Document processing method, device and equipment based on AI, RPA and AI Agent

Abstract

The invention provides a document processing method, device and equipment based on AI, RPA and AI Agent, and relates to the field of AI, RPA and AI Agent. The method comprises the following steps: executing a document processing task to extract a key field matched with a target business scene from a to-be-processed document; obtaining an execution code of the first auditing process; wherein the execution code is generated according to a self-defined field auditing condition in the target business scene; determining a second auditing process matched with the key field from the first auditing process; executing an execution code of a second auditing process on the key field to obtain auditing information; and generating a task execution result according to the auditing information and the key field. Therefore, pertinence, rationality and accuracy of key field extraction and auditing in the target business scene can be improved, and personalized document processing requirements in different business scenes can be met. In addition, document processing can be carried out based on AI, RPA and AI Agent technologies, and the accuracy of document processing is improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence (AI), robotic process automation (RPA) and artificial intelligence agent (AI agent), and in particular to a document processing method, device and equipment based on AI, RPA and AI agent. Background Art

[0002] Robotic Process Automation (RPA) uses specific "robot software" to simulate human operations on computers and automatically execute process tasks according to rules.

[0003] Artificial Intelligence (AI) is a technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

[0004] The Artificial Intelligence Agent (AI Agent) of the digital labor platform can perceive the environment, make decisions and perform actions. The difference from traditional artificial intelligence is that it has the ability to think and act independently and can call tools to complete given goals. AI Agent is based on Large Language Models (LLM) as its core computing engine, which enables it to have conversations, perform tasks, reason and show a certain degree of autonomy. It has the ability to understand, perceive, plan, remember and use tools independently, and can automatically perform complex tasks. Specifically, AI Agent, an intelligent assistant composed of multiple AI capabilities driven by LLM, can interact with employees in natural language, understand employees' instructions and needs, and provide feedback and responses; it can master domain knowledge related to the business to complete complex professional work; it can decompose complex tasks into several executable tasks and use data and tools to complete tasks; it can also collaborate with employees, and AI Agents can also collaborate to complete complex tasks, so as to enable digital employees to achieve a leap from automation to intelligence, help employees complete their work more efficiently, and fully realize human-machine collaboration.

[0005] With the rapid development of informatization, the importance of document processing and information extraction in enterprise automation processes has become increasingly prominent. The traditional method of manually reading, analyzing and extracting key fields from documents is not only inefficient, but also prone to errors and difficult to meet the needs of large-scale data processing. In order to overcome these challenges, machine learning methods in AI technology have emerged, providing a new solution for automatically extracting key fields required by enterprises from documents.

[0006] Specifically, the machine learning model is trained with labeled data, and the trained machine learning model is used to extract key fields from the document. A confidence score is returned for the key fields. The confidence score of each key field is determined to be greater than the confidence score threshold to review whether the key field is credible. If so, the key field is considered to have passed the review and is retained.

[0007] However, the confidence threshold is usually set in advance manually and is not universal and cannot be applied to different business scenarios. Summary of the invention

[0008] The embodiment of the present application provides a document processing method, apparatus and device based on AI, RPA and AI Agent to solve one of the technical problems existing in the related technology. The technical solution is as follows:

[0009] In a first aspect, an embodiment of the present application provides a document processing method based on AI, RPA, and AI Agent, including:

[0010] Execute a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task;

[0011] Obtaining at least one execution code of a first review process adapted to a target business scenario; wherein the execution code is generated according to a customized field review condition under the target business scenario;

[0012] Determine a second audit process matching any key field from the first audit process, and execute the execution code of the second audit process on any key field to obtain the first audit information;

[0013] Generate a task execution result based on the first audit information and each key field.

[0014] In a second aspect, an embodiment of the present application provides a document processing device based on AI, RPA, and AI Agent, including:

[0015] An extraction module, used to execute a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task;

[0016] A first acquisition module, used to acquire an execution code of at least one first review process adapted to a target business scenario; wherein the execution code is generated according to a customized field review condition under the target business scenario;

[0017] An execution module, used for determining a second audit process matching any key field from the first audit process, and executing an execution code of the second audit process on any key field to obtain first audit information;

[0018] The first generating module is used to generate a task execution result according to the first review information and each key field.

[0019] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising: a memory and a processor. The memory and the processor communicate with each other through an internal connection path, the memory is used to store instructions, the processor is used to execute the instructions stored in the memory, and when the processor executes the instructions stored in the memory, the processor executes the method in any one of the above-mentioned embodiments.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a computer, the method in any one of the above-mentioned embodiments is executed.

[0021] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which implements the method in any one of the above-mentioned embodiments when executed by a processor.

[0022] The advantages or beneficial effects of the above technical solution include at least:

[0023] For different business scenarios, the field review conditions adapted to each business scenario are configured in a targeted manner, and the execution code of the review process adapted to the business scenario is generated based on the field review conditions. Therefore, when processing any pending document, the target business scenario to which the pending document belongs is first determined. After that, only the key fields in the target business scenario can be extracted from the pending document, and the extracted key fields can be reviewed based on the execution code of the review process adapted to the target business scenario. This can improve the pertinence, rationality and accuracy of the key field extraction + review in the target business scenario, and meet the personalized document processing needs in different business scenarios. In addition, document processing can be performed based on AI, RPA and AIAgent technologies to improve the accuracy of document processing.

[0024] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present application will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0026] Figure 1 is a flowchart of a document processing method based on AI, RPA and AI Agent provided by an embodiment of the present application;

[0027] Figure 2 is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application;

[0028] Figure 3 is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application;

[0029] Figure 4 is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application;

[0030] Figure 5 is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application;

[0031] Figure 6 It is a schematic diagram of the review process of a field provided by an embodiment of the present application;

[0032] Figure 7 This is a schematic diagram of the key field extraction + review + human-computer interaction process provided by an embodiment of the present application;

[0033] Figure 8 is a structural diagram of a document processing device based on AI, RPA and AI Agent provided by an embodiment of the present application;

[0034] Fig. 9 A structural block diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0035] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0036] In related technologies, the following two methods are mainly used to extract key fields from documents:

[0037] The first type is rule and template matching: for some documents with special formats, predefined rules and templates are used to determine whether the document content meets the requirements. These rules are usually written manually by technicians and are only applicable to specific fields.

[0038] The second method is machine learning: use labeled data to train machine learning models, use the trained machine learning models to identify and extract key fields in documents, and return a confidence score for each key field. By judging whether the confidence score of each key field is greater than the confidence threshold, it is determined whether the key field is credible. If it is credible, the key field is retained. Among them, the confidence threshold is usually set in advance by humans and is not universal.

[0039] However, the first method mentioned above has at least the following disadvantages:

[0040] First, poor flexibility: Rules and templates are usually written manually by technicians and are only applicable to documents in specific fields and formats. When the document format or the field to which the document belongs changes, the rules need to be rewritten or adjusted.

