Examination task processing method and device, electronic equipment and storage medium

Abstract

The disclosure provides an approval task processing method and device, electronic equipment and storage medium, relates to the cross field of office automation and artificial intelligence, and particularly relates to the technical field of multi-system data integration, natural language processing and intelligent decision-making assistance. The method comprises the following steps: performing data cleaning, format standardization and task classification on original approval task data to generate structured approval task data; performing multi-dimensional logical conflict detection and differentiated automatic desensitization processing on the structured approval task data; performing element extraction and element completeness verification on the desensitized structured approval task data to generate an element completeness verification result; generating a reference approval opinion based on the element completeness verification result; and generating centralized display information based on the original approval task data, the conflict detection result, the element completeness verification result and the reference approval opinion.

Description

Technical Field

[0001] This disclosure relates to the intersection of office automation and artificial intelligence, specifically to the fields of multi-system data integration, natural language processing and intelligent decision support, and particularly to an approval task processing method, device, electronic device, computer-readable storage medium and computer program product. Background Technology

[0002] In modern enterprise operations, approval processes are a core element for standardizing business operations and controlling operational risks. Approving users, as key decision-makers, handle a large number of approval tasks from various business scenarios. With the deepening of digital transformation, enterprise business systems are becoming increasingly diversified. To adapt to different business needs, enterprises have successively introduced multiple independently operating business systems. Each of these systems has its own independent database and user interface, making data sharing impossible and creating information silos. Furthermore, the approval elements required for different types of approval tasks vary significantly, necessitating cross-platform operations and relying entirely on manual verification of various approval information.

[0003] In existing technologies, each independent business system can independently manage pending approval tasks within its own platform, sending reminders via in-system messages, emails, etc., and displaying task details within the system. Verification of approval elements is all done manually. Meanwhile, some general-purpose office collaboration tools can aggregate pending tasks from multiple systems, supporting manual addition or automatic synchronization of task information via API (Application Programming Interface), but only displaying basic task content, achieving centralized presentation of pending tasks from multiple systems. However, when approval data from multiple systems is centrally aggregated, sensitive information is directly exposed, posing risks of data leakage and unauthorized access; logical conflicts between elements cannot be detected, easily leading to erroneous approvals due to data inconsistencies. Summary of the Invention

[0004] This disclosure proposes an approval task processing method, apparatus, electronic device, computer-readable storage medium, and computer program product to solve the problems of easy leakage of sensitive information and inability to automatically detect logical conflicts of approval elements when approval data from multiple systems are aggregated. At the same time, it realizes automatic completeness verification of approval elements, thereby improving the security, accuracy, and efficiency of approval processing.

[0005] In a first aspect, this disclosure proposes an approval task processing method, comprising: collecting raw approval task data from multiple target systems; performing data cleaning, format standardization, and task classification on the raw approval task data to generate structured approval task data; performing multi-dimensional logical conflict detection on the structured approval task data across elements and fields to obtain conflict detection results, wherein the multi-dimensional logical conflict detection includes amount conflict detection, time conflict detection, subject conflict detection, clause conflict detection, and process node conflict detection; identifying sensitive information in the structured approval task data based on a preset sensitive element library, and executing different categories of sensitive information according to the approval user's permission level. Differentiated automatic desensitization processing yields desensitized approval task data. Sensitive information includes monetary data, customer information, project secrets, qualification certificates, bank accounts, and core contract terms. The desensitized approval task data undergoes element extraction and element completeness verification, generating element completeness verification results. Based on these results, reference approval opinions are generated. Finally, based on the original approval task data, conflict detection results, element completeness verification results, and reference approval opinions, centralized display information is generated. This centralized display information includes summary approval task information, element completeness verification results, and reference approval opinions. If a conflict detection result indicates a conflict, the corresponding approval task is automatically marked with a conflict type.

[0006] Secondly, this disclosure proposes an approval task processing device, comprising: a data acquisition module configured to acquire raw approval task data from multiple target systems; a data processing module configured to perform data cleaning, format standardization, and task classification on the raw approval task data to generate structured approval task data; a conflict detection module configured to perform multi-dimensional logical conflict detection on the structured approval task data across elements and fields to obtain conflict detection results, wherein the multi-dimensional logical conflict detection includes amount conflict detection, time conflict detection, subject conflict detection, clause conflict detection, and process node conflict detection; and a data desensitization module configured to identify sensitive information in the structured approval task data based on a preset sensitive element library, and to classify different categories of data according to the approval user's permission level. Sensitive information undergoes differentiated automatic de-identification processing to obtain de-identified approval task data. Sensitive information includes monetary data, customer information, project confidentiality, qualification certificates, bank accounts, and core contract terms. The element verification module is configured to extract elements and verify element completeness from the de-identified approval task data, generating element completeness verification results. The opinion generation module is configured to generate reference approval opinions based on the element completeness verification results. The information generation module is configured to generate centralized display information based on the original approval task data, conflict detection results, element completeness verification results, and reference approval opinions. The centralized display information includes approval task summary information, element completeness verification results, and reference approval opinions. If the conflict detection result indicates a conflict, the corresponding approval task is automatically marked with a conflict type.

[0007] Thirdly, embodiments of this disclosure provide an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to implement the approval task processing method as described in the first aspect.

[0008] Fourthly, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions that enable a computer to perform the approval task processing method as described in the first aspect when executed.

[0009] Fifthly, embodiments of this disclosure provide a computer program product including a computer program that, when executed by a processor, can implement the steps of the approval task processing method as described in the first aspect.

[0010] The approval task processing solution provided in this disclosure achieves secure, standardized, automated, and intelligent processing of pending approval tasks across multiple enterprise systems through a six-layer architecture: data acquisition layer, data processing layer, conflict detection layer, data anonymization layer, AI (Artificial Intelligence) analysis layer, and result display layer. The data acquisition layer supports multi-mode and multi-method data acquisition, ensuring secure and efficient data exchange across multiple systems and breaking down information silos. The data processing layer transforms unstructured raw data into analyzable structured data through standardized processing, laying the foundation for subsequent conflict detection, anonymization, and AI analysis. The conflict detection layer performs multi-dimensional logical conflict detection across elements and fields on structured approval task data, automatically identifying data contradictions such as amounts, times, subjects, clauses, and process nodes, achieving proactive risk interception. The data anonymization layer performs differentiated automatic anonymization processing on approval data aggregated from multiple systems based on a preset sensitive element database and approval user permission levels, effectively preventing the leakage of sensitive information and the risk of unauthorized access. The AI ​​analysis layer, as the core layer, replaces manual verification of approval elements and identification of business logic conflicts through a dynamically configurable approval element library, intelligent element extraction, and a dual verification mechanism based on anonymized data to check element completeness and detect multi-dimensional logical conflicts. It automatically identifies missing elements, incomplete information, and data contradictions, and generates targeted reference opinions based on task scenarios and user profiles to assist in approval decisions. The results display layer uses a visual and multi-dimensional presentation method to automatically mark conflicting approval tasks, enabling centralized management and convenient interaction. This allows approvers to quickly focus on core issues, significantly shortening approval cycles and reducing decision-making difficulty. Furthermore, this solution does not require modification of existing enterprise business systems, boasts strong compatibility and scalability, and ensures the security of approval data through multiple security mechanisms such as conflict detection, data anonymization, hierarchical access control, and encrypted transmission, meeting the compliance requirements of enterprise digital office practices.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is an exemplary system architecture to which this disclosure can be applied; Figure 2 A flowchart of an approval task processing method provided in this embodiment of the disclosure; Figure 3A flowchart of yet another approval task processing method provided in this disclosure embodiment; Figure 4 This is a schematic diagram of the overall framework of the intelligent approval assistant system provided in this embodiment; Figure 5 This is a schematic diagram of the approval result display interface of the intelligent approval assistant system provided in this embodiment of the disclosure; Figure 6 A structural block diagram of an approval task processing device provided in an embodiment of this disclosure; Figure 7 This is a schematic diagram of the structure of an electronic device suitable for performing an approval task processing method, provided as an embodiment of the present disclosure. Detailed Implementation

