the archiving and reconciliation process. In this way, the archiving and reconciliation system can perform archiving and reconciliation operations according to the preset archiving and reconciliation strategies, thereby achieving automatic archiving and reconciliation.

[0068] In the above embodiments, the document processing system includes a document recognition system and an archiving and verification system. The document recognition system of this application can perform text recognition on the original document uploaded by the user, classify the original document according to the recognition result, and split and rename the original document according to the classification result before sending the split document to the archiving and verification system. After receiving the split document, the archiving and verification system can archive and verify the split document according to the preset archiving and verification strategy. This process does not require manual operation and can be automated through the document recognition system and the archiving and verification system, thereby significantly reducing a large amount of manual work and effectively improving the efficiency of archiving and verification.

[0069] In one embodiment, such as Figure 2 As shown, Figure 2 This application provides a flowchart illustrating an AI-based document processing method. The application provides an AI-based document processing method applied to the document recognition system within the AI-based document processing system described in the above embodiments. The method may include:

[0070] S110: Retrieve the original document uploaded by the user.

[0071] S120: Perform text recognition on the original document and classify the original document based on the recognition results to obtain the classification results.

[0072] S130: After splitting and renaming the original documents according to the classification results, send the split documents to the archiving and verification system.

[0073] In this embodiment, the document recognition system can receive original documents uploaded by users. These original documents can be one page or multiple pages of a single original document, or one page or multiple pages of multiple original documents. Therefore, when performing text recognition on the original document, this application can recognize the text on each page of the original document, thereby obtaining the text information contained in each page of the original document.

[0074] Specifically, when performing text recognition on the original document, this application can use a variety of recognition methods to recognize the original document. For example, when the original document contains images, image recognition methods can be used to recognize the text in the images. When the original document contains other types of text, the corresponding recognition method can be determined by the corresponding text type, and the recognition method can be used to recognize the text information in the original document.

[0075] Next, this application can classify the original document according to the recognition results. After obtaining multiple different types of documents, this application can split the original document according to the classification results, rename the split document, and send the split document to the archiving and verification system. After receiving the split document, the archiving and verification system can call the pre-set archiving and verification strategy and archive and verify the split document according to the archiving and verification strategy, thereby effectively improving the archiving and verification efficiency and reducing labor costs.

[0076] In one embodiment, performing text recognition on the original document in step S120 to obtain the recognition result may include:

[0077] S121: Detect the content contained in each page of the original document to determine whether each page contains images.

[0078] S122: If included, then after recognizing the included image, obtain the text information in the image.

[0079] S123: If not included, then directly parse each page of the document and obtain the text information of each page of the document.

[0080] In this embodiment, when performing text recognition on the original document, the content contained in each page of the original document can be detected first. If each page contains an image, an image recognition method can be used to recognize the contained image in order to obtain the text information in the image. If no image is contained, each page of the document can be directly parsed to obtain the text information of each page of the document.

[0081] In one specific implementation, this application can detect the original document page by page and determine whether the current page contains images. If not, it parses the document using its format protocol encoding to obtain all text information for the current page. If images are present, it uses Optical Character Recognition (OCR) technology to obtain all text information for the current page. Once the text information for each page of the original document is determined, the original document can be classified based on the text information of each page.

[0082] In one embodiment, the recognition result may include text information corresponding to each page of the original document.

[0083] In step S120, the original document is classified based on the recognition result to obtain the classification result, which may include:

[0084] S124: Based on the text information corresponding to each page of the original document, perform preliminary classification on the original document, and group the original document according to the preliminary classification results to obtain at least one set of documents.

[0085] S125: For each set of documents, extract the index number from the set of documents, and perform secondary classification on the set of documents according to the extraction results to obtain the document types contained in the set of documents, the page number range corresponding to each type of document, and the index number.

