Multi-agent-based document processing methods, systems, devices, and storage media

Abstract

This application provides a document processing method, system, device, and storage medium based on multi-agent intelligence. The method includes: acquiring a first document; generating a task chain corresponding to the first document; based on the task chain, invoking a document processing intelligence to perform structured processing on the first document to obtain a second document; based on the task chain, invoking an indexing intelligence to construct multiple indexes corresponding to the second document; and based on the task chain, invoking a publishing intelligence to publish the second document and its corresponding indexes. In this application's embodiments, multiple dedicated intelligences are uniformly scheduled to form a standardized and automated processing flow from unstructured document to unstructured document, which simplifies the processing of unstructured documents, improves document processing efficiency, and reduces the operation and maintenance costs of the document processing system.

Description

Technical Field

[0001] This application relates to the field of text processing technology, and in particular to a multi-agent-based document processing method, system, device and storage medium. Background Technology

[0002] With the deepening of digital transformation in industries such as government affairs, finance, healthcare, and law, a massive amount of unstructured documents have been generated. The document types of these unstructured documents include, but are not limited to: scanned documents, PDFs, images, and Office files.

[0003] In existing technologies, each stage of unstructured document processing, such as collection, parsing, information extraction, index building, and service publishing, requires independent tools or scripts for each stage. Furthermore, the data formats for each stage are not uniform, which makes the processing of unstructured documents more complex, reduces the efficiency of document processing, and leads to higher maintenance costs for document processing systems. Summary of the Invention

[0004] This application provides a document processing method, system, device, and storage medium based on multi-agent technology, aiming to solve the technical problems that the processing of unstructured documents in the prior art is relatively complex, which reduces the efficiency of document processing and leads to high operation and maintenance costs of document processing systems.

[0005] In a first aspect, embodiments of this application provide a document processing method based on multiple agents, the method comprising:

[0006] Obtain the first document, which is an unstructured document;

[0007] Generate the task chain corresponding to the first document;

[0008] Based on the task chain, a document processing agent is invoked to perform structured processing on the first document to obtain a second document, which is a structured document.

[0009] Based on the task chain, the indexing agent is invoked to construct multiple indexes corresponding to the second document;

[0010] Based on the task chain, the publishing agent is invoked to publish the second document and the various indexes corresponding to the second document.

[0011] Optionally, the task chain for generating the first document includes:

[0012] Generate the task object corresponding to the first document;

[0013] The task object is input into a preset task orchestrator to generate a task chain corresponding to the first document;

[0014] The task chain includes each task node of the first document in the document processing process, and the intelligent agent called by each task node.

[0015] Optionally, the document processing agent includes a parsing agent, a semantic agent, and a verification agent;

[0016] Based on the task chain, the document processing agent is invoked to perform structured processing on the first document to obtain the second document, including:

[0017] Based on the task chain, the parsing agent is invoked to perform layout analysis and text extraction on the first document to obtain a structured content list;

[0018] The semantic intelligent agent is invoked to perform semantic segmentation, metadata extraction, and hierarchical relationship construction on the structured content list, generating candidate documents, which include structured document objects, chapter objects, and paragraph objects.

[0019] The verification agent is invoked to perform consistency verification on the candidate documents.

[0020] If the consistency check fails, the verification agent is invoked to repair the candidate document, resulting in a second document;

[0021] If the consistency check passes, the candidate document will be identified as the second document.

[0022] Optionally, the step of invoking the verification agent to perform consistency verification on the candidate documents includes:

[0023] Invoke the verification agent to compare the first document and the candidate document, and determine the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate document;

[0024] Based on the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate documents, the quality score of the candidate documents is determined.

[0025] If the quality score of the candidate document is lower than a preset threshold, the consistency check is determined to have failed.

[0026] If the quality score of the candidate document is equal to or higher than a preset threshold, the consistency check is deemed to have passed.

[0027] Optionally, based on the task chain, the step of invoking the indexing agent to construct multiple indexes corresponding to the second document includes:

[0028] Based on the task chain, the indexing agent is invoked to construct the full-text search index and aggregated statistical index corresponding to the second document;

[0029] The full-text search index includes paragraph-level indexes, chapter-level indexes, and document-level indexes, while the aggregated statistics index includes a time field index.