[0013] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding; these should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0014] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0015] Figure 1 An exemplary system architecture 100 is shown, in which embodiments of the approval task processing methods, apparatuses, electronic devices, and computer-readable storage media of this disclosure can be applied.

[0016] like Figure 1 As shown, system architecture 100 may include target systems 101, 102, and 103, networks 104 and 105, terminal device 106, and server 107. Network 104 serves as the medium for providing a communication link between target systems 101, 102, and 103 and server 107. Network 105 serves as the medium for providing a communication link between terminal device 106 and server 107. Networks 104 and 105 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0017] Target systems 101, 102, and 103 are various independent business systems introduced during enterprise operations. Each system has its own independent database, can independently manage pending approval tasks within its own platform, and can upload its original approval task data to server 107 via network 104. Users can use terminal device 106 to interact with server 107 via network 105 to receive or send messages, etc. Various applications, such as intelligent approval assistants, can be installed on terminal device 106 and server 107 to enable information communication between them.

[0018] Terminal device 106 and server 107 can be hardware or software. When terminal device 106 is hardware, it can be various electronic devices with a display screen, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal device 106 is software, it can be installed in the aforementioned electronic devices, and can be implemented as multiple software programs or software modules, or as a single software program or software module; no specific limitation is made here. When server 107 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When server 107 is software, it can be implemented as multiple software programs or software modules, or as a single software program or software module; no specific limitation is made here.

[0019] Server 107 can provide various services through its built-in applications. Taking the intelligent approval assistant, which provides approval task processing services for multiple business scenarios to approval users, as an example, server 107 can achieve the following effects when running this intelligent approval assistant: First, it receives raw approval task data collected from target systems 101, 102, and 103 via network 104; Second, it cleans, standardizes, and classifies the raw approval task data to generate structured approval task data; Third, it performs multi-dimensional logical conflict detection on the structured approval task data to obtain conflict detection results. Multi-dimensional logical conflict detection includes amount conflict detection, time conflict detection, subject conflict detection, and clause conflict detection. The process involves seven steps: 1) Detecting conflicts at process nodes; 2) Identifying sensitive information in structured approval task data based on a pre-defined sensitive element library, and performing differentiated automatic desensitization processing on different categories of sensitive information according to the approval user's permission level, resulting in desensitized approval task data. Sensitive information includes monetary data, customer information, project secrets, qualification certificates, bank accounts, and core contract terms; 3) Extracting elements and verifying element completeness in the desensitized approval task data, generating element completeness verification results; 4) Generating reference approval opinions based on the element completeness verification results; 5) Generating centralized display information based on the original approval task data, conflict detection results, element completeness verification results, and reference approval opinions. The centralized display information includes summary approval task information, element completeness verification results, and reference approval opinions. If the conflict detection result indicates a conflict, the corresponding approval task is automatically marked with a conflict type and sent to terminal device 106 via network 105 for visual display.

[0020] It should be noted that, in addition to being obtained from the target systems 101, 102, and 103 via network 104, the original approval task data can also be pre-stored in the blockchain through various means. Therefore, when server 107 detects that the blockchain already stores this data, it can choose to obtain this data directly from the blockchain. In this case, the exemplary system architecture 100 may also exclude the target systems 101, 102, and 103 and network 104.

[0021] Because complex analysis of approval task data requires significant computing resources and power, the approval task processing methods provided in the subsequent embodiments of this disclosure are generally executed by a server 107 with strong computing power and abundant computing resources. Correspondingly, the approval task processing device is also generally located within the server 107. However, it should also be noted that when the terminal device 106 also possesses sufficient computing power and resources, the terminal device 106 can also complete the aforementioned calculations performed by the server 107 through its installed intelligent approval assistant, thereby outputting the same results as the server 107. Especially when multiple terminal devices with different computing capabilities exist simultaneously, but the intelligent approval assistant determines that the terminal device has strong computing power and abundant remaining computing resources, it can allow the terminal device to perform the aforementioned calculations, thereby appropriately reducing the computing pressure on the server 107. Accordingly, the approval task processing device can also be located within the terminal device 106. In this case, the exemplary system architecture 100 may also exclude the server 107 and the network 105.

[0022] It should be understood that Figure 1 The number of target systems, terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of target systems, terminal devices, networks, and servers.

[0023] Please refer to Figure 2 , Figure 2 A flowchart of an approval task processing method provided in this disclosure embodiment, wherein process 200 includes the following steps: Step 201: Collect raw approval task data from multiple target systems.

[0024] This step is intended to be approved by the entity responsible for executing the task processing method (e.g., Figure 1 The server 107 shown, or the intelligent approval assistant installed on server 107, can be accessed from multiple target systems (e.g., Figure 1 The target systems 101, 102, and 103 shown collect raw approval task data. This raw approval task data includes approval text data and approval attachment data. The text data includes task ID, task name, applicant, submission time, task details, payment remarks, etc. The attachment data includes invoice photos, scanned copies of contracts, acceptance reports, supporting documents, etc.

[0025] The entity executing this step must ensure data collection security through an authentication mechanism. This involves a dual authentication mechanism: encrypted storage of account passwords and dynamic token verification. Data collection must first be authenticated by the target system before execution. Simultaneously, the collection process is logged in real-time. The collection log includes information such as collection time, target system, data type, and data volume for easy traceability later.

[0026] In some embodiments, by calling the target system's approval task query API, encrypted approval user identity information is sent to the target system, and the original approval task data is received from the target system after the account password and dynamic token in the approval user identity information have been successfully authenticated. This method is applicable to target systems that provide open APIs.

[0027] In some embodiments, raw approval task data is obtained through a dedicated data interface established with the target system. This method utilizes a dedicated interface built in cooperation with the system developer, suitable for scenarios with extremely high requirements for data collection stability and security, and serves as an alternative to web scraping technology for data collection.