[0041] Second, high maintenance costs: As data volume and complexity increase, maintaining and updating rules becomes increasingly difficult and time-consuming.

[0042] The third is poor scalability: it is difficult to handle unforeseen new types of information or complex language structures.

[0043] Fourth, limited coverage: Rules and templates may not cover all possible situations, resulting in limited accuracy and completeness of key field extraction.

[0044] The second method mentioned above has at least the following problems:

[0045] First, the selection of the confidence threshold is highly subjective: the confidence threshold is usually set manually and lacks objective standards. Different application scenarios and data sets may require different confidence thresholds, making it difficult to find a universal confidence threshold.

[0046] Second, lack of dynamic adjustment capability: fixed confidence thresholds cannot adapt to different input data and environmental changes. Although the method of dynamically adjusting the confidence threshold may be more effective, it is more complicated to implement.

[0047] The third one is sensitivity to imbalanced data: On an imbalanced dataset, the confidence threshold may lead to bias. For example, on minority class samples, the machine learning model may give a lower confidence score, causing these samples to be ignored.

[0048] In order to solve at least one of the above-mentioned problems, the embodiments of the present application propose a document processing method, device and equipment based on artificial intelligence (AI), robotic process automation (RPA) and digital labor platform intelligent assistant (AIAgent). These and other aspects of the embodiments of the present application will be clear with reference to the following description and drawings. In these descriptions and drawings, some specific implementations in the embodiments of the present application are specifically disclosed to represent some ways of implementing the principles of the embodiments of the present application, but it should be understood that the scope of the embodiments of the present application is not limited thereto. On the contrary, the embodiments of the present application include all changes, modifications and equivalents that fall within the spirit and connotation of the appended claims.

[0049] Before describing the embodiments of the present application in detail, in order to facilitate understanding, common technical terms are first introduced:

[0050] In the description of the present application, the term "plurality" refers to two or more.

[0051] In the description of this application, the term "OCR" is the abbreviation of Optical Character Recognition, which specifically refers to the process in which an electronic device examines characters printed on paper, determines their shapes by detecting dark and light patterns, and then translates the shapes into computer text using character recognition methods; that is, for printed characters, an optical method is used to convert the text in a paper document into a black and white dot matrix image file, and recognition software is used to convert the text in the image into a text format for further editing and processing by word processing software.

[0052] In the description of this application, the term "document processing task" refers to a task to be performed, which is used to process a document to be processed.

[0053] The "undecided document" refers to a document that needs to be processed, which may be a long document or a short document, and this application does not limit this. For example, the unprocessed document may be a document generated by an enterprise when performing its business functions and providing services, including but not limited to: contracts, reports, customer information, transaction records and other documents.

[0054] There is no restriction on the document format of the document to be processed. For example, the document format includes but is not limited to: Word, Portable Document Format (PDF) and other formats.

[0055] In the description of this application, the term "key field" refers to the key information that needs to be extracted in document processing. Taking the document to be processed as a contract document as an example, the key fields may include Party A, Party B, payment method, payment amount, etc.

[0056] In the description of this application, the term "target business scenario" refers to the business scenario to which the document to be processed belongs. Taking the document to be processed as a contract document as an example, the target business scenario may be a contract scenario.

[0057] In the description of this application, the term "field review condition" refers to the review condition customized by relevant personnel for key fields in the target business scenario, which is used to review the key fields.

[0058] In the description of this application, the term "review process" refers to a process generated based on field review conditions, wherein each review process includes at least one process link, and each process link is used to review key fields.

[0059] For example, taking the target business scenario as the contract scenario and the field review condition as "Party B must be a corporate entity and have no violation record within 2 years", the review process generated based on this field review condition (marked as Review Process 1) includes 2 process links, one of which is used to query whether Party B in the contract document is a corporate entity, and the other process link is used to query whether Party B in the contract document has any violation record within the past 2 years.

[0060] For another example, taking the target business scenario as a contract scenario and the field review condition as "the writing of the amount in capital letters meets the requirements", the review process (marked as review process 2) generated according to the field review condition includes 1 process link, which is used to query whether the writing of various amounts in capital letters in the contract document meets the requirements.

[0061] In the description of this application, the term "first reference example" may also be referred to as a "positive example", which is used to describe key fields that meet field review conditions.

[0062] In the description of this application, the term "second reference example" may also be referred to as a "counter-example", which is used to describe key fields that do not meet the field review conditions.

[0063] These and other aspects of the embodiments of the present application will be apparent with reference to the following description and accompanying drawings. In these descriptions and accompanying drawings, some specific implementations in the embodiments of the present application are specifically disclosed to represent some ways of implementing the principles of the embodiments of the present application, but it should be understood that the scope of the embodiments of the present application is not limited thereto. On the contrary, the embodiments of the present application include all changes, modifications and equivalents that fall within the spirit and connotation of the appended claims.

[0064] The following describes the document processing method, device and equipment based on AI, RPA and AI Agent according to the embodiments of the present application in conjunction with the accompanying drawings.

[0065] Figure 1 This is a flowchart of a document processing method based on AI, RPA and AI Agent provided by an embodiment of the present application.

[0066] In a possible implementation method provided in an embodiment of the present application, the present application takes the document processing method based on AI, RPA and AI Agent as an example, and the document processing device based on AI, RPA and AI Agent is configured. The document processing device can be applied to any electronic device with computing capabilities.

[0067] The electronic device may be a personal computer, a mobile terminal, a service end (or server), etc. The mobile terminal may be, for example, a mobile phone, a tablet computer, a personal digital assistant, or other hardware device with various operating systems.

[0068] In another possible implementation of the embodiment of the present application, the document processing method based on AI, RPA and AI Agent can be applied to AI Agent, wherein the AI ​​Agent can run in any electronic device with computing capabilities.

[0069] In another possible implementation of the embodiment of the present application, the document processing method based on AI, RPA and AI Agent can be applied to a Robotic Process Automation (RPA) robot, wherein the RPA robot can run in any electronic device with computing capabilities.

[0070] In another possible implementation of the embodiment of the present application, the document processing method based on AI, RPA and AI Agent can be applied to an intelligent automation platform; wherein the intelligent automation platform can realize seamless integration of multiple capabilities such as RPA, intelligent document processing (IDP), conversational AI (CoAI), and process mining (Process Mining), and has five major functions of "business understanding", "process creation", "run anywhere", "centralized control", and "human-machine collaboration", which can realize end-to-end intelligent automation of business processes for enterprises, replace manual operations, further improve business efficiency, and accelerate digital transformation.

[0071] Among them, intelligent document processing (IDP) is one of the core capabilities of the intelligent automation platform. Intelligent document processing (IDP) is a new generation of automation technology based on AI technologies such as optical character recognition (OCR), computer vision (CV), natural language processing (NLP), and knowledge graph (KG). It can identify, classify, extract elements, verify, compare, and correct errors of various documents, helping enterprises realize the intelligence and automation of document processing.