[0030] Optionally, the step of invoking the publishing agent based on the task chain to publish the second document and various indexes corresponding to the second document includes:

[0031] Based on the task chain, the publishing agent publishes the network interface to transmit the second document, the various indexes corresponding to the second document, and the mapping data through the network interface;

[0032] The mapping data includes mappings between paragraph objects and page indices of the first document, mappings between paragraph objects and chapter objects, and mappings between paragraph objects and document objects.

[0033] Optionally, the method further includes:

[0034] The first document, the task chain corresponding to the first document, the second document, and the various indexes corresponding to the second document are backed up.

[0035] Secondly, embodiments of this application provide a document processing system, the document processing system comprising:

[0036] The acquisition module is used to acquire the first document, which is an unstructured document;

[0037] The generation module is used to generate the task chain corresponding to the first document;

[0038] The processing module is used to invoke a document processing agent based on the task chain to perform structured processing on the first document to obtain a second document, wherein the second document is a structured document;

[0039] The construction module is used to invoke the indexing agent based on the task chain to construct multiple indexes corresponding to the second document;

[0040] The publishing module is used to invoke the publishing agent based on the task chain to publish the second document and the various indexes corresponding to the second document.

[0041] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the multi-agent-based document processing method described in the first aspect.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-agent-based document processing method described in the first aspect.

[0043] This application provides a document processing method, system, device, and storage medium based on multi-agent intelligence. The method includes: acquiring a first document, which is an unstructured document; generating a task chain corresponding to the first document; based on the task chain, invoking a document processing intelligence to perform structured processing on the first document to obtain a second document, which is a structured document; based on the task chain, invoking an indexing intelligence to construct multiple indexes corresponding to the second document; and based on the task chain, invoking a publishing intelligence to publish the second document and its corresponding indexes. In this embodiment, the task chain is used to invoke a document processing intelligence to perform structured processing on the unstructured document, invoke an indexing intelligence to construct multiple indexes, and invoke a publishing intelligence to publish the structured document and its indexes. This unifies the scheduling of multiple dedicated intelligences, forming a standardized and automated processing flow from unstructured document to unstructured document. This simplifies the processing of unstructured documents, improves document processing efficiency, and reduces the operation and maintenance costs of the document processing system. Furthermore, this application can fully utilize computing resources based on the parallel scheduling of task chains, significantly improving the processing efficiency of batch documents. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart of a document processing method based on multiple agents provided in an embodiment of this application;

[0046] Figure 2 This is a schematic diagram of the architecture of a document processing system provided in an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the workflow of the verification agent provided in the embodiments of this application;

[0048] Figure 4 This is a schematic diagram of the workflow of the publishing agent provided in the embodiments of this application;

[0049] Figure 5 This is a schematic diagram of the multi-agent workflow provided in the embodiments of this application;

[0050] Figure 6 This is a schematic diagram of the document processing flow provided in the embodiments of this application;

[0051] Figure 7 This is a schematic diagram of the structure of a document processing system provided in an embodiment of this application;

[0052] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] See Figure 1 , Figure 1 This is a flowchart of a multi-agent-based document processing method provided in an embodiment of this application, such as... Figure 1 As shown in the figure, a document processing method based on multiple agents provided in this application includes the following steps:

[0055] Step S101: Obtain the first document, which is an unstructured document.

[0056] In this step, the first document is obtained, wherein the first document is an unstructured document.

[0057] Alternatively, the first document can be entered via a local file system, a network address, or in batches; that is, there can be multiple first documents.

[0058] Step S102: Generate the task chain corresponding to the first document.

[0059] In this step, after obtaining the first document, a task chain corresponding to the first document is generated. This task chain includes each task node in the document processing of the first document, and the intelligent agent invoked by each task node.

[0060] Step S103: Based on the task chain, the document processing agent is invoked to perform structured processing on the first document to obtain a second document, which is a structured document.

[0061] It should be understood that the task chain includes a document processing agent, which is used to convert unstructured documents into structured documents.

[0062] In this step, based on the task chain corresponding to the first document, the document processing agent is invoked to perform structured processing on the first document, resulting in a structured second document.

[0063] Step S104: Based on the task chain, invoke the indexing agent to construct multiple indexes corresponding to the second document.