[0028] In some embodiments, raw approval task data pre-uploaded by the target system is obtained from the blockchain node. Each target system uploads its pending task data to the blockchain node in advance, and the executing entity retrieves it directly from the blockchain, improving data synchronization efficiency.

[0029] Furthermore, data acquisition employs a multi-mode strategy, including scheduled acquisition, real-time triggered acquisition, and manual triggered acquisition, to balance the real-time nature of data acquisition with system performance. Scheduled acquisition automatically executes acquisition operations at preset time intervals, defaulting to once every 30 minutes, with the interval adjustable by the administrator. Real-time triggered acquisition is triggered immediately via system message push when a new task awaiting approval is assigned to the approving user, ensuring timely synchronization of urgent tasks. Manual triggered acquisition allows the approving user to trigger immediate acquisition through operation commands via a visual interface, meeting personalized operational needs.

[0030] In some embodiments, the data acquisition process also incorporates a system adaptation adaptive learning mechanism: by constructing a target system interface feature library, it automatically identifies differences in interface protocols and data formats between different systems, dynamically generates adaptive acquisition scripts, and eliminates the need to develop separate adaptation modules for new systems. Simultaneously, for high-concurrency scenarios, a hybrid strategy of incremental acquisition and full verification is adopted, synchronizing only newly added or changed approval task data, and ensuring data integrity through hash value verification, significantly reducing system interaction pressure and improving acquisition efficiency.

[0031] Step 202: Clean the original approval task data, standardize the format, and classify the tasks to generate structured approval task data.

[0032] Based on step 201, this step aims to have the aforementioned executing entity perform data cleaning, format standardization, and task classification on the original approval task data, transforming the unstructured raw data into structured, analyzable approval task data, and storing it in the system database with hierarchical access control, laying the data foundation for subsequent element extraction and verification. The structured approval task data has a unified format and clear classification, and can be directly recognized and processed by AI analysis models.

[0033] The data cleaning process aims to remove invalid data, correct erroneous data, and filter core data to ensure data quality. This includes redundant data removal, erroneous data correction, and valid data filtering. Redundant data removal involves deleting invalid characters, automatically generated redundant prompts, and duplicate fields from the original approval task data. Erroneous data correction involves identifying and correcting garbled text, inconsistently formatted time fields, and missing key information from the original approval task data. Valid data filtering involves removing fields irrelevant to the approval process and retaining core approval data such as task description, applicant, attachment list, payment amount, and contract counterparty.

[0034] The format standardization process aims to unify the data formats of different target systems, eliminating analytical obstacles caused by format differences. This includes basic information standardization and text data structuring. Basic information standardization involves unifying the different formats of basic information from various systems into a pre-defined standard. This includes converting time fields to the internationally standardized date and time combination format "YYYY-MM-DD HH:MM:SS", standardizing applicant information into a string format of "Name-Department-Position", and standardizing attachment names into a string format of "Task ID-Attachment Type-File Name". Text data structuring involves splitting the unstructured task details from various systems into a fixed structure of "Task Title-Core Description-Key Data List", extracting structured information such as amount, date, project name, and contract counterparty, giving the text data analyzable structured characteristics.

[0035] The task classification stage aims to categorize raw approval task data according to approval business type, providing a basis for subsequent element extraction and verification by type. It employs a combination of keyword matching and machine learning classification. Initial classification is achieved through keyword matching, followed by final classification using a machine learning model, improving classification accuracy. Pre-set task type tags include basic tags such as payment approval, contract approval, and routine tasks, each configured with corresponding feature keywords. For example, "reimbursement," "request for payment," and "payment" correspond to the payment approval tag, while "contract," "cooperation," and "agreement" correspond to the contract approval tag.

[0036] In some embodiments, a multi-dimensional feature fusion mechanism is introduced in the task classification process. In addition to traditional keyword matching and Naive Bayes models, it also integrates the business-related features of approval tasks and historical classification feedback data to construct a dynamically weighted classification model. Simultaneously, it supports user-defined classification label systems, allowing for rapid training of classification models adapted to specific business scenarios with only a small number of labeled samples, improving classification accuracy, and automatically identifying new task types and triggering label creation reminders.

[0037] Step 203: Perform multi-dimensional logical conflict detection across elements and fields on the structured approval task data to obtain the conflict detection results.

[0038] Building upon step 202, this step aims to have the aforementioned executing entity perform multi-dimensional logical conflict detection across elements and fields on the structured approval task data, automatically identifying business data contradictions and avoiding erroneous approvals due to data conflicts. Multi-dimensional logical conflict detection includes: amount conflict detection for contract amount, payment amount, and invoice amount; time conflict detection for application time, contract signing time, acceptance time, and payment time; subject conflict detection for applicant, contract counterparty, and payee; clause conflict detection for text descriptions, attachment clauses, and approval opinions; and process node conflict detection for preceding approval nodes and process flow sequence. Upon completion of the detection, a structured conflict detection result containing the conflict location, conflict type, and conflict level is generated.

[0039] For example, amount conflict detection can identify discrepancies such as a contract amount of 1 million yuan being entered as 1.5 million yuan; time conflict detection can identify time-series discrepancies such as a payment date earlier than the contract signing date or an acceptance date later than the payment date; entity conflict detection can identify discrepancies such as a contract counterparty being Company A but the payee's account being Company B; clause conflict detection can identify clause discrepancies such as a task text describing a one-time payment but an attached contract clause stipulating installment payments; and process node conflict detection can identify process omissions such as submission directly to the finance payment node without department head approval. Through these multi-dimensional detections, various logical anomalies in approval tasks can be comprehensively identified, providing risk warnings for approval decisions.

[0040] Step 204: Based on the preset sensitive element library, identify sensitive information in the structured approval task data, and perform differentiated automatic desensitization processing on different categories of sensitive information according to the approval user's permission level to obtain desensitized approval task data.

[0041] Building upon step 202, this step aims to enable the aforementioned executing entity to accurately identify and hierarchically anonymize sensitive information in structured approval task data based on a pre-defined sensitive element database and the approval user's permission levels. This prevents the risk of sensitive information leakage and unauthorized access arising from the aggregation of data from multiple systems. The pre-defined sensitive element database includes identification rules for monetary data, customer information, project confidentiality, qualification certificates, bank accounts, and core contract clauses. Sensitive fields are located using a combination of keyword matching and semantic parsing. Different anonymization strategies are employed based on the approval user's permission level: high-privilege users retain necessary plaintext information, while low-privilege users undergo anonymization using mask replacement, partial hiding, and feature retention. The anonymized data does not affect subsequent element extraction, completeness verification, and logical conflict detection, while simultaneously achieving data usability without visibility.

[0042] For example, regarding sensitive information such as bank account details, high-privilege users can view the complete account number, while low-privilege users will only see "6222". The mask format is "1234"; for sensitive information such as contract amounts, high-privilege users can view the complete value, while low-privilege users only see "yuan"; for customer ID information, high-privilege users can view the complete number, while low-privilege users only see "110". 1234”; For confidential project information, high-privilege users can view the complete project content, while low-privilege users will only see “

Project Confidentiality Has Been De-identified

[0043] Step 205: Extract elements and verify element completeness from the de-identified approval task data, and generate element completeness verification results.

