Method, system and device for realizing intelligent circulation of business documents

Abstract

The invention discloses a method, system and device for realizing intelligent circulation of business receipts, and the method comprises the following steps: S1, uploading to-be-processed file data through diversified transmission modes by means of a file uploading module provided by the system, and automatically completing the processing of automatic classification of documents in the uploading process; and S2, calling an information extraction module under the action of the corresponding category label. The method has the beneficial effects that high-quality and high-efficiency intelligent circulation of business receipts and data is realized mainly by combining an artificial intelligence technology with business knowledge, common problems caused by manual processing are avoided, high expansibility, high availability and high linkage of the business data are improved, and enterprise cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of business process automation and information processing, and in particular to a method, system and device for realizing intelligent circulation of business documents. Background Art

[0002] In the torrent of the digital age, enterprises are facing unprecedented challenges and opportunities. With the rapid development of information technology, digital transformation has become a key path for enterprises to enhance their competitiveness and achieve sustainable development. In this process, the optimization and automation of business processes are particularly important. Business documents, including invoices, orders, contracts, etc., are the cornerstone of daily operations of enterprises. They are the basis for enterprises to carry out core businesses such as financial management, supply chain management, and customer relationship management. However, traditional manual processing methods are not only inefficient but also prone to errors, which greatly limits the development potential of enterprises. Fortunately, with the development of cutting-edge technologies such as artificial intelligence, machine learning, and natural language processing, the intelligent circulation of business documents has become a reality.

[0003] Business documents are key documents in business operations. They record the transaction details between the enterprise and its suppliers and customers, and are an important basis for the enterprise to conduct financial accounting, inventory management, contract execution and other activities. However, with the increase in business volume, the workload of manually processing these documents has also increased, which not only consumes a lot of manpower and time, but also errors caused by human factors are difficult to avoid. For example, a medium-sized enterprise may need to process hundreds of orders and invoices every day. If these documents need to be manually entered into the system, it is not only inefficient, but also prone to input errors, leading to problems in subsequent financial processing and inventory management. In addition, manual processing of documents may also lead to information islands, that is, information between different departments cannot be shared in a timely manner, affecting the timeliness and accuracy of decision-making.

[0004] Therefore, how to use artificial intelligence technology combined with business knowledge to efficiently and high-quality complete the intelligent circulation of business documents and seamlessly connect with third-party systems is a very important issue to be solved. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and to design a method, system and device for realizing intelligent circulation of business documents.

[0006] The technical solution of the present invention to achieve the above-mentioned purpose is a method, system and device for realizing intelligent circulation of business documents, including the following steps: Step S1: Relying on the file upload module provided by the system, the file data to be processed is uploaded through a variety of transmission methods, and the document automatic classification processing will be automatically completed during the uploading process.

[0007] The various transmission methods in step S1 include manual upload and automatic upload. The automatic upload is further divided into: upload via email; batch upload via URL address; upload via private cloud; direct upload via scanner docking. The type and format of the file will be automatically determined during the file upload.

[0008] The automatic classification of documents in step S1 refers to: automatically identifying the uploaded files through a pre-set document classification model, completing the classification of the files, and obtaining specific fine-grained categories of the files.

[0009] Before the document classification model identifies the file category, it also includes: Document classification model training: collect image data of different styles, perform binarization, noise reduction and tilt correction on the images, and complete the labeling of image category information; divide the labeled image data into training set, validation set and test set; use the divided training set and validation set to train the document automatic classification model, input the images to be identified into the document automatic classification model to obtain category classification results for model evaluation and optimization, and finally obtain the document classification model.

[0010] Step S2: After the file classification is completed, under the action of the corresponding category label, the corresponding information extraction module is called, the template and verification rules applicable to the current document are automatically matched, and the recognition results are structured. The key information is extracted with the support of the document template and business knowledge.

[0011] The information extraction module in step S2: automatically identifies the classified image files through a customized key information extraction model, extracts the key information that people are concerned about, and enables the rapid circulation and storage of business data; Before the key information extraction model performs key information identification and extraction, it also includes: Training of key information extraction model: collecting image data of different styles and preprocessing the images, wherein the preprocessing includes adaptive threshold binarization processing, perspective correction processing, rotation adjustment processing and image contrast enhancement processing; annotating image data, wherein the required annotation information is a rectangular box that can represent the text position, the text information corresponding to the rectangular box, and the information category corresponding to the rectangular box; dividing the annotated data into a training set, a validation set and a test set; using the divided training set and validation set to train a text detection model, inputting the test set of images to be detected into the key information extraction model to obtain inference results for model evaluation and optimization, and finally obtaining a key information extraction model.