[0064] It should be understood that the task chain includes an indexing agent, which is used to generate the index.

[0065] In this step, based on the task chain corresponding to the first document, the indexing agent is invoked to construct multiple indexes corresponding to the second document.

[0066] Step S105: Based on the task chain, call the publishing agent to publish the second document and the various indexes corresponding to the second document.

[0067] It should be understood that the task chain includes a publishing agent, which is used to publish documents and indexes.

[0068] In this step, based on the task chain corresponding to the first document, the publishing agent is invoked to publish the second document and the various indexes corresponding to the second document.

[0069] The multi-agent-based document processing method provided in this application can be applied to document processing systems. Please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram of the architecture of a document processing system provided in an embodiment of this application. For example... Figure 2 The document processing system shown includes a task entry layer, an intelligent agent system processing layer, a data and indexing layer, and a service output layer.

[0070] The task entry layer is responsible for receiving document processing requests and defining task parameters. The agent execution layer includes acquisition agents, parsing agents, semantic agents, validation agents, indexing agents, and publishing agents. The data and indexing layer includes shared object storage, an event log library, a full-text index library, a vector index library, and a statistical index library. The service output layer includes retrieval interfaces, statistical interfaces, citation interfaces, and traceability interfaces.

[0071] Furthermore, the aforementioned document processing system also incorporates a feedback learning mechanism, allowing for manual corrections of chapter boundaries, time fields, and attribution relationships. These corrections are recorded and written to a feedback sample library. The system prioritizes loading feedback rules when processing similar document tasks, thereby continuously optimizing document output quality and adapting to more diverse document types and business needs.

[0072] In this embodiment, a task chain is used to invoke a document processing agent to perform structuring on unstructured documents, an indexing agent to build various indexes, and a publishing agent to publish structured documents and various indexes. This unified scheduling of multiple dedicated agents forms a standardized and automated processing flow from unstructured documents to unstructured documents, simplifying the processing of unstructured documents, improving document processing efficiency, and reducing the operational costs of the document processing system. Furthermore, this application can fully utilize computing resources based on the parallel scheduling of the task chain, significantly improving the processing efficiency of batch documents.

[0073] It should be understood that the document processing method based on multi-agent provided in this application can be applied to scenarios such as government document processing, enterprise archive digitization, legal document retrieval, scientific research data knowledge transformation, and medical document structuring, and has clear industrial application value and engineering feasibility.

[0074] Optionally, the task chain for generating the first document includes:

[0075] Generate the task object corresponding to the first document;

[0076] The task object is input into a preset task orchestrator to generate a task chain corresponding to the first document;

[0077] The task chain includes each task node of the first document in the document processing process, and the intelligent agent called by each task node.

[0078] In this embodiment, after obtaining the first document, a task object and a task identifier corresponding to the first document can be generated. Furthermore, the task object is input into a preset task orchestrator to generate a task chain corresponding to the first document.

[0079] Optionally, the aforementioned task orchestrators include, but are not limited to, LangGraph, CrewAI, and AutoGen.

[0080] Optionally, the above task chain is a directed acyclic graph, which includes a collection stage, a parsing stage, a semantic enrichment stage, a quality verification stage, an index building stage, and a service publishing stage. The above task chain can call the intelligent agent corresponding to each task node.

[0081] In this embodiment, multiple agents are uniformly scheduled by a task orchestration controller to form a standardized and automated processing flow from unstructured document to unstructured document, which improves the efficiency of document processing.

[0082] Optionally, the document processing agent includes a parsing agent, a semantic agent, and a verification agent;

[0083] Based on the task chain, the document processing agent is invoked to perform structured processing on the first document to obtain the second document, including:

[0084] Based on the task chain, the parsing agent is invoked to perform layout analysis and text extraction on the first document to obtain a structured content list;

[0085] The semantic intelligent agent is invoked to perform semantic segmentation, metadata extraction, and hierarchical relationship construction on the structured content list, generating candidate documents, which include structured document objects, chapter objects, and paragraph objects.

[0086] The verification agent is invoked to perform consistency verification on the candidate documents.

[0087] If the consistency check fails, the verification agent is invoked to repair the candidate document, resulting in a second document;

[0088] If the consistency check passes, the candidate document will be identified as the second document.