[0044] Building upon step 204, this step aims to have the aforementioned implementing entity extract elements and verify element completeness from the de-identified approval task data, generating element completeness verification results that include element problem annotations and specific descriptions. This is the core AI analysis step of this method, replacing manual identification and verification of approval elements, and improving the accuracy and efficiency of verification.

[0045] Element extraction involves extracting element information from anonymized approval task data and matching it with a pre-defined approval element database. The approval element database is a standardized database categorized by task type tags, storing corresponding required and optional elements; it serves as the standard for element verification. Element completeness verification compares the extracted element information with the required elements for the corresponding task type in the approval element database, determining if any elements are missing or incomplete, recording relevant issues, and generating structured verification results.

[0046] The approval element library supports both manual and automatic updates, as well as customized configurations to suit the business needs and approval rules of different enterprises. Manual updates allow administrators to add, modify, and delete approval types, required elements, and optional elements in the library. Automatic updates use an element recognition model to identify newly added high-frequency elements in historical approval task data, generate element addition prompts, and automatically update the approval element library after administrator confirmation. The element recognition model is trained using historical approval task data and corresponding high-frequency elements. Customized configurations allow enterprises to customize the configuration rules for required and optional elements for different approval types based on their own business characteristics; for example, listing business trip applications as a required element for expense reimbursement approvals.

[0047] For different approval types, the approval element library pre-defines corresponding required and optional elements, as shown in the following example configuration: Contract approval: Required elements include business overview, cooperation model, payment milestones, consistency with the meeting conclusions, and prior key approvals. Optional elements include other supplementary information.

[0048] Payment approval: Required elements include business overview, project stage, current payment milestones, current acceptance conclusions, whether subsequent milestones have deviated, and prior key approvals. Optional elements include other supplementary information.

[0049] Routine business approval: Required elements include theme, meal amount, number of people, applicant, and amount per person. Optional element is the amount exceeding the plan.

[0050] In some embodiments, the element extraction stage employs a dual-model collaborative extraction and contextual semantic completion mechanism: the extraction results of the general BERT model and the industry-specific pre-trained model are cross-validated, and for ambiguous elements (such as unclear amounts or dates), semantic completion is performed by combining the task context and historical expression habits of similar tasks. The element verification stage not only performs completeness verification but also introduces a reasonableness intelligent judgment function. Based on the enterprise's preset compliance rule library (such as expense reimbursement limits and payment ratio restrictions) and a threshold model built from historical approval data, it automatically judges the reasonableness of element values ​​and generates three types of verification results: "missing," "incomplete," and "unreasonable," achieving a deep verification upgrade from "element presence or absence" to "element quality." Step 206: Based on the element completeness verification results, generate reference approval opinions.

[0051] Building upon step 205, this step aims to have the aforementioned implementing entity generate targeted and actionable reference approval opinions based on the element completeness verification results, providing auxiliary basis for the approval user's approval decision-making, reducing the cost of manually sorting information, and lowering the difficulty of decision-making.

[0052] The generation of reference approval opinions is not a simple feedback of results, but a result optimized from multiple dimensions, combining element verification results, approval task type, business scenario, and approval user profile. Initial opinions are generated by first matching a preset basic template, and then adjusted according to personalized needs to ensure that the opinions align with the actual approval scenario and the decision-making habits of the approval users.

[0053] The preset basic templates are divided into three categories based on the verification results: normal approval templates, missing element templates, and incomplete element templates, covering all verification result scenarios. The normal approval template is suitable for approval tasks with all elements and complete information. The missing element template is suitable for approval tasks where essential elements have not been extracted. The incomplete element template is suitable for approval tasks where essential elements have been extracted but element information is missing. Each template is a populateable structured template, allowing users to supplement element information based on specific verification results.

[0054] The user profile for approvals is built based on the user's historical approval preferences, decision-making habits, and priorities. For example, some users prioritize cost control, while others prioritize process compliance. When generating reference approval opinions, the wording, information display priority, and problem highlighting dimensions of the opinions will be adjusted according to the user profile. For users who prioritize cost control, budget-related and monetary-related elements will be emphasized first. For users who prioritize process compliance, approval process and qualification certification-related elements will be emphasized first, achieving personalized decision-making assistance.

[0055] In some embodiments, the generation of reference approval opinions introduces a multi-dimensional context fusion mechanism: in addition to element verification results and user profiles, it automatically associates business background data of the task (such as the progress status of associated projects, contract performance, and applicant's historical approval records) and real-time compliance requirements of enterprises (such as the latest financial policies and industry regulatory provisions) to generate a four-dimensional opinion system that includes "problem identification - risk level - handling suggestions - reference cases". At the same time, it supports adaptive adjustment of opinion complexity, generating highly condensed core conclusion-type opinions for approval users and detailed step-by-step opinions for execution-level users, and can dynamically optimize the opinion expression style based on user feedback to improve the accuracy of decision support.

[0056] Step 207: Based on the original approval task data, conflict detection results, element completeness verification results, and reference approval opinions, generate centralized display information.

[0057] Building upon steps 203, 205, and 206, this step aims to have the aforementioned executing entity generate centralized display information based on the original approval task data, conflict detection results, element completeness verification results, and reference approval opinions, and send it to the terminal device (e.g., terminal device 106) for visual display. The centralized display information includes summary approval task information, element completeness verification results, and reference approval opinions. If the conflict detection result indicates a conflict, the corresponding approval task is automatically marked with a conflict type. The centralized display information is divided into summary approval task display information and complete display information for individual approval tasks, enabling centralized management of approval tasks across multiple systems, intuitive presentation of verification results, and convenient viewing of approval opinions. Simultaneously, it supports interactive approval operations, allowing approval users to complete all approval-related operations on a single interface without switching platforms.

[0058] The centralized information display design follows the principles of intuitiveness, convenience, and efficiency. It uses color labels to visually identify the status of elements, displays different types of information in separate zones, and supports operations such as sorting, searching, and jumping. It adapts to the office habits of approval users and greatly improves the convenience of approval operations.

[0059] In addition, the centralized information display also supports a variety of flexible and adaptable display solutions to meet the needs of different office scenarios and user habits: Firstly, it adopts a voice interaction method, allowing users to query the list of tasks to be done, the results of element verification, and reference approval opinions through voice commands. The system provides key information through voice broadcast, eliminating the need for manual operation. This is especially suitable for scenarios such as mobile office and when both hands are occupied, thus improving the convenience of operation.

[0060] Secondly, the system adds an intelligent push function. Based on the urgency of the approval task (such as large payments or urgent contract approvals) and the importance of the business, combined with the approval user's work habits (such as high-frequency approval periods), the system will proactively push the core information of important approval tasks to the user's frequently used devices (such as mobile apps, smartwatches, and pop-ups in office software), and mark the completeness of the elements to remind the user to prioritize the processing, thereby further improving the timeliness and response efficiency of task processing.