[0086] In this embodiment, since the original documents uploaded by users may contain documents of the same type, documents of different types, or multiple documents of the same type, this application can first perform a preliminary classification of the original documents based on the text information corresponding to each page of the original documents, and then group the original documents according to the preliminary classification results to obtain at least one set of documents. Next, for each set of documents, this application can extract the index number from that set, and then perform a secondary classification of that set of documents based on the extraction results, thereby obtaining the document types contained in that set of documents, the page range corresponding to each type of document, and the index number.

[0087] For example, regarding the aforementioned 20-page PDF file, this application can first perform a preliminary classification of the PDF file based on the text information of each page, thereby dividing the PDF file into expense settlement statements, income and expenditure statements, and invalid documents. Further, this application can extract the settlement statement number, settlement date, or other identifying information from the expense settlement statement as index numbers, and then perform a secondary classification of the expense settlement statement based on the extracted index numbers, thereby dividing the original expense settlement statement into one or more expense settlement statements with different settlement statement numbers and the corresponding page range and index number for each expense settlement statement. Similarly, this application can also perform a secondary classification of other types of documents in the original document, such as income and expenditure statements, thereby obtaining one or more income and expenditure statements with different settlement statement numbers.

[0088] In one embodiment, step S124, which involves performing a preliminary classification of the original document based on the text information corresponding to each page of the original document to obtain a preliminary classification result, may include:

[0089] S1241: Based on the text information corresponding to each page of the original document, extract key information from each page of the document and determine the semantic information corresponding to the key information of each page of the document.

[0090] S1242: Obtain user requirements, perform preliminary classification of each page of documents based on the semantic information of each page and the user requirements, and obtain preliminary classification results.

[0091] In this embodiment, when performing preliminary classification of the original document, key information can be extracted from each page of the original document based on the text information of each page. Then, semantic analysis is performed on this key information to determine the semantic information corresponding to the key information on each page. Furthermore, this application can also obtain user requirements, and then perform preliminary classification of each page based on the semantic information of each page and the user requirements, obtaining preliminary classification results.

[0092] The key information in this application includes, but is not limited to, titles and key fields, with key fields including, but not limited to, invoice numbers and customs numbers. When extracting key information from the text information of each page of the original document, this application can utilize a feature extraction model. This model can extract key information using Principal Component Analysis (PCA), which projects the text information into a new coordinate system through linear transformation, maximizing the variance of the text information in the new coordinate system and thus extracting the main feature components. Alternatively, it can use Linear Discriminant Analysis (LDA), a supervised feature extraction method that aims to find a projection direction that maximizes the distance between different categories while minimizing the distance within the same category. Of course, deep learning can also be used for feature extraction, such as Convolutional Neural Networks (CNNs) which automatically learn image features through convolutional and pooling layers; Recurrent Neural Networks (RNNs) and Long Short-Term Memory Networks (LSTMs) can extract features from time-series data. The specific choice depends on the actual situation and is not limited here.

[0093] After extracting key information, this application can also use a feature classification model to perform semantic parsing on the key information, thereby determining the semantic information corresponding to the key information on each page of the document. Specifically, this application can use decision trees for feature classification; for example, it can construct a tree structure, branch based on feature values, and ultimately classify the data into different categories. Alternatively, this application can use Naive Bayes for feature classification; for example, it can use Bayes' theorem, assuming that features are independent, to calculate the probability of features in different categories for classification. Of course, this application can also use artificial neural networks for classification, especially structures such as multilayer perceptrons (MLPs), which can learn complex patterns of features and perform classification. The specific choice depends on the actual situation and is not limited here.

[0094] Once this application determines the semantic information corresponding to the key information of each page of the original document, it can obtain user requirements. Then, based on the semantic information of each page and the user requirements, it can perform a preliminary classification of each page and obtain a preliminary classification result. It can be understood that the user requirements here can be the document type to be classified in this document classification, the document type to be classified for multiple original documents sent within a certain period, or the document type to be classified for a certain type of original document. Thus, after obtaining the user requirements, the document recognition system can perform a preliminary classification of the original document based on the user requirements and the pre-determined semantic information of each page, and identify the document types mentioned in the user requirements as valid documents, determining the document type of that category based on the user requirements.