[0089] It should be understood that document processing agents include parsing agents, semantic agents, and verification agents.

[0090] In this embodiment, the parsing agent performs layout analysis and text extraction on the input first document, outputting a page-level structured content list. This structured content list includes at least page indexes, block types, text levels, and text content. Optionally, the parsing agent can dynamically select different parsing backends based on the quality level of the input first document (e.g., plain text PDF, scanned image), such as a text extraction backend, an OCR backend, or a backend that integrates a visual language model.

[0091] The semantic intelligent agent performs semantic segmentation, metadata extraction, and hierarchical relationship construction based on a structured content list, generating structured document objects, chapter objects, and paragraph objects, and then generates candidate documents based on these objects.

[0092] Optionally, the semantic agent jointly determines the semantic segmentation boundary by coordinating the execution of two sub-processes: directory recognition and text recognition, and performs a unified format standardization conversion on the extracted time field of the first document summary.

[0093] The validation agent performs consistency checks on document objects, chapter objects, and paragraph objects, including paragraph boundaries, page number alignment, field integrity, and time normalization.

[0094] If the consistency check passes, the candidate document will be selected as the second document.

[0095] If the consistency check fails, the verification agent will repair the candidate document. It should be understood that the scope of the verification agent's repair of the candidate document includes, but is not limited to, re-estimation of the directory range, correction of page number offsets, correction of item folding, and field backfilling based on rules and knowledge base.

[0096] The verification agent performs a consistency check on the repaired candidate document again. If the check passes, the repaired candidate document is determined as the second document; if the check fails, the candidate document continues to be repaired.

[0097] In this embodiment, a quantifiable quality check and automated repair process is introduced during document processing to ensure consistency between the unstructured document before conversion and the structured document after conversion. This effectively reduces file problems such as segmentation deviation and missing fields, and ensures the reliability of the output structured document.

[0098] Optionally, the step of invoking the verification agent to perform consistency verification on the candidate documents includes:

[0099] Invoke the verification agent to compare the first document and the candidate document, and determine the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate document;

[0100] Based on the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate documents, the quality score of the candidate documents is determined.

[0101] If the quality score of the candidate document is lower than a preset threshold, the consistency check is determined to have failed.

[0102] If the quality score of the candidate document is equal to or higher than a preset threshold, the consistency check is deemed to have passed.

[0103] In this embodiment, the quality score corresponding to the candidate document can be calculated using the following formula:

[0104] Q = w1×Rcov + w2×(1-Rmiss) + w3×(1-Ralign) + w4×Rtime

[0105] Where Q represents the quality score, Rcov represents the coverage rate, Rmiss represents the field missing rate, Ralign represents the alignment deviation rate, Rtime represents the time standardization success rate, and w1, w2, w3, and w4 are preset weight coefficients.

[0106] If the quality score of a candidate document is lower than a preset threshold, the consistency check is determined to have failed; if the quality score of a candidate document is equal to or higher than the preset threshold, the consistency check is determined to have passed.

[0107] In this embodiment, a consistency check process is introduced during document processing to ensure consistency between the unstructured document before conversion and the structured document after conversion, thereby guaranteeing the reliability of the output structured document.

[0108] To facilitate understanding of the solution provided in this embodiment, please refer to [link / reference]. Figure 3 ,like Figure 3 As shown, a consistency check is performed on the document to obtain a quality score. A threshold is determined based on the quality score, which is compared with a preset threshold. If the quality score is lower than the preset threshold, the consistency check is considered passed. If the quality score is equal to or higher than the preset threshold, the document is repaired, and the consistency check is performed again on the repaired document.

[0109] Optionally, based on the task chain, the step of invoking the indexing agent to construct multiple indexes corresponding to the second document includes:

[0110] Based on the task chain, the indexing agent is invoked to construct the full-text search index and aggregated statistical index corresponding to the second document;

[0111] The full-text search index includes paragraph-level indexes, chapter-level indexes, and document-level indexes, while the aggregated statistics index includes a time field index.