[0061] All of the above solutions can achieve the core objectives of "centralized information display and convenient interactive operation". They can be flexibly selected and deployed according to the actual office scenarios and user needs of enterprises, further expanding the scope of application and flexibility of implementation of this method.

[0062] The approval task processing method provided in this disclosure achieves a secure, automated, and intelligent upgrade in approval task processing through a closed-loop design encompassing "intelligent data collection from multiple systems - structured processing - multi-dimensional logical conflict detection - hierarchical and differentiated desensitization - AI-driven verification - personalized decision support - centralized visualization display." The technical effects are significant. Breaking down information silos across multiple systems significantly improves approval preparation efficiency: Through multi-mode data collection strategies and dual security authentication, non-intrusive integration of approval data from multiple target systems is achieved, eliminating the need for manual cross-platform logins and manual aggregation. This completely solves the pain points of fragmented tasks and cumbersome operations in traditional approval processes, allowing approval users to obtain all pending data in one stop, significantly reducing the time cost of the approval preparation stage.

[0063] Standardized data processing lays a solid foundation for AI analysis: Through data cleaning, format standardization, and precise classification, unstructured raw data is transformed into structured data with a unified format and clear classification, eliminating the analytical obstacles caused by differences in data formats across different systems. This provides high-quality data support for subsequent element extraction and verification, while structured storage facilitates data traceability and system maintenance.

[0064] Multi-dimensional logical conflict detection, proactively intercepting approval risks: The system performs multi-dimensional logical conflict detection on structured approval task data across elements and fields, including amount, time, subject, clauses, and process nodes. It automatically identifies data contradictions and process anomalies, provides early warnings of erroneous approval risks, and avoids approval errors caused by data conflicts and missing processes from the source, significantly improving approval compliance and accuracy.

[0065] Differentiated anonymization based on access permissions enhances data security and compliance: Based on a pre-defined sensitive element database, sensitive data such as amounts, customer information, project secrets, and bank accounts are automatically identified, and differentiated anonymization is performed according to the approval user's permission level. Without affecting normal approval judgment, it effectively prevents the risk of leakage of sensitive information and unauthorized access, and fully meets the enterprise's data security and privacy compliance requirements.

[0066] Intelligent replacement of manual verification improves approval accuracy and efficiency: With the help of a dynamically configurable approval element library and the semantic understanding capabilities of AI models, the system automatically completes the extraction and completeness verification of approval elements, replacing the traditional manual verification mode. This avoids the problems of element omission and judgment errors caused by human experience bias and fatigue, improves the accuracy of element extraction, significantly reduces the repetition rate of the approval process, and improves the efficiency of approval flow.

[0067] Personalized decision support reduces approval difficulty: Based on verification results and user profiles, targeted reference opinions are generated, clearly marking key issues and improvement directions. This reduces the time cost for users to sort through complex information and assess risks, especially for complex approval tasks, which can shorten the decision-making cycle. At the same time, it adapts to the decision-making habits of different users, improving decision satisfaction and quality.

[0068] Visualized interaction optimizes the approval experience while balancing compatibility and security: Through categorized and partitioned visual displays and user-friendly interactive design, approval tasks with logical conflicts are automatically marked with conflict types. Personalized operations such as sorting, searching, and navigation are supported, allowing approval users to intuitively grasp task status, conflict details, and core issues, achieving a one-stop experience of "centralized viewing - rapid decision-making - convenient operation." The solution supports multi-system adaptation and personalized configuration without requiring modifications to existing systems, reducing deployment costs. Furthermore, the entire process of data collection, desensitization, transmission, and storage is secure and controllable, meeting enterprise data compliance requirements and possessing broad application scenarios and practical value.

[0069] Further reference Figure 3 , Figure 3 A flowchart of another approval task processing method provided in this disclosure embodiment, wherein process 300 includes the following steps: Step 301: Call the target system's approval task query API, send the encrypted approval user identity information to the target system, and receive the original approval task data sent by the target system after the account password and dynamic token in the approval user identity information have been successfully authenticated.

[0070] This step aims to have the entity executing the approval task processing method call the target system's approval task query API, send encrypted approval user identity information to the target system, and receive the original approval task data sent by the target system after the account password and dynamic token in the approval user identity information have been successfully authenticated. It is one of the core methods of multi-system data collection, suitable for target systems that provide open APIs, and achieves non-intrusive data collection without modifying the original architecture of the target system.

[0071] The user identity information for approval includes encrypted account passwords and dynamic tokens. A dual authentication mechanism is used to ensure the security of data collection and prevent unauthorized access to approval data in the target system. The data transmission process uses HTTPS (Hypertext Transfer Protocol Secure) encryption to prevent data leakage during transmission. The collected original approval task data covers all pending approval task data in the target system that matches the approval user, including text data and attachment data, to ensure data integrity.

[0072] In some embodiments, a dynamic permission adaptation mechanism is introduced in the identity authentication process: based on the user's job level and business permission scope, the system automatically requests the minimum necessary data access permissions from the target system, only acquiring the pending task data corresponding to that user, thus avoiding the collection of data beyond the user's permissions. Simultaneously, the dynamic token adopts a time-sensitive and scenario-bound strategy. For highly sensitive systems (such as financial payment systems), the token validity period is set to 15 minutes, and it is only bound to the IP (Internet Protocol) address and device fingerprint of the current data collection scenario, further enhancing the security of identity authentication and increasing the rate of unauthorized access interception.

[0073] Step 302: Remove redundant data, correct erroneous data, and filter valid data from the original approval task data.

[0074] Building upon step 301, this step aims to have the aforementioned implementing entity remove redundant data, correct erroneous data, and filter valid data from the original approval task data. This data cleaning process removes invalid data, corrects erroneous data, and filters core data to ensure the quality of data for subsequent analysis.

[0075] Redundant data removal involves deleting invalid characters and duplicate fields, specifically including irrelevant system characters acquired during data collection, garbled placeholders, redundant prompt text automatically generated by the system, and duplicate approval information in different fields. Error data correction involves identifying and correcting garbled text and inconsistently formatted fields, focusing on correcting garbled task detail text, inconsistently formatted time fields, missing applicant information, and payment amount information. Valid data filtering involves removing irrelevant information and retaining core data related to the approval process. This includes removing system logs, advertising information, and non-approval fields unrelated to approval from the target system, while retaining core approval data such as task ID, task name, applicant, submission time, task details, attachment list, payment amount, contract counterparty, and approval records.

[0076] In some embodiments, the data cleaning process incorporates intelligent error correction and data traceability mechanisms: for common format errors (such as inconsistent date formats or mixed monetary units), the system automatically corrects them using correction rules learned from historical data; for complex errors (such as garbled characters or missing key information), the system marks the error type and associates it with the original data source, allowing users to manually correct errors and then automatically learn correction rules; simultaneously, a cleaning operation log is established to record the cleaning process for each piece of data (such as deleted redundant fields and corrected error content), facilitating data quality traceability and cleaning rule optimization, thereby improving the accuracy of data cleaning.