[0072] like Figure 1 As shown, the document processing method based on AI, RPA and AI Agent may include the following steps S101 to S104:

[0073] Step S101: executing a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task.

[0074] In the embodiment of the present application, a document processing task can be obtained. For example, the document processing task can be a task submitted by the human-machine collaboration center. In the present application, the document processing task sent by the human-machine collaboration center can be received and executed to extract various key fields adapted to the target business scenario from the document to be processed in the document processing task. For example, taking the target business scenario as a contract scenario as an example, the key fields can be key fields such as Party A, Party B, payment method, and payment amount in the document to be processed.

[0075] Step S102: Obtain execution code of at least one first review process adapted to the target business scenario.

[0076] Among them, the execution code of the first review process is generated according to the customized field review conditions in the target business scenario.

[0077] It should be noted that the audit requirements for fields in different business scenarios may be different. If the same field audit conditions are used to audit key fields in different business scenarios, the accuracy of the key field audit will be reduced and the personalized application requirements in different business scenarios will not be met.

[0078] To address this issue, in this application, in order to improve the accuracy and pertinence of field audits, relevant personnel (such as staff, business personnel, etc.) can set field audit conditions for each business scenario respectively for different business scenarios, and generate the corresponding audit process execution code based on each field audit condition. Afterwards, the association relationship between the business scenarios, field audit conditions, audit processes, and the audit process execution code can be established.

[0079] Therefore, in this application, the above-mentioned correspondence can be queried according to the target business scenario to determine the execution code of the review process (referred to as the first review process in this application) associated with or adapted to the target business scenario.

[0080] Step S103: determining a second audit process matching any key field from the first audit process, and executing the execution code of the second audit process on any key field to obtain first audit information.

[0081] The first audit information is used to indicate whether each key field has passed the audit of the corresponding second audit process.

[0082] In an embodiment of the present application, for any key field, a second review process adapted to the key field can be determined from each first review process. For example, the key field can be Party B in the document to be processed, and the second review process can be the review process 1 mentioned above.

[0083] In an embodiment of the present application, the execution code of the second audit process can be executed on the above-mentioned key fields to obtain the first audit information; wherein the first audit information is used to indicate whether each key field has passed the audit of the corresponding second audit process.

[0084] Step S104: Generate a task execution result according to the first audit information and each key field.

[0085] In the embodiment of the present application, the task execution result can be generated according to the first audit information and each key field.

[0086] The document processing method based on AI, RPA and AI Agent in the embodiment of the present application configures field review conditions that are adapted to each business scenario for different business scenarios, and generates execution code for the review process that is adapted to the business scenario based on the field review conditions, so that when processing any document to be processed, the target business scenario to which the document to be processed belongs is first determined, and then only the key fields in the target business scenario can be extracted from the document to be processed, and the extracted key fields can be reviewed according to the execution code of the review process that is adapted to the target business scenario, which can improve the pertinence, rationality and accuracy of the key field extraction + review in the target business scenario and meet the personalized document processing needs in different business scenarios. In addition, document processing can be performed based on AI, RPA and AI Agent technologies to improve the accuracy of document processing.

[0087] In order to clearly explain how any embodiment of the present application generates a task execution result based on the first review information and various key fields, the present application also proposes a document processing method based on AI, RPA and AI Agent.

[0088] Figure 2 This is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application.

[0089] It should be noted that the document processing method based on AI, RPA and AI Agent can be executed alone, or it can be executed in combination with any embodiment in the present application or a possible implementation method in the embodiment, or it can be executed in combination with any technical solution in the relevant technology, and the embodiments of the present application are not limited to this.

[0090] like Figure 2 As shown, the document processing method based on AI, RPA and AI Agent may include the following steps S201 to S210:

[0091] Step S201: executing a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task.

[0092] Step S202: Obtain execution code of at least one first review process adapted to the target business scenario.

[0093] Among them, the execution code of the first review process is generated according to the customized field review conditions in the target business scenario.

[0094] Step S203: determining a second audit process matching any key field from the first audit process, and executing the execution code of the second audit process on any key field to obtain first audit information.

[0095] The first audit information is used to indicate whether each key field has passed the audit of the corresponding second audit process.

[0096] The explanation of steps S201 to S203 can be found in the relevant description in any embodiment of the present application and will not be repeated here.

[0097] Step S204: According to the first audit information, determine whether the first target field exists in each key field. If so, execute steps S206 to S210; if not, execute step S205.

[0098] The first target field refers to a key field that has not passed the review of the corresponding second review process.

[0099] The number of the first target fields may be one or more, and this application does not impose any limitation on this.

[0100] It should be noted that step S205 and steps S206 to S210 are two parallel implementation methods. In actual application, only one needs to be executed.

[0101] Step S205: Generate task execution results based on the key fields that have passed the review.

[0102] In the embodiment of the present application, when the first audit information indicates that all key fields have passed the audit, the task execution result can be directly generated based on the key fields that have passed the audit. In other words, when all key fields have passed the audit, the execution of the document processing task can be terminated to obtain the task execution result.

[0103] Step S206: marking the target content segment in the document to be processed, and determining the reason why the first target field fails the review; wherein the target content segment is the content segment where the first target field is located.

[0104] In an embodiment of the present application, when the first review information indicates that there is a first target field in each key field that has not passed the review, the target content segment in the document to be processed can be further marked (such as highlighting, special color marking, etc.); wherein the target content segment is the content segment where the first target field is located.

[0105] In the embodiment of the present application, the reason why the first target field failed to pass the review can also be determined. For example, taking the first target field as the relevant amount in the contract as an example, the reason can be "the lowercase amount is inconsistent with the uppercase amount".

[0106] Step S207: Generate a modification suggestion for the first target field according to the reason, and generate a review task according to the modification suggestion and the annotated document to be processed.

[0107] In an embodiment of the present application, a modification suggestion for each first target field can be generated based on the reason why each first target field failed the review. Still taking the above example as an example, the modification suggestion may be "Please modify the lowercase amount or the uppercase amount to keep the two consistent."

[0108] In an embodiment of the present application, an audit task may also be generated based on the modification suggestions for the first target field and the annotated documents to be processed, that is, the audit task includes both the modification suggestions for the first target field and the annotated documents to be processed.

[0109] Step S208: dispatching the review task to the target review object, so that the target review object modifies the target content segment in the marked document to be processed based on the modification suggestion.

[0110] The target audit object refers to a user object with audit authority. For example, taking the document processing task as a task submitted by the human-machine collaboration center as an example, the target audit object may be a relevant auditor on the human-machine collaboration center side.

[0111] In an embodiment of the present application, the audit task can be scheduled to the target audit object so that the target audit object performs the audit task to modify the target content segment marked in the document to be processed based on the modification suggestion to obtain a modified task to be processed.