[0112] In this embodiment, after obtaining the second document, the indexing agent is invoked to construct a full-text search index and an aggregated statistical index corresponding to the second document. The full-text search index includes paragraph-level, chapter-level, and document-level indexes. Furthermore, a separate time field index is established to support efficient time series analysis.

[0113] Optionally, the step of invoking the publishing agent based on the task chain to publish the second document and various indexes corresponding to the second document includes:

[0114] Based on the task chain, the publishing agent publishes the network interface to transmit the second document, the various indexes corresponding to the second document, and the mapping data through the network interface;

[0115] The mapping data includes mappings between paragraph objects and page indices of the first document, mappings between paragraph objects and chapter objects, and mappings between paragraph objects and document objects.

[0116] In this embodiment, after obtaining the second document and its corresponding indexes, a publishing agent is invoked. This publishing agent provides a unified application programming interface (API) to output the various indexes and mapping data corresponding to the second document. The mapping data includes mappings between paragraph objects and page indexes of the first document, mappings between paragraph objects and chapter objects, and mappings between paragraph objects and document objects.

[0117] In this embodiment, by calling the publishing agent's publishing network interface, the function of uniformly outputting retrieval, statistics, citation, and tracing interfaces to the upper-layer application is realized, enabling the document processing method provided in this application embodiment to have cross-scenario reuse capability.

[0118] In other embodiments, stage events during document processing can be pushed to a visual interactive interface in real time via an event stream, and the event push can be turned off after the task is completed.

[0119] For a better understanding of the technical solutions provided in this embodiment, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of the workflow of the publishing agent provided in the embodiments of this application. For example... Figure 4 As shown, the publishing agent outputs the retrieval results object (i.e., Figure 4 Search services in the middle), statistical result objects (i.e. Figure 4 Statistical services in the data), referencing result objects (i.e. Figure 4 Reference services in the middle), traceability result objects (i.e. Figure 4 (Tracing services within).

[0120] The search results object includes the hit paragraph, its chapter, its document, and the sorting field; the statistical results object includes aggregated statistics by time, document, and chapter; the citation results object includes standardized citation text; and the traceability results object includes the mapping from the hit paragraph to the original page index and the original document location. When a terminal initiates a traceability request, the system reads intermediate artifacts from the archive based on the task identifier and returns the processing evidence chain.

[0121] Optionally, the method further includes:

[0122] The first document, the task chain corresponding to the first document, the second document, and the various indexes corresponding to the second document are backed up.

[0123] In this embodiment, an archiving agent can also be invoked to version archive the stage events, exception information, and intermediate / final artifacts generated during document processing, forming a replayable task chain.

[0124] In this embodiment, by fully recording the event chain and artifact version chain during document processing, it supports detailed replay and auditing at the task level, thus meeting the requirements for compliance and problem investigation.

[0125] For a better understanding of the overall technical solution, please refer to [link / reference]. Figure 5 and Figure 6 .

[0126] An optional application scenario of the document processing method provided in this application includes the following steps:

[0127] Step A1: Task Standardization

[0128] The input document is encapsulated as a standard task object, which includes at least: task identifier, input path, document type, processing strategy, target index, and output path.

[0129] Step A2: Task Graph Generation and Scheduling

[0130] A directed acyclic task graph is generated based on the task objects. Nodes include collection, parsing, enrichment, verification, indexing, and publishing; edges represent stage dependencies. The scheduler performs concurrent allocation based on resource quotas.

[0131] Step A3: Structured Analysis

[0132] The analytical agent performs layout understanding, text extraction, and block-level structure recognition on the document, outputting a structured content list. The content list records page index, block type, text level, text content, and related information.

[0133] Step A4: Semantic Enrichment

[0134] The semantic intelligent agent performs directory recognition, text boundary recognition, metadata extraction, time field normalization, and hierarchical relationship construction on the structured content list, generating document objects, chapter objects, paragraph objects, and relationship objects.

[0135] Step A5: Quality Check and Repair

[0136] The system verifies the agent's computational coverage, missing rate, bias rate, and time standardization success rate; when these fall below a threshold, a repair sub-process is triggered. The repair sub-process can reassess the directory range, correct page number offsets, perform entry folding, or backfill fields.

[0137] Step A6: Index Building

[0138] The indexing agent creates paragraph-level, chapter-level, and document-level indexes on the verified data, and also creates a time field index for sorting and trend analysis.