[0077] Step 303: Standardize the basic information and structure the text data of the original approval task data.

[0078] Based on step 302, this step aims to standardize the basic information and structure the text data of the original approval task data by the aforementioned implementing entity, namely the format standardization step, to unify the data format of different target systems, eliminate format differences, and generate standardized approval data.

[0079] The standardization of basic information includes converting time fields into a date-time combination format and standardizing applicant information and file names into string concatenation formats. Specifically, all time fields are uniformly converted to the internationally standardized format "YYYY-MM-DD HH:MM:SS", applicant information is uniformly standardized into a string concatenation format of "Name-Department-Position", and attachment names are uniformly standardized into a string concatenation format of "Task ID-Attachment Type-File Name", ensuring consistency in the basic information format across different target systems. Text data structuring involves splitting task details into string concatenation formats and extracting structured information. Specifically, unstructured task details are split into "Task Title-Core Description-Key Data List", extracting structured information such as amount, date, project name, contract counterparty, and acceptance conclusion, giving the text data structured characteristics that can be analyzed by AI models.

[0080] Step 304: First, use keyword matching to perform initial task classification on the original approval task data, and then use machine learning classification to perform final task classification on the original approval task data.

[0081] Based on step 303, this step aims to have the aforementioned executing entity first perform initial task classification on the original approval task data using keyword matching, and then perform final task classification on the original approval task data using machine learning classification. That is, the task classification stage improves the accuracy of task classification through initial screening and fine-tuning, providing a basis for subsequent extraction and verification of elements by type.

[0082] In some embodiments, the task classification step specifically includes: First, the original approval task data is matched with the feature keywords corresponding to multiple task type tags to determine the preliminary task type of the original approval task data. The preset task type tags include payment approval, contract approval, and routine tasks, etc., and each tag is configured with exclusive feature keywords. The executing entity performs similarity matching between the task title and core description of the original approval task data and the feature keywords. If the matching degree reaches a preset threshold, it is determined as the corresponding preliminary task type.

[0083] Then, the preliminary task type and original approval task data are input into the task classification model, which outputs the final task type of the original approval task data. The task classification model is trained using historical approval task data and corresponding historical task types on a Naive Bayes classification model. The model trained on historical data can accurately classify approval tasks by combining semantic features, correcting keyword matching errors, and improving classification accuracy.

[0084] Step 305: Perform multi-dimensional logical conflict detection across elements and fields on the structured approval task data to obtain the conflict detection results.

[0085] Step 306: Based on the preset sensitive element library, identify sensitive information in the structured approval task data, and perform differentiated automatic desensitization processing on different categories of sensitive information according to the approval user's permission level to obtain desensitized approval task data.

[0086] In this embodiment, the specific operations of steps 305-306 have been described. Figure 2 Steps 203-204 in the illustrated embodiments are described in detail and will not be repeated here.

[0087] Step 307: Extract the element information corresponding to the approval element database from the de-identified approval task data.

[0088] Building upon step 306, this step aims to have the aforementioned executing entity extract element information corresponding to the approval element database from the de-identified approval task data. This is the core step of element analysis, achieving accurate extraction of approval elements through a combination of various AI technologies, covering approval data of different types and scenarios.

[0089] The approval element library stores required and optional elements categorized by task type tags. It supports both manual and automatic updates. Manual updates include adding, modifying, and deleting elements. Automatic updates utilize an element recognition model to identify new elements in historical approval task data. This model is trained using historical approval task data and corresponding high-frequency elements. Furthermore, the approval element library allows enterprises to customize its configuration to meet their specific business needs and adapt to different approval rules.

[0090] In some embodiments, a keyword extraction algorithm and a semantic analysis model are used to extract textual elements from the task titles and core descriptions in the anonymized approval task data. These textual elements are then stored in a concatenated string format to generate element information. The semantic analysis model is trained using BERT (Bidirectional Encoder Representations from Transformers) with historical approval task data and corresponding textual elements. The keyword extraction algorithm employs TF-IDF (Term Frequency-Inverse Document Frequency), and the combination of these two methods enables multi-dimensional element extraction. TF-IDF facilitates the rapid extraction of high-frequency keyword elements, while the BERT model leverages its bidirectional semantic understanding capabilities to extract semantically related elements from context, overcoming the limitations of traditional keyword extraction and improving the accuracy of element extraction. The extracted textual elements are stored in a pre-defined concatenated string storage structure: "Task ID - Approval Type - Element Name - Element Value - Extraction Source," facilitating subsequent verification, traceability, and display.

[0091] In some embodiments, an industry-specific pre-trained model corresponding to the task type of the original approval task data is used to extract textual elements from the task titles and core descriptions in the anonymized approval task data. These textual elements are then stored in a string concatenated format to generate element information. Different task types correspond to different industry-specific pre-trained models. These models are trained using historical approval task data and corresponding textual elements for the corresponding task type, such as a NLP (Natural Language Processing) model for financial approvals or an OCR (Optical Character Recognition) model for invoice recognition. The industry-specific pre-trained model is trained on features specific to a particular approval type, improving the accuracy of extracting specific approval elements. For image attachment data, the characters in the image are first recognized and converted into text data using an OCR model, and then element information is extracted using the aforementioned model, achieving a full-link extraction of "image-text-element".

[0092] In some embodiments, a cross-modal element fusion extraction function is added to the element extraction process: For approval tasks that include multimodal attachments such as text, images, and tables, text information in images / tables is extracted using OCR technology, the logical relationship of the data is restored using a table structure recognition algorithm, and then fused with the task text data to extract cross-modal related elements (such as consistency verification between the amount in the invoice image and the payment amount in the text); at the same time, an element extraction confidence assessment mechanism is introduced to score the reliability of the extraction results. Elements with a confidence score lower than a preset threshold are marked as "to be confirmed" and the user is prompted to supplement the verification, further improving the reliability of element extraction and the accuracy of core element extraction.

[0093] Step 308: Compare the element information with the required elements of the corresponding task type tag in the approval element library to generate element completeness verification results.

[0094] Based on step 307, this step aims to have the aforementioned executing entity compare the element information with the required elements of the corresponding task type label in the approval element library, generate element completeness verification results, i.e. element completeness verification step, verify the extracted elements according to a unified standard, identify issues such as missing elements and incomplete elements, and generate structured verification results.

[0095] In some embodiments, the element completeness verification result step specifically includes: First, the element information is compared one by one with the essential elements of the corresponding task type label. If any essential elements are not extracted, they are marked as missing elements. The executing entity retrieves the corresponding essential element list from the approval element library according to the final type of the approval task. The extracted element information is compared one by one with the list. Essential elements not appearing in the element information are marked as missing elements. For example, if the element "Conclusion of this Acceptance" is not extracted in a payment approval task, then this element is marked as missing.