[0012] The document template and verification rules in step S2: Since there are many styles and layouts of documents, and the key information in the documents is mostly presented in groups, such as the quantity, unit price and subtotal of a certain product need to be assigned to the same group of information. Therefore, it is necessary to pre-set a document template that conforms to the grouping information and a verification rule responsible for business knowledge verification in combination with business knowledge, and apply it in the process of extracting key information to improve the accuracy of the obtained key information and its grouping. The document template and verification rules can change with the change of the document layout.

[0013] Step S3: For the extracted key information, the next step of information flow direction is determined by the pre-set judgment conditions. When the judgment result is "yes", the extracted key information will flow to the data connection module; when the judgment result is "no", the extracted key information will flow to the auxiliary correction module, and then flow to the data connection module after manual intervention verification.

[0014] The determination conditions in step S3 include: the confidence level of the extracted information is lower than a preset threshold; the format of the extracted key information does not conform to a predetermined template or verification rule.

[0015] The auxiliary correction module in step S3: For those that do not meet the judgment conditions, that is, the judgment result is "no", the extracted key information will be transferred to the auxiliary correction module, and the position that does not meet the judgment conditions will be highlighted in the system interface, which is convenient for quick positioning and modification. Among them, low confidence does not directly mean that the recognition content is wrong. In fact, the recognition content may be correct, but because the model is not "confident" enough in recognizing specific data, the system marks a low-confidence warning. At this time, the data can be directly saved and transferred to the next module. In addition, the system will verify the recognized data according to preset formulas. These formulas may involve total calculations, percentage calculations, tax calculations, etc., to ensure that the recognition results are mathematically reasonable. If the formula verification finds a calculation error, the system will mark the erroneous data item and deduce the possible correct value based on the given data.

[0016] Step S4: The extracted data that meets the judgment conditions and the data corrected by the auxiliary correction module will be transferred to the data connection module. After entering the data connection module, the data can be connected with the third-party system to realize the subsequent automatic flow of business data; the data that has completed business adaptation will be automatically archived to cope with future audit work.

[0017] The data connection module in step S4 supports seamless connection of various business systems. The business data extracted from the document is connected to the business system to realize the automatic flow and archiving of business data. In addition, the archived data information can be presented in a variety of forms, such as quickly understanding the overall status of the file through the visualization of the report; more intuitive visualization of data statistical analysis results to help management make decisions that are conducive to the future development of the enterprise.

[0018] Step S5: The system provides a novel function, namely, the model self-learning function. Relying on the three data sources of existing training data, normal data, and correction data, the model self-learning can be performed periodically or after accumulating a certain amount of data. After the model training is completed, the model is evaluated using the reserved test data. At this time, an evaluation index N is pre-set. When the evaluation result of the new model is greater than the evaluation index, the corresponding model is updated; when the evaluation result is less than the evaluation index, the model is not updated. Through regular self-training and continuous optimization, the system can quickly adapt to the format changes of newly added recognition objects and original recognition objects.

[0019] A method, system and device for realizing intelligent circulation of business documents are made by utilizing the technical solution of the present invention. The present invention mainly realizes high-quality and efficient intelligent circulation of business documents and data through artificial intelligence technology combined with business knowledge. The use of an intelligent circulation system to assist manual work can greatly avoid the problems of input errors and low efficiency in the manual processing process, and can also avoid the problems of information retention and information islands caused by slow manual processing. At the same time, the self-learning function of the model is realized to quickly respond to the frequent changes in business scenarios in the industry, thereby reducing enterprise costs. The integration of the above methods improves the high scalability, high availability and high linkage of business data, and solves the problems existing in this specific field and existing implementation technologies and processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a block diagram of a method, system and device for realizing intelligent circulation of business documents according to the present invention; Figure 2 A method, system and device for realizing intelligent circulation of business documents, a flow for realizing the model self-learning function in step S5; DETAILED DESCRIPTION

[0021] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, a method, system and device for realizing intelligent circulation of business documents include the following steps: Step S1: Relying on the file upload module provided by the system, the file data to be processed is uploaded through a variety of transmission methods, and the document automatic classification processing will be automatically completed during the uploading process.

[0022] It should be noted that the file data can be image data in various formats such as PDF files, Word files, PNG, etc.

[0023] The various transmission methods in step S1 include manual upload and automatic upload. The automatic upload is further divided into: upload via email; batch upload via URL address; upload via private cloud; direct upload via scanner docking. The type and format of the file will be automatically determined during the file upload.