[0139] Step A7: Service Publication

[0140] Release intelligent agent generation retrieval interface, statistics interface, citation interface and traceability interface, and return result set and evidence mapping to the terminal.

[0141] Step A8: Status Archiving and Replay

[0142] The state archiving agent archives stage events, exception information, intermediate artifacts and final artifacts in a versioned manner to form a replayable task chain.

[0143] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a document processing system provided in an embodiment of this application, such as... Figure 7 As shown, the document processing system 700 includes:

[0144] The acquisition module 701 is used to acquire a first document, wherein the first document is an unstructured document;

[0145] Generation module 702 is used to generate the task chain corresponding to the first document;

[0146] Processing module 703 is used to invoke a document processing agent based on the task chain to perform structured processing on the first document to obtain a second document, wherein the second document is a structured document;

[0147] Module 704 is used to invoke the indexing agent based on the task chain to construct multiple indexes corresponding to the second document;

[0148] The publishing module 705 is used to call the publishing agent based on the task chain to publish the second document and the various indexes corresponding to the second document.

[0149] Optionally, the generation module 702 is specifically used for:

[0150] Generate the task object corresponding to the first document;

[0151] The task object is input into a preset task orchestrator to generate a task chain corresponding to the first document;

[0152] The task chain includes each task node of the first document in the document processing process, and the intelligent agent called by each task node.

[0153] Optionally, the document processing agent includes a parsing agent, a semantic agent, and a verification agent;

[0154] The processing module 703 is specifically used for:

[0155] Based on the task chain, the parsing agent is invoked to perform layout analysis and text extraction on the first document to obtain a structured content list;

[0156] The semantic intelligent agent is invoked to perform semantic segmentation, metadata extraction, and hierarchical relationship construction on the structured content list, generating candidate documents, which include structured document objects, chapter objects, and paragraph objects.

[0157] The verification agent is invoked to perform consistency verification on the candidate documents.

[0158] If the consistency check fails, the verification agent is invoked to repair the candidate document, resulting in a second document;

[0159] If the consistency check passes, the candidate document will be identified as the second document.

[0160] Optionally, the processing module 703 is further specifically used for:

[0161] Invoke the verification agent to compare the first document and the candidate document, and determine the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate document;

[0162] Based on the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate documents, the quality score of the candidate documents is determined.

[0163] If the quality score of the candidate document is lower than a preset threshold, the consistency check is determined to have failed.

[0164] If the quality score of the candidate document is equal to or higher than a preset threshold, the consistency check is deemed to have passed.

[0165] Optionally, the construction module 704 is specifically used for:

[0166] Based on the task chain, the indexing agent is invoked to construct the full-text search index and aggregated statistical index corresponding to the second document;

[0167] The full-text search index includes paragraph-level indexes, chapter-level indexes, and document-level indexes, while the aggregated statistics index includes a time field index.

[0168] Optionally, module 705 is published, specifically for:

[0169] Based on the task chain, the publishing agent publishes the network interface to transmit the second document, the various indexes corresponding to the second document, and the mapping data through the network interface;

[0170] The mapping data includes mappings between paragraph objects and page indices of the first document, mappings between paragraph objects and chapter objects, and mappings between paragraph objects and document objects.

[0171] Optionally, the document processing system 700 further includes:

[0172] The backup module is used to back up the first document, the task chain corresponding to the first document, the second document, and various indexes corresponding to the second document.

[0173] The document processing system is capable of implementing each process of the above-described multi-agent-based document processing method, with one-to-one correspondence of technical features and achieving the same technical effect. To avoid repetition, it will not be described in detail here.

[0174] For details, see Figure 8 This application also provides an electronic device, including a bus 801, a transceiver 802, an antenna 803, a bus interface 804, a processor 805, and a memory 806.

[0175] The transceiver 802 is used to acquire a first document, which is an unstructured document.

[0176] The processor 805 is used to generate the task chain corresponding to the first document;

[0177] Based on the task chain, a document processing agent is invoked to perform structured processing on the first document to obtain a second document, which is a structured document.

[0178] Based on the task chain, the indexing agent is invoked to construct multiple indexes corresponding to the second document;

[0179] Based on the task chain, the publishing agent is invoked to publish the second document and the various indexes corresponding to the second document.