[0096] Next, the completeness of the extracted essential elements is checked. If any element is missing information, it is marked as incomplete. For the extracted essential elements, the completeness and standardization of their element values ​​are checked. If the element value has missing information, the unit is not marked, the attachment is empty, or the content is not standard, it is marked as incomplete. For example, the "Payment Amount" element is marked as incomplete if it only shows "200" without specifying the currency unit, or the "Qualification Certificate" element has an empty attachment.

[0097] Finally, the names and problem descriptions of missing and incomplete elements are recorded to generate the element completeness verification results. The verification results are structured data, clearly recording the name, problem type (missing / incomplete), and specific problem description of each problem element, providing an accurate basis for the generation of subsequent reference approval opinions.

[0098] In some embodiments, the element verification process adds a dynamic rule engine and associated element verification function: the approval element library supports configuring conditional verification rules according to business scenarios, and the rules support logical operator combinations, which can be flexibly adjusted without modifying the code; at the same time, it supports associated element verification, automatically identifies the logical relationships between elements, and generates early warning prompts when logical contradictions are found, realizing the upgrade from "single-point element verification" to "full-link logical verification", improving the coverage of approval risk identification. Step 309: Match the element completeness verification results with multiple basic templates to determine the target basic template that matches the element completeness verification results.

[0099] Based on step 308, this step aims to have the aforementioned executing entity match the element completeness verification results with multiple basic templates to determine the target basic template that matches the element completeness verification results. This is the first step in generating reference approval opinions, and the standardized generation of opinions is achieved through template matching.

[0100] The basic templates include a normal approval template, a template with missing elements, and a template with incomplete elements, each corresponding to different verification result scenarios. The normal approval template matches verification results with complete element information and no issues. The template with missing elements matches verification results with missing element issues. The template with incomplete elements matches verification results with incomplete element issues. If both missing and incomplete element issues exist in the verification result, the template with missing elements is matched, and the relevant information for the incomplete element is added to the template. Each basic template is a structured, fillable template, reserving space for filling in element issue information to ensure that the subsequently generated opinions are structured and standardized.

[0101] Step 310: Fill in the target basic template with content based on the element completeness verification results and generate initial approval reference opinions.

[0102] Building upon step 309, this step aims to have the aforementioned implementing entity fill in the target basic template with content based on the element completeness verification results, generate initial approval reference opinions, and transform the structured verification results into natural language opinion content to ensure that the opinions can clearly and accurately reflect the problems found in the element verification.

[0103] The core of content filling is to accurately fill the element problem names and specific problem descriptions from the verification results into the reserved positions of the target basic template. For example, in the element missing template, fill the missing element names one by one into "Missing the following essential elements", and fill the incomplete element names and specific problem descriptions into "The following element information is incomplete" to ensure that the initial approval reference opinion can present all element problems completely and clearly, without omissions or errors.

[0104] Step 311: Based on the task type and approval user profile of the original approval task data, adjust the initial approval reference opinions and generate reference approval opinions.

[0105] Based on step 310, this step aims to have the aforementioned executing entity adjust the initial approval reference opinions based on the task type and approval user profile of the original approval task data, generate the final reference approval opinions, realize the personalization and scenario-based optimization of the opinions, and make the reference opinions more in line with the actual approval needs.

[0106] The approval user profile is built based on the user's historical approval preferences, decision-making habits, and priorities. The executing entity adjusts the wording of its comments, information display priorities, and key issues based on the type of approval task (e.g., payments, contracts, routine tasks) and the approval user profile. For users who prioritize cost control, budget, amount, and expense-related elements are emphasized. For users who prioritize process compliance, elements related to the approval process, qualification certificates, and acceptance conclusions are emphasized. Appropriate professional wording is used for different types of approval tasks; for example, more rigorous business language is used for contract approvals, while more concise and general language is used for routine task approvals.

[0107] Step 312: Generate summary display information for approval tasks.

[0108] Building upon step 311, this step aims to generate summary display information of approval tasks by the aforementioned executing entities. It is an important component of centralized information display, enabling centralized presentation of pending approval tasks from multiple systems. This allows approval users to quickly browse the core information of all pending tasks without having to log in to different systems across platforms.

[0109] The summary display fields include task name, approval type, applicant user, submission time, element completeness status, and processing priority, covering all the core information needed for quick browsing by approval users. The summary display information can be sorted by submission time, approval type, processing priority, and keyword search, allowing approval users to switch freely according to their needs and improve browsing efficiency. The status indicators for the summary display information use different color labels to distinguish different element statuses: green indicates elements are complete and reasonable, while yellow indicates elements are missing / incomplete, making it intuitive and eye-catching, allowing approval users to quickly identify the element status of tasks and prioritize important tasks.

[0110] Processing priorities are automatically assigned by the executing entity based on the urgency and importance of the approval task. For example, approvals for large payments and urgent contracts are marked as high priority, while approvals for routine administrative tasks are marked as normal priority, making it easier for users to process tasks according to priority.

[0111] In some embodiments, the summary display stage introduces intelligent task sorting and personalized view customization functions: in addition to traditional sorting methods, it supports sorting by a composite weight of approval urgency, business relevance, and user attention preferences, and automatically puts tasks that are highly relevant to the user's current business and are about to expire at the top; at the same time, it allows users to customize the display view, select the fields to be displayed, the status identification method (color / icon), and grouping rules (by approval type / urgency), and supports saving and switching views to adapt to the operating habits of different users and improve the efficiency of task search.

[0112] Step 313: Generate complete display information and verification results for a single approval task.

[0113] Building upon step 312, this step aims to generate complete display information and verification results for a single approval task from the aforementioned executing entity. It is a core component of the centralized display information, providing approval users with detailed information, element verification results, and reference approval opinions for a single approval task, and supporting in-depth viewing and approval operations.

[0114] The complete information display and verification results are divided into sections: basic task information, approval element list, verification result details, and approval reference opinions. The information in each section is independent yet interconnected, allowing approval users to view information according to their needs. The approval element list is categorized into required and optional elements, with each element's value and extraction source clearly indicated. Extraction sources include task detail text, attachments, and approval records, facilitating users' ability to trace the original source of elements. Verification result details include the names and descriptions of missing or incomplete elements, clearly presenting all element issues. Users can also jump to the corresponding target system to view the original approval task data; clicking the link allows users to directly view the original document details in the target system without manually logging in.

[0115] In addition, the complete display interface of a single approval task also supports interactive approval operations. Approving users can directly perform approval operations such as agreeing, rejecting, and returning for supplementation on the interface. The operation instructions will be synchronized to the corresponding target system, realizing one-stop approval of "view-decision-operation" and greatly improving approval efficiency.

[0116] The approval task processing method provided in this disclosure, through more refined process design and technology selection, combined with logical conflict detection and permission-based anonymization, further enhances the accuracy, security, efficiency, and practicality of approval task processing. The specific technical effects are as follows: Secure and efficient multi-system data collection ensures data integrity and timeliness: The collection method adopts API integration and dual authentication to achieve non-intrusive data acquisition. Combined with HTTPS encrypted transmission, it ensures the security and authorization compliance of the approval data collection process. At the same time, through a multi-mode collection strategy, it takes into account both data real-time performance and system performance, ensuring timely synchronization of urgent tasks and laying a reliable data foundation for subsequent processing.