[0095] In addition, users can classify unnecessary document types as invalid documents. This way, such invalid documents can be discarded during subsequent document grouping and splitting, thereby improving the accuracy of archiving and reconciliation.

[0096] In one embodiment, such as Figure 3 As shown, Figure 3 This is a schematic diagram of a process for grouping documents according to an embodiment of this application; the preliminary classification result may include valid documents and invalid documents, as well as the document type of valid documents.

[0097] In step S124, the original documents are grouped according to the preliminary classification results to obtain at least one set of documents, which may include:

[0098] S1243: When the preliminary classification results contain at least one type of valid documents, the valid documents of the same type are grouped together from front to back according to the page number order of the original documents to obtain at least one set of documents.

[0099] In this step, since the preliminary classification results of this application include valid documents and invalid documents, as well as the document type of valid documents, a multi-page valid document often has a more obvious first page feature and weaker features on subsequent pages, making it easy to be classified as an invalid page in the preliminary classification. Therefore, when grouping the original documents based on the preliminary classification results, this application can first determine the document types included in the preliminary classification results. If only at least one type of valid document is included, then it is only necessary to group the valid documents of the same type together according to the page number order of the original document from front to back to obtain at least one set of documents, where the document types in each set of documents are the same and all are valid documents.

[0100] S1244: When the preliminary classification results contain invalid documents and at least one type of valid documents, the valid documents of the same type and the invalid documents following the valid documents of that type are grouped together according to the page number order of the original documents, to obtain at least one set of documents.

[0101] In this step, when the preliminary classification results contain both invalid documents and at least one valid document, this application can initially group the invalid documents forward for subsequent secondary verification. Specifically, this application can group valid documents of the same type and the invalid documents following them according to the page number order of the original documents, thus obtaining at least one set of documents. Each set of documents may contain only one type of valid document, or it may contain one valid document and invalid documents.

[0102] It should be noted that if the preliminary classification results all contain invalid documents, since the invalid documents do not match the user's needs, such preliminary classification results can be discarded directly after feedback to the user, without any further action.

[0103] In one embodiment, extracting index numbers from the set of documents in step S125 and obtaining the extraction results may include:

[0104] S1251: Determine the key information of each page of the document in this set of documents.

[0105] S1252: Extract index numbers from the key information of each page of the document using the number extraction model to obtain the extraction results for each page of the document set.

[0106] In this embodiment, after obtaining at least one set of documents, this application can perform secondary classification on each set of documents. Specifically, when performing secondary classification on each set of documents, this application can first determine the key information of each page of the document in that set of documents. The specific determination process can be found in the above-described process of extracting key information from each page of the document based on the text information of each page of the original document, and will not be repeated here.

[0107] Once the key information of each page of the document set is determined, this application can use a number extraction model to extract the index number from the key information of each page of the document set, thereby obtaining the extraction result of each page of the document set.

[0108] The aforementioned index extraction model incorporates functions such as key-value classification, table removal, index number feature extraction, and index number verification. Therefore, this model can quickly extract the index number from the key information of each page of a document. The index number can be one or multiple, depending on the actual extraction results; no restriction is imposed here. Furthermore, different document types in this application can use independent models to extract index numbers, or a fusion model can be used to extract index numbers from different document types; no restriction is imposed here either.

[0109] In one embodiment, step S125 involves secondary classification of the document set based on the extraction results to obtain the document types contained in the document set, the page number range corresponding to each type of document, and the index number. This may include:

[0110] S1253: Determine if there are any documents in the extraction results that do not contain an index number.

[0111] S1254: If it exists, discard the documents in the document set that do not contain an index number, and take the documents with the same index number as the same type of documents in the remaining documents, so as to obtain the document types contained in the document set, the page number range corresponding to each type of document and the index number.

[0112] S1255: If it does not exist, then the documents with the same index number in the document set are directly regarded as documents of the same type, and the document types, page number ranges and index numbers of the document set are obtained.