[0112] Step S209: receiving the modified document to be processed sent by the target audit object in response to the audit task.

[0113] In an embodiment of the present application, a modified document to be processed sent by the target audit object in response to the audit task may be received.

[0114] Step S210: Generate a task execution result based on the modified document to be processed and the key fields that have passed the review.

[0115] In an embodiment of the present application, the modified documents to be processed and the key fields that have passed the review can be integrated to generate a task execution result.

[0116] In any embodiment of the present application, the task execution result can be generated by following steps A to E:

[0117] Step A: extracting a modified target content segment from the modified document to be processed.

[0118] As an example, the target content segment is marked in the document to be processed, and the modified target content segment is also marked in the document to be processed. Based on the above marks, the modified target content segment can be extracted from the modified document to be processed.

[0119] Step B: extracting a second target field adapted to the target business scenario from the modified target content segment, that is, the second target field is the modified first target field.

[0120] The method of extracting the second target field is similar to the method of extracting the key field, and will not be described in detail here.

[0121] Step C: Determine a third audit process matching the second target field from each first audit process.

[0122] Among them, the method for determining the third audit process matching the second target field is similar to the method for determining the second audit process matching the key field, and will not be described in detail here.

[0123] Step D: Execute the execution code of the third audit process on the second target field to obtain second audit information; wherein the second audit information is used to indicate whether the second target field has passed the audit of the third audit process.

[0124] It should be noted that the implementation principle of step D is similar to that of step S103, which will not be described in detail here.

[0125] Step E: When the second audit information indicates that the second target field has passed the audit, a task execution result is generated according to the second target field that has passed the audit and the key field that has passed the audit. That is, when all the key fields have passed the audit, the execution of the document processing task can be terminated to obtain the task execution result.

[0126] In summary, when all key fields have passed the review, the execution of the document processing task is terminated and the task execution result is obtained. This can ensure the accuracy of the key fields finally obtained in the task execution result and meet the actual field extraction requirements in the target business scenario.

[0127] The document processing method based on AI, RPA and AI Agent in the embodiment of the present application, when the extracted key fields fail to pass the review, dispatches human resources to manually review and modify the documents to be processed, and re-extracts and reviews the key fields of the modified documents to be processed, which can further improve the accuracy of the key fields finally obtained in the task execution results and meet the actual field extraction requirements in the target business scenario. In addition, the review task carries modification suggestions and documents marked with key fields to be modified. The modification suggestions are generated based on the reasons why the key fields failed the review. Human resources can directly locate, review and modify the key fields to be modified in the document based on the modification suggestions, which can improve the review efficiency of human resources.

[0128] In order to clearly explain how to generate the execution code of the first review process adapted to the target business scenario in any embodiment of the present application, the present application also proposes a document processing method based on AI, RPA and AI Agent.

[0129] Figure 3 This is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application.

[0130] like Figure 3 As shown, based on any embodiment of the present application, the execution code of the first audit process can be generated by following steps S301 to S304:

[0131] Step S301: Acquire at least one natural language text input in response to a condition configuration operation; wherein the natural language text is used to indicate a field review condition in a target business scenario.

[0132] Among them, natural language text is manually input by relevant personnel, and the input methods include but are not limited to: touch input (such as sliding, clicking, etc.), keyboard input, voice input, etc.

[0133] There is a one-to-one correspondence between natural language text and field review conditions, that is, each natural language text is used to indicate a field review condition.

[0134] In an embodiment of the present application, in response to a conditional configuration operation triggered by a relevant person, natural language text input by a relevant person can be obtained, wherein each natural language text is used to indicate a field review condition in a target business scenario.

[0135] Step S302: Generate first prompt information according to the natural language text; wherein the first prompt information is used to prompt the large language model to execute a process generation task.

[0136] In an embodiment of the present application, for any natural language text, a first prompt information (prompt) can be generated based on the natural language text; wherein the first prompt information is used to prompt the large language model for task information to be executed, that is, to prompt the large language model to execute a process to generate a task.

[0137] In any embodiment of the present application, the first prompt information may be generated by following steps a to b:

[0138] Step a: Obtain at least one reference example adapted to the natural language text.

[0139] Among them, the reference examples may include a first reference example (or positive example) and / or a second reference example (or negative example), the first reference example includes key fields that meet the field review conditions indicated by the natural language text, and the second reference example includes key fields that do not meet the field review conditions indicated by the natural language text.

[0140] It should be noted that the manually configured field review conditions in natural language form may be too colloquial and broad, which may result in the large language model being unable to understand the user's actual intentions. To address this issue, in this application, reference examples can be supplemented for the large language model. The reference examples can be used as prior information to guide the learning of the large language model, thereby improving the prediction accuracy of the large language model.

[0141] As an example, at least one reference example adapted to the natural language text in the target business scenario may be manually provided by relevant personnel.

[0142] Step b: Generate first prompt information based on the reference example and the natural language text. For example, the reference example and the natural language text may be assembled to obtain the first prompt information.

[0143] In summary, by introducing positive and negative examples in the target business scenarios for the large language model, not only can the understanding ability of the large language model in the target business scenarios be improved, but the positive and negative examples also provide a clear contrast, allowing the large language model to more clearly understand which situations are reasonable and which are unreasonable. This helps to reduce errors in the prediction and generation process of the large language model and improve the accuracy, robustness, and generalization ability of the model.

[0144] Step S303: calling the large language model to process the first prompt information to obtain a link code of at least one process link in the first review process.

[0145] In an embodiment of the present application, a large language model may be called to process the first prompt information to obtain a link code for each process link in the first review process output by the large language model.

[0146] Step S304: Generate the execution code of the first review process according to the link code of each process link.

[0147] In the embodiment of the present application, the execution code of the first audit process can be generated according to the link code of each process link in the first audit process.

[0148] In any embodiment of the present application, in order to further improve the accuracy and reliability of the review of each key field in the target business scenario, after generating the execution code of each first review process, a testing process can also be performed to evaluate the generation quality of the execution code of the first review process.

[0149] As an example, the testing process may include the following steps H to J:

[0150] Step H: Use the execution code of the first audit process to test the sample key field to obtain the test result. That is, the execution code of the first audit process can be executed on the sample key field to obtain the test result.

[0151] The test result is used to indicate whether the key fields of the sample have passed the review of the first review process in the test scenario.

[0152] Among them, the sample key fields refer to the key fields in the target business scenario. The sample key fields can be key fields extracted from documents in the target business scenario, or they can be key fields manually entered, or they can be key fields collected online, etc. This application does not impose any restrictions on this.

[0153] Step I: Determine whether the test result meets the set conditions. If so, it indicates that the generation quality of the execution code of the first audit process is high. At this time, the execution code of the first audit process can be directly retained; if not, continue to execute the subsequent step J.

[0154] Among them, the setting conditions include that the test results are consistent with the annotation results of the sample key fields.