[0024] The automatic document classification in step S1 refers to: automatically identifying the uploaded files through a pre-set document classification model, completing the classification of the files, and obtaining specific fine-grained categories of the files, such as invoices, supply documents, applications, etc.

[0025] Furthermore, before the document classification model identifies the file category, it also includes: Document classification model training: collect image data of different styles, perform binarization, noise reduction and tilt correction on the images, and complete the labeling of image category information; divide the labeled image data into training set, validation set and test set; use the divided training set and validation set to train the document automatic classification model, input the images to be identified into the document automatic classification model to obtain the category classification results for model evaluation and optimization, and finally obtain the document classification model.

[0026] Step S2: After the file classification is completed, under the action of the corresponding category label, the corresponding information extraction module is called, the template and verification rules applicable to the current document are automatically matched, and the recognition results are structured. The key information is extracted with the support of the document template and business knowledge.

[0027] The information extraction module in step S2: automatically identifies the classified image files through a customized key information extraction model, extracts the key information that people are concerned about, and facilitates the rapid circulation and storage of business data; Furthermore, before the key information extraction model performs key information identification and extraction, it also includes: Training of the key information extraction model: Collect image data of different styles and preprocess the images, including adaptive threshold binarization, perspective correction, rotation adjustment and image contrast enhancement; annotate the image data, and the required annotation information is a rectangular box that can represent the text position, the text information corresponding to the rectangular box, and the information category corresponding to the rectangular box; divide the annotated data into a training set, a validation set and a test set; use the divided training set and validation set to train the text detection model, input the test set of the images to be detected into the key information extraction model to obtain the inference results for model evaluation and optimization, and finally obtain the key information extraction model.

[0028] Document templates and validation rules in step S2: Since there are many styles and layouts of documents, and the key information in the documents is mostly presented in groups, such as the quantity, unit price and subtotal of a certain product need to be assigned to the same group of information. Therefore, it is necessary to pre-set a document template that conforms to the grouping information and a validation rule responsible for business knowledge verification in combination with business knowledge, and apply it in the process of extracting key information to improve the accuracy of the obtained key information and its grouping. The document template and validation rules can change with the change of the document layout.

[0029] Step S3: For the extracted key information, the next step of information flow direction is determined by the pre-set judgment conditions. When the judgment result is "yes", the extracted key information will flow to the data connection module; when the judgment result is "no", the extracted key information will flow to the auxiliary correction module, and then flow to the data connection module after manual intervention verification.

[0030] The judgment conditions in step S3 include: the confidence level of the extracted information is lower than a preset threshold; the format of the extracted key information does not conform to the predetermined template or verification rule.

[0031] Auxiliary correction module in step S3: For key information data with a judgment result of "no" in step S3, it will be directly transferred to the auxiliary correction module, and the position that does not meet the judgment conditions will be highlighted in the system interface to facilitate quick positioning and modification. Among them, low confidence does not directly mean that the recognition content is wrong. In fact, the recognition content may be correct, but because the model is not "confident" enough in recognizing specific data, the system marks a low-confidence warning. At this time, the data can be directly saved and transferred to the next module. In addition, the system will verify the recognized data according to preset formulas. These formulas may involve total calculations, percentage calculations, tax calculations, etc., to ensure that the recognition results are mathematically reasonable. If the formula verification finds a calculation error, the system will mark the erroneous data item and deduce the possible correct value based on the given data.

[0032] Step S4: The extracted data that meets the judgment conditions and the data corrected by the auxiliary correction module will be transferred to the data connection module. After entering the data connection module, the data can be connected with the third-party system to realize the subsequent automatic flow of business data; the data that has completed business adaptation will be automatically archived to cope with future audit work.

[0033] The data connection module in step S4 supports seamless connection of various business systems. The business data extracted from the document is connected with the business system (such as SAP, CRM, etc.) to realize the automatic flow and archiving of business data. In addition, the archived data information can be presented in a variety of forms, such as quickly understanding the overall status of the file through the visualization of the report; more intuitive visualization of data statistical analysis results to help management make decisions that are conducive to the future development of the enterprise.

[0034] Step S5: The system provides a novel function, namely, the model self-learning function. Relying on the three data sources of existing training data, normal data, and correction data, the model self-learning can be performed periodically or after accumulating a certain amount of data. After the model training is completed, the model is evaluated using the reserved test data. At this time, an evaluation index N is pre-set. When the evaluation result of the new model is greater than the evaluation index, the corresponding model is updated; when the evaluation result is less than the evaluation index, the model is not updated. Through regular self-training and continuous optimization, the system can quickly adapt to the format changes of newly added recognition objects and original recognition objects.