[0113] In this embodiment, after obtaining the extraction results for each document set, since the extraction results contain one or more index numbers extracted from each page of that document set, for each document set, this application can determine whether there are documents in that document set that do not contain index numbers based on the extraction results. If such documents exist, they are discarded directly, and then the remaining documents in that document set with the same index numbers are treated as documents of the same type, thereby obtaining the document types contained in that document set, the page range corresponding to each type of document, and the index numbers. If no such documents exist, the documents in that document set with the same index numbers can be directly treated as documents of the same type, thus also obtaining the document types contained in that document set, the page range corresponding to each type of document, and the index numbers.

[0114] In one embodiment, after splitting and renaming the original document according to the classification result in S130, the resulting split document may include:

[0115] S131: After splitting and renaming the original document according to the classification result through the rule splitting service, the split document is obtained.

[0116] In this embodiment, when splitting and renaming the original document, the document recognition system can call a rule-based splitting service. This service splits the original document into smaller, more manageable units according to pre-set rules. Using this rule-based splitting service, the original document can be quickly split into multiple documents according to the classification results, and each document can be renamed to obtain clearer and more accurate split documents.

[0117] In one embodiment, such as Figure 4 As shown, Figure 4 This application provides a flowchart illustrating another AI-based document processing method according to an embodiment of the present application. The application also provides an AI-based document processing method applied to the archiving and verification system within the AI-based document processing system described in the above embodiments. The method may include:

[0118] S210: Receives the split document sent by the document recognition system.

[0119] S220: Archive and verify the split documents according to the preset archiving and verification strategy.

[0120] In this embodiment, in order to further improve the accuracy of archiving and reconciliation, an archiving and reconciliation strategy can be set in advance in the archiving and reconciliation system. The archiving and reconciliation strategy can be a rule set for the archiving and reconciliation process, as well as a solution set for problems that may be encountered in the archiving and reconciliation process. In this way, the archiving and reconciliation system can perform archiving and reconciliation operations according to the preset archiving and reconciliation strategy, thereby realizing automatic archiving and reconciliation.

[0121] In one embodiment, S220, which involves archiving and revoking the split documents according to a preset archiving and revocation strategy, may include:

[0122] S221: Parse the index number corresponding to the split document.

[0123] S222: Compare the index number with the pre-stored document records to determine whether the index number exists in the document records.

[0124] S223: If it exists, then after parsing the document type corresponding to the split document, archive and verify the split document.

[0125] S224: If not, determine whether the document classification or numbering of the split document is incorrect.

[0126] S225: If the document classification and numbering of the split documents are correct, then the split documents are supplemented with data and archived for verification.

[0127] S226: If the document classification or numbering of the split documents is incorrect, then the split documents shall be corrected and archived for verification.

[0128] In this embodiment, when archiving and revoking the split documents according to the preset archiving and revoking strategy, the index number corresponding to the split document can be parsed first. Then, the index number is compared with the index number in the pre-stored document records to determine whether the index number exists in the document records. If it exists, the document type corresponding to the split document is parsed, and the split document is archived and revoked. If it does not exist, it can first determine whether the document classification or number identification of the split document is incorrect. If the document classification and number identification of the split document are both correct, the data supplementation and archiving and revoking of the split document can continue. If the document classification or number identification of the split document is incorrect, the data correction and archiving and revoking of the split document can be performed.

[0129] Specifically, when this application detects that the index number is not present in the document record, it can first notify the corresponding manual review exception form via email or other means. If the classification is correct but the data is missing in the system, the data will be supplemented and archived for verification. If the document classification or number identification is incorrect, the document will be manually corrected and archived to further improve the accuracy of archiving and verification.

[0130] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0131] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0132] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-based document processing system, characterized in that, The system includes: The document recognition system is used to perform text recognition on original documents uploaded by users, classify the original documents according to the recognition results, split and rename the original documents according to the classification results, and send the split documents to the archiving and verification system. The archiving and verification system is used to receive the split documents sent by the document recognition system and to archive and verify the split documents according to a preset archiving and verification strategy.