[0155] The annotation results are used to indicate whether the key fields of the sample have passed the review in the actual target business scenario.

[0156] Step J: If the test result does not meet the set conditions, that is, the test result is inconsistent with the annotation result of the sample key field, it indicates that the generation quality of the execution code of the first review process is poor. At this time, the natural language text can be updated until the test result meets the set conditions, that is, the updated natural language text can be used to re-execute steps S302, S303, S304, step H, step I and step J until the test result is consistent with the annotation result of the sample key field.

[0157] In summary, testing the execution code of the first audit process and adjusting the natural language text to regenerate the execution code of the first audit process when the test results do not meet expectations has significant benefits in ensuring data accuracy, improving system stability, enhancing user experience, reducing maintenance costs, and promoting compliance.

[0158] In the document processing method based on AI, RPA and AI Agent in the embodiment of the present application, the prompt information can be used as prior information to prompt the large language model of task information to be executed, which can improve the prediction accuracy of the large language model, that is, improve the accuracy of link code generation in each process link in the first review process, and then improve the accuracy of execution code generation in the first review process.

[0159] In order to clearly explain how to extract various key fields adapted to the target business scenario from the document to be processed in any embodiment of the present application, the present application also proposes a document processing method based on AI, RPA and AI Agent.

[0160] Figure 4 This is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application.

[0161] It should be noted that the document processing method based on AI, RPA and AI Agent can be executed alone, or it can be executed in combination with any embodiment in the present application or a possible implementation method in the embodiment, or it can be executed in combination with any technical solution in the relevant technology, and the embodiments of the present application are not limited to this.

[0162] like Figure 4 As shown, the document processing method based on AI, RPA and AI Agent may include the following steps S401 to S407:

[0163] Step S401: executing a document processing task to perform character recognition on a document to be processed in the document processing task based on OCR technology to obtain document content.

[0164] It should be noted that, considering that enterprises have to deal with a variety of document formats in daily work, such as Word, PDF and other different document formats, even the same type of document format may have multiple different types, such as image-type PDF, text-type PDF, etc. If each document format is processed separately, it will generate a huge workload and limit the scalability of the entire system.

[0165] In response to the above problems, in this application, OCR technology can be used to identify the text content in the document to be processed, so that different document formats can be ignored. While simplifying the complexity of the system, it also enhances the compatibility and scalability of the entire system.

[0166] Step S402: Acquire a prompt template adapted to the target business scenario; wherein the prompt template is used to indicate field information to be extracted by the large language model in the target business scenario.

[0167] The target business scenario refers to the business scenario to which the document to be processed belongs.

[0168] Among them, the prompt templates adapted to different business scenarios may be different, and the prompt templates adapted to each business scenario are pre-configured to indicate the field information to be extracted by the large language model in the corresponding business scenario.

[0169] It should be noted that, in actual application, the prompt template adapted for each business scenario can also be dynamically maintained and updated according to actual business needs to meet dynamically changing business needs in different business scenarios.

[0170] Taking the target business scenario as an example of a contract scenario, a prompt template adapted to the contract scenario can be used to instruct the large language model to extract field information such as Party A, Party B, payment method, and payment amount.

[0171] In an embodiment of the present application, prompt templates corresponding to different business scenarios can be pre-configured, and a correspondence between the business scenarios and the prompt templates can be established. Therefore, in the present application, the above correspondence can be queried according to the target business scenario to obtain a prompt template adapted to the target business scenario.

[0172] Step S403: Fill the prompt template according to the document content to obtain second prompt information.

[0173] In an embodiment of the present application, the above prompt template may be filled with document content to obtain second prompt information, wherein the second prompt information is used to instruct the large language model to perform a field extraction task.

[0174] Step S404: calling the large language model to process the second prompt information to obtain at least one key field adapted to the target business scenario.

[0175] In an embodiment of the present application, a large language model may be called to process the second prompt information to obtain various key fields output by the large language model that are adapted to the target business scenario.

[0176] Step S405: Obtain execution code of at least one first review process adapted to the target business scenario.

[0177] The execution code is generated based on the customized field review conditions in the target business scenario.

[0178] Step S406: Determine a second audit process that matches any key field from the first audit process, and execute the execution code of the second audit process on any key field to obtain first audit information.

[0179] Step S407: Generate a task execution result according to the first audit information and each key field.

[0180] The explanation of steps S405 to S407 can be found in the relevant description in any embodiment of the present application and will not be repeated here.

[0181] The document processing method based on AI, RPA and AI Agent in the embodiment of the present application can guide the large language model to understand business needs more accurately by designing special prompt templates for different business scenarios, thereby accurately extracting the required field information and improving the accuracy and efficiency of field information extraction. In addition, by configuring different prompt templates for different business scenarios, the large language model can easily adapt to various complex business environments. This flexibility enables the large language model to be widely used in multiple fields and industries, enhancing the adaptability and flexibility of the model.

[0182] In order to clearly explain how to extract various key fields adapted to the target business scenario from the document to be processed in any embodiment of the present application, the present application also proposes a document processing method based on AI, RPA and AI Agent.

[0183] Figure 5 This is a flowchart of a document processing method based on AI, RPA and AI Agent provided in another embodiment of the present application.

[0184] It should be noted that the document processing method based on AI, RPA and AI Agent can be executed alone, or it can be executed in combination with any embodiment in the present application or a possible implementation method in the embodiment, or it can be executed in combination with any technical solution in the relevant technology, and the embodiments of the present application are not limited to this.

[0185] like Figure 5 As shown, the document processing method based on AI, RPA and AI Agent may include the following steps S501 to S507:

[0186] Step S501: executing a document processing task, and performing character recognition on a document to be processed in the document processing task based on OCR technology to obtain document content.

[0187] For the explanation of step S501, reference may be made to the relevant description in the above embodiment, which will not be elaborated here.

[0188] Step S502: Generate an input sequence according to the document content.

[0189] As an example, each character in the document content may be written into the input sequence in sequence.

[0190] Step S503: Use a sequence labeling model to label the input sequence to obtain a labeled sequence.

[0191] Among them, each label in the annotation sequence is used to indicate whether the character at the corresponding position in the input sequence belongs to the key field.

[0192] Among them, the sequence labeling model is trained using the corpus (i.e., sample text) in the target business scenario.

[0193] Exemplarily, the following steps L to O may be used to train the sequence labeling model.

[0194] Step L: Obtain sample text in the target business scenario and generate a sample input sequence based on the sample text.

[0195] Step M: Obtain a marked label sequence corresponding to the sample input sequence; wherein each label in the marked label sequence is used to indicate whether the character at the corresponding position in the sample input sequence belongs to a key field in the target business scenario.

[0196] As a possible implementation method, the labeled label sequence can be obtained by manually labeling the sample input sequence, or it can be obtained by labeling using a high-precision deep learning model, and this application does not impose any restrictions on this.