[0035] The above technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention. Some changes that may be made to certain parts thereof by technicians in this technical field all reflect the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A method, system and device for realizing intelligent circulation of business documents, characterized in that: The following steps are involved: Step S1: Relying on the file upload module provided by the system, the file data to be processed is uploaded through a variety of transmission methods, and the document automatic classification process is automatically completed during the upload process; Step S2: After the document classification is completed, under the action of the corresponding category label, the corresponding information extraction module is called, the template and verification rules applicable to the current document are automatically matched, the recognition results are structured, and the key information is extracted with the support of the document template and business knowledge; Step S3: For the extracted key information, the next information flow direction is determined by the pre-set determination conditions; when the determination result is "yes", the extracted key information will flow to the data connection module; when the determination result is "no", the extracted key information will flow to the auxiliary correction module, and then flow to the data connection module after manual intervention verification; Step S4: The extracted data that meets the judgment conditions and the data corrected by the auxiliary correction module will be transferred to the data connection module. After entering the data connection module, the data can be connected with the third-party system to realize the subsequent automatic flow of business data; the data that has completed business adaptation will be automatically archived to cope with future audit work; Step S5: The system provides a novel function, namely, the model self-learning function; relying on the three data sources of existing training data, normal data and corrected data, the model self-learning can be performed periodically or after accumulating a certain amount of data. After the model training is completed, the model is evaluated using the reserved test data. At this time, an evaluation index N is pre-set. When the evaluation result of the new model is greater than the evaluation index, the corresponding model is updated; when the evaluation result is less than the evaluation index, the model is not updated; through regular self-training and continuous optimization, the system can quickly adapt to the format changes of newly added recognition objects and original recognition objects.

2. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The various transmission methods in step S1 include two file uploading methods: manual uploading and automatic uploading. The automatic uploading is further divided into: uploading via email; batch uploading via URL address; uploading via private cloud; direct uploading via scanner docking, wherein the type and format of the file will be automatically determined during the file uploading.

3. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The automatic classification of documents in step S1 refers to: automatically identifying the uploaded files through a pre-set document classification model, completing the classification of the files, and obtaining specific fine-grained categories of the files; Before the document classification model identifies the file category, it also includes: Document classification model training: collect image data of different styles, perform binarization, noise reduction and tilt correction on the images, and complete the labeling of image category information; divide the labeled image data into training set, validation set and test set; use the divided training set and validation set to train the document automatic classification model, input the images to be identified into the document automatic classification model to obtain category classification results for model evaluation and optimization, and finally obtain the document classification model.

4. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The information extraction module in step S2: automatically identifies the classified image files through a customized key information extraction model, extracts the key information that people are concerned about, and enables the rapid circulation and storage of business data; Before the key information extraction model performs key information identification and extraction, it also includes: Training of key information extraction model: collecting image data of different styles and preprocessing the images, wherein the preprocessing includes adaptive threshold binarization processing, perspective correction processing, rotation adjustment processing and image contrast enhancement processing; Annotate image data, where the required annotation information is a rectangular box that can represent the text location, the text information corresponding to the rectangular box, and the information category corresponding to the rectangular box; divide the annotated data into a training set, a validation set, and a test set; use the divided training set and validation set to train a text detection model, input the test set of images to be detected into the key information extraction model to obtain inference results for model evaluation and optimization, and finally obtain a key information extraction model.

5. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The document template and verification rules in step S2: There are many styles and layouts of documents, and the key information in the documents is mostly presented in groups, that is, the quantity, unit price and subtotal of a certain product need to be assigned to the same group of information; it is necessary to pre-set the document template and verification rules and apply them in the process of extracting key information to improve the accuracy of the obtained key information and its grouping.

6. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The determination conditions in step S3 include: the confidence level of the extracted information is lower than a preset threshold; the format of the extracted key information does not conform to a predetermined template or verification rule.

7. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The auxiliary correction module in step S3: For the key information data with a "no" judgment result in step S3, it will be directly transferred to the auxiliary correction module, and the position that does not meet the judgment conditions will be highlighted in the system interface, which is convenient for quick positioning and modification; Among them, low confidence does not directly mean that the recognized content is wrong; in fact, the recognized content may be correct, but because the model is not "confident" enough in recognizing specific data, the system marks a low confidence warning, and the data can be directly saved and transferred to the next module. In addition, the system will verify the recognized data according to the preset formula.

8. A method, system and device for realizing intelligent circulation of business documents according to claim 1, characterized in that: The data connection module in step S4 supports seamless connection of various business systems. The business data extracted from the document is connected with the business system to realize automatic flow and archiving of business data. In addition, the archived data information can be presented in a variety of forms.

Images