2. An AI-based document processing method, applied to the document recognition system in the AI-based document processing system of claim 1, characterized in that, The method includes: Retrieve the original document uploaded by the user; The original document is subjected to text recognition, and the original document is classified according to the recognition results to obtain the classification results; After splitting and renaming the original document according to the classification results, the split document is sent to the archiving and verification system.

3. The AI-based document processing method according to claim 2, characterized in that, The text recognition process for the original document, to obtain the recognition result, includes: The content of each page in the original document is detected to determine whether each page contains images; If it is included, then the included image is identified and the text information in the image is obtained; If not included, the text information of each page of the document will be obtained directly after parsing each page.

4. The AI-based document processing method according to claim 2, characterized in that, The recognition result includes the text information corresponding to each page of the original document; The step of classifying the original document based on the recognition result to obtain the classification result includes: Based on the text information corresponding to each page of the original document, the original document is initially classified, and then grouped according to the initial classification results to obtain at least one set of documents; For each document set, extract the index number from the document set, and perform secondary classification on the document set according to the extraction results to obtain the document types contained in the document set, the page number range corresponding to each type of document, and the index number.

5. The AI-based document processing method according to claim 4, characterized in that, The preliminary classification of the original document based on the text information corresponding to each page of the original document, to obtain a preliminary classification result, includes: Based on the text information corresponding to each page of the original document, key information is extracted from each page of the document, and semantic information corresponding to the key information of each page of the document is determined; Obtain user requirements, and perform preliminary classification of each page of documents based on the semantic information of each page and the user requirements to obtain preliminary classification results.

6. The AI-based document processing method according to claim 4, characterized in that, The preliminary classification results include valid documents and invalid documents, as well as the document type of valid documents; The original documents are grouped according to the preliminary classification results to obtain at least one set of documents, including: When the preliminary classification results contain at least one type of valid documents, the valid documents of the same type are grouped together from front to back according to the page number order of the original documents to obtain at least one set of documents. When the preliminary classification results contain invalid documents and at least one type of valid documents, the valid documents of the same type and the invalid documents following the valid documents of that type are grouped together according to the page number order of the original documents, to obtain at least one set of documents.

7. The AI-based document processing method according to claim 4, characterized in that, The process of extracting index numbers from this set of documents and obtaining the extraction results includes: Identify the key information in each page of the document collection; The index number is extracted from the key information of each page of the document using the number extraction model, and the extraction result corresponding to each page of the document set is obtained.

8. The AI-based document processing method according to claim 4, characterized in that, The document set is then classified a second time according to the extraction results to obtain the document types contained in the document set, the page number range corresponding to each type of document, and the index number, including: Determine if any documents without an index number exist in the extraction results; If it exists, discard the documents in the document set that do not contain an index number, and take the documents with the same index number as the same type of documents in the remaining documents, so as to obtain the document types contained in the document set, the page number range corresponding to each type of document and the index number. If none exists, documents with the same index number in the document set are directly treated as documents of the same type, thus obtaining the document types contained in the document set, the page number range corresponding to each type of document, and the index number.

9. The AI-based document processing method according to any one of claims 2-8, characterized in that, The process of splitting and renaming the original document based on the classification result yields the split document, including: The original document is split and renamed according to the classification results using the rule-based splitting service to obtain the split document.

10. An AI-based document processing method, applied to the archiving and verification system within the AI-based document processing system of claim 1, characterized in that, The method includes: Receive the segmented document sent by the document recognition system; The split documents are archived and revoked according to the preset archiving and revocation strategy.

11. The AI-based document processing method according to claim 10, characterized in that, The step of archiving and verifying the split documents according to a preset archiving and verification strategy includes: Parse the index numbers corresponding to the split documents; The index number is compared with the pre-stored document records to determine whether the index number exists in the document records; If it exists, then after parsing the document type corresponding to the split document, the split document is archived and cancelled. If it does not exist, then determine whether the document classification or numbering of the split document is incorrect; If the document classification and numbering of the split documents are correct, then the split documents will be supplemented with data and archived for verification. If the document classification or numbering of the split documents is incorrect, the split documents will be corrected and archived for verification.

Images