[0197] Step N: Use the sequence labeling model to label the sample input sequence to obtain the predicted label sequence.

[0198] Among them, each label in the predicted label sequence is used to indicate whether the character at the corresponding position in the sample input sequence belongs to the key field in the target business scenario.

[0199] Step O: Train the sequence labeling model based on the difference between the predicted label sequence and the labeled label sequence.

[0200] As an example, the value of the loss function (referred to as the loss value in this application) can be determined based on the difference between the predicted label sequence and the labeled label sequence, wherein the loss value is positively correlated with the above-mentioned difference. Thus, in this application, the sequence labeling model can be trained based on the loss value to minimize the loss value.

[0201] It should be noted that the above only uses the termination condition of model training as the minimization of the loss value as an example. In actual application, other termination conditions can also be set. For example, the termination conditions can also include: the training time reaches the set time, the training rounds reach the set rounds, etc. This application does not limit this.

[0202] In an embodiment of the present application, a trained sequence labeling model may be used to label an input sequence to obtain a labeled sequence.

[0203] Step S504: extract at least one key field adapted to the target business scenario from the input sequence according to the marked sequence.

[0204] In an embodiment of the present application, at least one key field adapted to the target business scenario can be extracted from the input sequence according to the marked sequence.

[0205] As an example, assuming that the sequence labeling model uses the BIO (B represents Begin (i.e., the beginning of the key field), I represents Inside (i.e., the middle or end of the key field), and O represents Outside (not belonging to the key field or entity)) labeling method to label the input sequence, the key field can be determined based on the continuous labels containing only B and I in the labeling sequence.

[0206] As another example, assuming that the sequence labeling model uses the BIOES (B stands for Begin (i.e., the beginning of the key field), I stands for Inside (i.e., the middle of the key field), E stands for End (i.e., the end position of the key field), O stands for Outside (not belonging to the key field, used to mark irrelevant characters), and S stands for Single (a single character, which is itself a key field or entity)) labeling method to label the input sequence, the key field can be determined based on the continuous labels containing only B, I and E in the labeling sequence, and / or the key field can be determined based on S in the labeling sequence.

[0207] Step S505: Obtain execution code of at least one first review process adapted to the target business scenario.

[0208] Among them, the execution code of the first review process is generated according to the customized field review conditions in the target business scenario.

[0209] Step S506: Determine a second audit process that matches any key field from the first audit process, and execute the execution code of the second audit process on any key field to obtain first audit information.

[0210] Step S507: Generate a task execution result according to the first audit information and each key field.

[0211] The execution process of steps S505 to S507 can refer to the relevant description in any embodiment of the present application and will not be repeated here.

[0212] In a first possible implementation method of any embodiment of the present application, the task execution result can also be sent to a designated device, which can ensure the timeliness and timeliness of the task execution result synchronization to relevant personnel, so that relevant personnel can view the task execution results in a timely manner through the designated device, and execute subsequent business processes based on the task execution results, thereby promoting the execution progress of the business.

[0213] In a second possible implementation method of any embodiment of the present application, the task execution result can also be sent to a designated account, which can ensure the timeliness and timeliness of the task execution result synchronization to the relevant personnel, so that the relevant personnel bound to the designated account can view the task execution results in time through the terminal devices they use, so as to execute subsequent business processes based on the task execution results, thereby promoting the execution progress of the business.

[0214] In a third possible implementation method of any embodiment of the present application, the task execution results can also be persistently stored so that relevant personnel can conduct data queries in a timely and effective manner, thereby executing subsequent business processes based on the query results, thereby promoting the execution progress of the business.

[0215] It should be noted that, in actual application, the above-mentioned multiple possible implementation methods can also be executed simultaneously, and the embodiments of the present application do not limit this.

[0216] In summary, different methods can be used to provide task execution results to relevant personnel, which can improve the flexibility and applicability of the method and meet personalized application needs in different scenarios.

[0217] The document processing method based on AI, RPA and AI Agent in the embodiment of the present application uses the corpus in the target business scenario to train the sequence annotation model, which can improve the accuracy of the field information extracted by the sequence annotation model in the target business scenario.

[0218] With the popularity of large language models, large language models are increasingly used in document processing. Since large language models may have problems such as hallucinations and unstable output, it is impossible to fully trust the results generated by large language models, which greatly limits the application of large language models in document processing. In order to ensure the accuracy of the work, manual assistance may be required when using large language models, which is contrary to the use of large language models to improve work efficiency. Based on this, in this application, intelligent review based on human-computer interaction is introduced to reduce manual intervention in document processing by large language models, improve work efficiency, and give full play to the advantages of large language models in document processing.

[0219] This application aims to solve the following problems: 1. Use AI Agent to determine which key fields require manual review and give review suggestions (recorded as modification suggestions in this application); 2. Use human-computer collaboration to complete the closed loop of human-computer interaction.

[0220] As an example, before officially running the document processing process, first of all, you need to set the audit requirements for the fields in a document processing process. After that, for the document processing tasks submitted by the human-machine collaboration center, after completing the extraction of key fields in the document, each key field will be audited using the AI ​​Agent. Key fields that fail the audit will be returned to the human-machine collaboration center to enter the manual processing process. Key fields that pass the audit will directly enter the subsequent business process without manual processing. The implementation principles of this application mainly include the following parts:

[0221] Part 1: Setting audit requirements.

[0222] Different from the traditional rule system, the manually set audit requirements are in the form of natural language, which greatly reduces the cost of use. For example, the field audit conditions can be as follows:

[0223] Example 1: Party B must be a corporate entity and have no violation record within 2 years;

[0224] Example 2: The amount in capital letters meets the requirements;

[0225] Example 3: The lowercase amount must be consistent with the uppercase amount.

[0226] It can be seen that for the setting of field review conditions, there is no need to master professional Internet Technology (IT) knowledge, only business background is required. After completing the setting of field review conditions, AI Agent technology can be used to generate the corresponding review process for each field review condition. After completing the generation of the review process, the staff can test it on the intelligent automation platform to determine whether the generated review process meets the requirements. If not, continue to modify the field review conditions until the requirements are met.

[0227] It should be noted that the audit conditions for some fields are relatively simple, and the AI ​​Agent can directly call the large language model to generate the corresponding audit process. Even in the vertical field, it is usually sufficient to add some positive and negative examples. However, the audit conditions for some fields are more complicated. Taking the audit condition in Example 1 above as an example, when generating the audit process, some process links need to generate code before they can be completed.

[0228] As an example, the review process generated according to Example 1 can be as follows Figure 6 shown.

[0229] Part 2: Review results of AI Agent.