[0180] exist Figure 8In this document, a bus architecture (represented by bus 801) is used. Bus 801 can include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 805 and memory represented by memory 806. Bus 801 can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 804 provides an interface between bus 801 and transceiver 802. Transceiver 802 can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 805 is transmitted over a wireless medium via antenna 803, which further receives data and transmits data to processor 805.

[0181] The processor 805 manages the bus 801 and handles general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 806 can be used to store data used by the processor 805 during operation.

[0182] Optionally, the processor 805 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD).

[0183] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described multi-agent document processing method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may include read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0184] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0185] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0186] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A document processing method based on multi-agent intelligence, characterized in that, The method includes: Obtain the first document, which is an unstructured document; Generate the task chain corresponding to the first document; Based on the task chain, a document processing agent is invoked to perform structured processing on the first document to obtain a second document, which is a structured document. Based on the task chain, the indexing agent is invoked to construct multiple indexes corresponding to the second document; Based on the task chain, the publishing agent is invoked to publish the second document and the various indexes corresponding to the second document.

2. The method according to claim 1, characterized in that, The task chain for generating the first document includes: Generate the task object corresponding to the first document; The task object is input into a preset task orchestrator to generate a task chain corresponding to the first document; The task chain includes each task node of the first document in the document processing process, and the intelligent agent called by each task node.

3. The method according to claim 1, characterized in that, The document processing agent includes a parsing agent, a semantic agent, and a verification agent; Based on the task chain, the document processing agent is invoked to perform structured processing on the first document to obtain the second document, including: Based on the task chain, the parsing agent is invoked to perform layout analysis and text extraction on the first document to obtain a structured content list; The semantic intelligent agent is invoked to perform semantic segmentation, metadata extraction, and hierarchical relationship construction on the structured content list, generating candidate documents, which include structured document objects, chapter objects, and paragraph objects. The verification agent is invoked to perform consistency verification on the candidate documents. If the consistency check fails, the verification agent is invoked to repair the candidate document, resulting in a second document; If the consistency check passes, the candidate document will be identified as the second document.

4. The method according to claim 3, characterized in that, The process of invoking the verification agent to perform consistency verification on the candidate documents includes: Invoke the verification agent to compare the first document and the candidate document, and determine the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate document; Based on the coverage, field missing rate, alignment deviation rate and time standardization success rate of the candidate documents, the quality score of the candidate documents is determined. If the quality score of the candidate document is lower than a preset threshold, the consistency check is determined to have failed. If the quality score of the candidate document is equal to or higher than a preset threshold, the consistency check is deemed to have passed.

5. The method according to claim 1, characterized in that, Based on the task chain, the indexing agent is invoked to construct multiple indexes corresponding to the second document, including: Based on the task chain, the indexing agent is invoked to construct the full-text search index and aggregated statistical index corresponding to the second document; The full-text search index includes paragraph-level indexes, chapter-level indexes, and document-level indexes, while the aggregated statistics index includes a time field index.

6. The method according to claim 1, characterized in that, Based on the task chain, the publishing agent is invoked to publish the second document and its corresponding indexes, including: Based on the task chain, the publishing agent publishes the network interface to transmit the second document, the various indexes corresponding to the second document, and the mapping data through the network interface; The mapping data includes mappings between paragraph objects and page indices of the first document, mappings between paragraph objects and chapter objects, and mappings between paragraph objects and document objects.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: The first document, the task chain corresponding to the first document, the second document, and the various indexes corresponding to the second document are backed up.

8. A document processing system based on multi-agent intelligence, characterized in that, The system includes: The acquisition module is used to acquire the first document, which is an unstructured document; The generation module is used to generate the task chain corresponding to the first document; The processing module is used to invoke a document processing agent based on the task chain to perform structured processing on the first document to obtain a second document, wherein the second document is a structured document; The construction module is used to invoke the indexing agent based on the task chain to construct multiple indexes corresponding to the second document; The publishing module is used to invoke the publishing agent based on the task chain to publish the second document and the various indexes corresponding to the second document.

9. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the multi-agent-based document processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the multi-agent document processing method as described in any one of claims 1 to 7.

Images