[0230] After completing the setting of the field review conditions, the actual document processing process can be executed. After receiving the document processing task, the human-computer collaboration center will first call the corresponding document extraction capability to complete the extraction of key fields, and then call the set review process in sequence to obtain the review results of the key fields. The key fields that fail the review will be marked with the content fragments in the original document and the modification suggestions returned by the review process and sent to manual processing. The key fields that pass the review and the key fields that are manually confirmed will continue to be processed in the subsequent business process under the control of the human-computer collaboration center.

[0231] As an example, the key field extraction + review + human-computer interaction process can be as follows Figure 7 shown.

[0232] In summary, the technical solution provided by this application has at least the following advantages:

[0233] 1. High flexibility: Workers can interact using natural language without the need for professional IT knowledge.

[0234] 2. High scalability: New field audit conditions can be added quickly, even completely new audit content can be supported.

[0235] 3. High coverage: Large language models have strong generalization and semantic understanding capabilities, and do not need to be set up for all situations.

[0236] 4. Strong adaptability: There is no threshold introduced for confidence level that is limited to a specific data set, and it can adapt to a variety of different application scenarios and data sets.

[0237] In order to implement the above embodiments, the present application also provides a document processing device based on AI, RPA and AI Agent.

[0238] Figure 8 This is a structural diagram of a document processing device based on AI, RPA and AI Agent provided by an embodiment of the present application.

[0239] like Figure 8 As shown, the document processing device 800 based on AI, RPA and AI Agent includes: an extraction module 810, a first acquisition module 820, an execution module 830 and a first generation module 840.

[0240] The extraction module 810 is used to execute a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task;

[0241] A first acquisition module 820 is used to acquire an execution code of at least one first review process adapted to a target business scenario; wherein the execution code is generated according to a customized field review condition under the target business scenario;

[0242] An execution module 830, configured to determine a second audit process matching any key field from the first audit process, and execute an execution code of the second audit process on any key field to obtain first audit information;

[0243] The first generating module 840 is used to generate a task execution result according to the first review information and each key field.

[0244] In any one of the embodiments of the present application, the document processing device 800 based on AI, RPA and AI Agent is applied to an RPA robot.

[0245] In any embodiment of the present application, the first audit information is used to indicate whether each key field has passed the audit of the corresponding second audit process; the first generation module 840 is used to: when the first audit information indicates that there is a first target field that has not passed the audit in each key field, mark the target content segment in the document to be processed and determine the reason why the first target field has not passed the audit; wherein the target content segment is the content segment where the first target field is located; according to the reason, generate a modification suggestion for the first target field, and generate a review task based on the modification suggestion and the marked document to be processed; schedule the review task to the target review object so that the target review object modifies the target content segment in the marked document to be processed based on the modification suggestion; receive the modified document to be processed sent by the target review object in response to the audit task; generate a task execution result based on the modified document to be processed and the key fields that have passed the audit.

[0246] In any embodiment of the present application, the first generation module 840 is used to: extract a modified target content segment from the modified document to be processed, and extract a second target field adapted to the target business scenario from the modified target content segment; determine a third review process that matches the second target field from each first review process; execute the execution code of the third review process on the second target field to obtain second review information; wherein the second review information is used to indicate whether the second target field has passed the review of the third review process; when the second review information indicates that the second target field has passed the review, generate a task execution result based on the second target field that has passed the review and the key fields that have passed the review.

[0247] In any embodiment of the present application, the first generation module 840 is further used to: when the first audit information indicates that all key fields have passed the audit, generate a task execution result according to the key fields that have passed the audit.

[0248] In any embodiment of the present application, the execution code of the first audit process is generated using the following modules:

[0249] A second acquisition module is used to acquire at least one natural language text input in response to the condition configuration operation; wherein the natural language text is used to indicate the field review condition;

[0250] A second generation module is used to generate first prompt information according to the natural language text; wherein the first prompt information is used to prompt the large language model to execute a process generation task;

[0251] A first processing module, used to call the large language model to process the first prompt information, and obtain a link code of at least one process link in the first review process;

[0252] The third generation module is used to generate the execution code of the first review process according to the link code of each process link.

[0253] In any one embodiment of the present application, the second generation module is used to: obtain at least one reference example adapted to the natural language text; wherein the reference example includes a first reference example and / or a second reference example, the first reference example includes key fields that meet the field review conditions indicated by the natural language text, and the second reference example includes key fields that do not meet the field review conditions indicated by the natural language text; generate a first prompt message based on the reference example and the natural language text.

[0254] In any one of the embodiments of the present application, the document processing device 800 based on AI, RPA and AI Agent further includes:

[0255] A test module, used to test the sample key fields using the execution code of the first audit process to obtain a test result; wherein the test result is used to indicate whether the sample key fields have passed the audit of the first audit process in the test scenario;

[0256] A judgment module is used to judge whether the test results meet the set conditions; wherein the set conditions include that the test results are consistent with the annotation results of the sample key fields; the annotation results are used to indicate whether the sample key fields have passed the review in the actual target business scenario;

[0257] The updating module is used to update the natural language text if the set conditions are not met until the test results meet the set conditions.

[0258] In any embodiment of the present application, the extraction module 810 is used to: perform character recognition on the document to be processed based on optical character recognition (OCR) technology to obtain document content; obtain a prompt template adapted to the target business scenario; wherein the prompt template is used to indicate the field information to be extracted by the large language model in the target business scenario; fill the prompt template according to the document content to obtain second prompt information; wherein the second prompt information is used to prompt the large language model to perform a field extraction task; call the large language model to process the second prompt information to obtain at least one key field adapted to the target business scenario.

[0259] In any embodiment of the present application, the extraction module 810 is used to: perform character recognition on the document to be processed based on optical character recognition (OCR) technology to obtain the document content, and generate an input sequence based on the document content; use a sequence labeling model to label the input sequence to obtain a labeling sequence; wherein each label in the labeling sequence is used to indicate whether the character at the corresponding position in the input sequence belongs to a key field; the sequence labeling model is obtained by training with corpus under the target business scenario; according to the labeling sequence, extract at least one key field adapted to the target business scenario from the input sequence.

[0260] In any one of the embodiments of the present application, the document processing device 800 based on AI, RPA and AI Agent further includes:

[0261] The second processing module is used to instruct at least one of the following: sending the task execution result to a designated device; persistently storing the task execution result; and sending the task execution result to a designated account.

[0262] It should be noted here that the document processing device based on AI, RPA and AI Agent provided in the embodiment of the present application can implement all the method steps implemented in any of the above-mentioned method embodiments, and can achieve the same technical effects. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0263] Fig. 9 FIG. 1 is a block diagram showing a structure of an electronic device according to an embodiment of the present application. Fig. 9 As shown, the electronic device includes: a memory 910 and a processor 920, and the memory 910 stores a computer program that can be run on the processor 920. When the processor 920 executes the computer program, the document processing method based on AI, RPA and AI Agent in the above embodiment is implemented. The number of the memory 910 and the processor 920 can be one or more.

[0264] The electronic device also includes:

[0265] The communication interface 930 is used to communicate with external devices and perform data exchange transmission.

[0266] If the memory 910, the processor 920 and the communication interface 930 are implemented independently, the memory 910, the processor 920 and the communication interface 930 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig. 9 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0267] Optionally, in a specific implementation, if the memory 910, the processor 920 and the communication interface 930 are integrated on a chip, the memory 910, the processor 920 and the communication interface 930 can communicate with each other through an internal interface.

[0268] An embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the document processing method based on AI, RPA, and AI Agent provided in any embodiment of the present application.

[0269] An embodiment of the present application also provides a chip, which includes a processor for calling and executing instructions stored in the memory from the memory, so that a communication device equipped with the chip executes the document processing method based on AI, RPA and AI Agent provided in any embodiment of the present application.

[0270] An embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected through an internal connection path, and the processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the document processing method based on AI, RPA and AIAgent provided in any embodiment of the application.

[0271] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. It is worth noting that the processor may be a processor supporting an advanced RISC machines (ARM) architecture.

[0272] Further, optionally, the above-mentioned memory may include a read-only memory and a random access memory, and may also include a non-volatile random access memory. The memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus RAM (DR RAM).

[0273] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0274] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0275] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0276] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0277] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0278] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0279] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0280] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A document processing method based on artificial intelligence AI, robotic process automation RPA and digital labor platform intelligent assistant AIAgent, characterized in that: include: Executing a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task; Obtaining at least one execution code of a first review process adapted to the target business scenario; wherein the execution code is generated according to a customized field review condition under the target business scenario; Determine a second audit process matching any key field from the first audit process, and execute the execution code of the second audit process on the any key field to obtain first audit information; A task execution result is generated according to the first audit information and each of the key fields.

2. The method according to claim 1, characterized in that The method is performed by an RPA robot.

3. The method according to claim 1, characterized in that The first audit information is used to indicate whether each of the key fields has passed the audit of the corresponding second audit process; Generating a task execution result according to the first audit information and each of the key fields includes: In the case where the first audit information indicates that there is a first target field that has not passed the audit in each of the key fields, marking the target content segment in the to-be-processed document and determining the reason why the first target field has not passed the audit; wherein the target content segment is the content segment where the first target field is located; Generate a modification suggestion for the first target field according to the reason, and generate a review task according to the modification suggestion and the annotated document to be processed; Dispatching the review task to a target review object, so that the target review object modifies the target content segment in the annotated document to be processed based on the modification suggestion; Receiving the modified document to be processed sent by the target audit object in response to the audit task; The task execution result is generated according to the modified document to be processed and the key fields that have passed the review.

4. The method according to claim 3, characterized in that The step of generating the task execution result according to the modified document to be processed and the key fields that have passed the review includes: Extracting a modified target content segment from the modified document to be processed, and extracting a second target field adapted to the target business scenario from the modified target content segment; Determine a third audit process matching the second target field from each of the first audit processes; Executing the execution code of the third audit process on the second target field to obtain second audit information; wherein the second audit information is used to indicate whether the second target field has passed the audit of the third audit process; In the case where the second audit information indicates that the second target field has passed the audit, the task execution result is generated according to the second target field that has passed the audit and the key field that has passed the audit.

5. The method according to claim 3, characterized in that: The step of generating a task execution result according to the first audit information and each of the key fields further includes: When the first audit information indicates that all the key fields have passed the audit, the task execution result is generated according to the key fields that have passed the audit.

6. The method according to claim 1, characterized in that The execution code of the first audit process is generated by the following steps: Acquire at least one natural language text input in response to a condition configuration operation; wherein the natural language text is used to indicate the field review condition; Generate first prompt information according to the natural language text; wherein the first prompt information is used to prompt the large language model to execute a process generation task; Calling the large language model to process the first prompt information to obtain a link code of at least one process link in the first review process; According to the link code of each process link, an execution code of the first review process is generated.

7. The method according to claim 6, characterized in that The step of generating first prompt information according to the natural language text includes: Acquire at least one reference example adapted to the natural language text; wherein the reference example includes a first reference example and / or a second reference example, the first reference example includes a key field that satisfies the field review condition indicated by the natural language text, and the second reference example includes a key field that does not satisfy the field review condition indicated by the natural language text; The first prompt information is generated according to the reference example and the natural language text.

8. The method according to claim 6, characterized in that The method further comprises: Using the execution code of the first audit process to test the sample key field to obtain a test result; wherein the test result is used to indicate whether the sample key field has passed the audit of the first audit process in the test scenario; Determine whether the test result meets the set conditions; wherein the set conditions include that the test result is consistent with the annotation result of the sample key field; the annotation result is used to indicate whether the sample key field has passed the review in the actual target business scenario; If the set condition is not met, the natural language text is updated until the test result meets the set condition.

9. The method according to any one of claims 1 to 8, characterized in that The step of extracting at least one key field adapted to the target business scenario from the document to be processed in the document processing task includes: Based on optical character recognition (OCR) technology, character recognition is performed on the document to be processed to obtain document content; Acquire a prompt template adapted to the target business scenario; wherein the prompt template is used to indicate field information to be extracted by the large language model in the target business scenario; Filling the prompt template according to the document content to obtain second prompt information; wherein the second prompt information is used to prompt the large language model to perform a field extraction task; The large language model is called to process the second prompt information to obtain at least one key field adapted to the target business scenario.

10. The method according to any one of claims 1 to 8, characterized in that The step of extracting at least one key field adapted to the target business scenario from the document to be processed in the document processing task includes: Based on optical character recognition (OCR) technology, character recognition is performed on the document to be processed to obtain document content, and an input sequence is generated according to the document content; The input sequence is annotated using a sequence annotation model to obtain an annotation sequence; wherein each label in the annotation sequence is used to indicate whether the character at the corresponding position in the input sequence belongs to the key field; the sequence annotation model is obtained by training with the corpus in the target business scenario; At least one key field adapted to the target business scenario is extracted from the input sequence according to the marked sequence.

11. The method according to any one of claims 1 to 8, characterized in that The method further comprises at least one of the following: Sending the task execution result to a designated device; Persistently storing the task execution results; The task execution result is sent to the specified account.

12. A document processing device based on AI, RPA and AI Agent, characterized in that: include: An extraction module, used to execute a document processing task to extract at least one key field adapted to a target business scenario from a document to be processed in the document processing task; A first acquisition module, used to acquire an execution code of at least one first review process adapted to the target business scenario; wherein the execution code is generated according to a customized field review condition under the target business scenario; an execution module, configured to determine a second audit process matching any key field from the first audit process, and execute an execution code of the second audit process on the any key field to obtain first audit information; The first generating module is used to generate a task execution result according to the first audit information and each of the key fields.

13. The device according to claim 12, characterized in that The device is applied to an RPA robot.

14. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the method according to any one of claims 1 to 11.

15. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images