[0117] Refined data processing improves data quality and enhances the effectiveness of AI analysis: Through a refined cleaning process that removes redundant data, corrects erroneous data, and filters effective data, as well as standardized processing of basic information and structured text data, the problems of inconsistent data formats and messy information across different systems are thoroughly solved. The generated high-quality structured data can be directly recognized by AI models, providing solid support for subsequent conflict detection, desensitization, feature extraction and verification, and significantly improving the accuracy of feature extraction and verification.

[0118] Precise classification and professional model empowerment enhance the targeting of element extraction: A two-level classification strategy using keyword matching and Naive Bayes model ensures the accuracy of approval task classification, providing a precise basis for subsequent conflict detection, desensitization processing, and element extraction; combined with TF-IDF algorithm, BERT semantic analysis model, and industry-specific pre-trained models (such as financial approval NLP model and invoice OCR model), it achieves accurate extraction of multiple types of approval elements, including text and images, with a significantly higher accuracy rate for element extraction in specific industry approval scenarios than general models.

[0119] Multi-dimensional conflict detection and permission desensitization build a security defense for approvals: Through multi-dimensional logical conflict detection across elements and fields, abnormal conflicts such as amounts, times, subjects, terms, and process nodes are comprehensively identified to achieve risk interception in advance; combined with a permission-based differentiated desensitization mechanism, sensitive information is accurately identified and desensitized in a hierarchical manner, taking into account both data availability and security, and ensuring the reliability of approvals from the dual dimensions of process risk and data security.

[0120] Dual verification and personalized opinion generation enhance decision support value: Through dual verification logic of missing elements and information completeness, the system comprehensively covers approval element issues and generates structured verification results; the three-level opinion generation mechanism based on basic template filling, task type adaptation and user profile adjustment makes the reference opinions more targeted and operable, which not only reduces the difficulty of approval decision-making, but also guides the standardized advancement of the approval process and reduces decision-making errors caused by information asymmetry.

[0121] Layered display and convenient interaction achieve a dual improvement in approval efficiency and user experience: Layered information is generated, displaying both summary and individual task details. The summary page synchronously displays element status and conflict type indicators, supporting multi-dimensional sorting and visual status indicators for quick filtering of key tasks. The details page displays basic information, element lists, verification results, and reference opinions by module, supporting jumps to view original documents. Combined with direct approval operation functions, it achieves a one-stop approval loop. The solution balances technical expertise with practicality, is compatible with existing system architectures of enterprises of different sizes, and has robust data security and risk control mechanisms. It can effectively reduce the transformation costs and risks of digital approval for enterprises, and the protective barriers formed by core technologies further ensure the uniqueness and market competitiveness of the solution.

[0122] Further reference Figure 4 , Figure 4 This diagram illustrates the overall framework of the intelligent approval assistant system provided in this embodiment. The diagram shows the six-layer architecture of the intelligent approval assistant system implementing the approval task processing method of this disclosure, including a data acquisition layer, a data processing layer, a conflict detection layer, a data anonymization layer, an AI analysis layer, and a result display layer. Each layer operates independently yet works collaboratively. The system also includes a system database and an approval user terminal, enabling end-to-end processing and display of approval task data.

[0123] The system comprises several layers: The data acquisition layer serves as the system's data source, securely and efficiently collecting raw approval task data from multiple enterprise business systems (System 1, System 2, System 3, and System 4) through API integration, web scraping, and identity authentication. It supports scheduled, real-time, and manual collection modes, forming the system's data foundation. The data processing layer receives the raw approval task data from the data acquisition layer and transforms unstructured raw data into structured approval task data through cleaning, standardization, and classification. This data is then stored in the system database, providing standardized data for the AI ​​analysis layer. The conflict detection layer performs multi-dimensional logical conflict detection across elements and fields in the structured approval task data, obtaining conflict detection results including conflict location, type, and level, enabling automatic identification and proactive risk interception of approval data contradictions. The data anonymization layer, based on a preset sensitive element library and approval user permission levels, identifies sensitive information in the structured data and performs differentiated automatic anonymization processing, resulting in anonymized approval task data, ensuring that sensitive data is not leaked or access violated. The AI ​​analysis layer is the core functional layer of the system. Through core technologies such as element extraction, element completeness verification, and opinion generation from anonymized data, it achieves intelligent extraction of approval elements, completeness verification, and generation of reference approval opinions, representing the core manifestation of the system's intelligence. The results display layer receives the processing results from the conflict detection layer, data anonymization layer, and AI analysis layer. It presents the summary information of approval tasks, conflict detection results, element completeness verification results, and reference opinions to the approval user's terminal through a visual interface. It automatically marks conflicting tasks with conflict types and supports operations such as summarizing, detailing, and interacting, serving as the interaction interface between the system and the user.

[0124] The system database provides data storage services for the entire system, storing structured approval task data, approval element library, historical approval data, user profile data, sensitive element library, conflict detection rules, etc. It adopts MySQL database, supports data backup and recovery, and adopts security mechanisms such as hierarchical permission management and sensitive information de-identification. The approval user terminal is the operating terminal for approval users, including smartphones, computers, etc. Approval users can view centrally displayed information and perform approval operations through this terminal, realizing the convenience of approval work.

[0125] Data transmission between layers is achieved through encrypted data interfaces. The transmission process uses encryption protocols such as HTTPS to ensure data transmission security. The six-layer architecture design makes the responsibilities of each module of the system clear and the coupling low, which facilitates the maintenance, upgrading and expansion of the system.

[0126] Further reference Figure 5 , Figure 5 This diagram illustrates the approval result display interface of the intelligent approval assistant system provided in this embodiment. Taking an approval user processing a pending approval task as an example, the diagram visually demonstrates the centralized display information generated by the approval task processing method of this disclosure on the approval user's terminal. The corresponding complete execution flow and display logic are as follows: Data collection: The intelligent approval assistant collects payment tasks from the target system 1 through the API interface according to the scheduled collection rules. At the same time, it collects data from the corresponding application through API connection. After dual authentication of account password encryption storage and dynamic token verification, it obtains the original data of the task and another task. Data Processing: Cleaning: Remove redundant system prompt text and correct incorrectly formatted time fields in payment details; Standardization: Filter invalid field information and extract core content such as project name, total project budget, project background, project objectives, contract name, contract amount, contract remarks, contract counterparty, current payment amount, current payment remarks, payee, and approval records; Classification: Determine the task as a "payment" category using a combination of keyword matching and machine learning classification. Conflict detection: Perform multi-dimensional logical conflict detection on structured payment approval data, including amount, time, subject, terms, and process nodes, to identify data contradictions and process anomalies and generate conflict detection results; Data anonymization: Sensitive information such as contract amount, payee account, and project confidentiality is anonymized according to the current approval user's permissions and presented in an anonymized form on the display interface, without affecting the approval judgment and avoiding the leakage of sensitive information; AI Analysis: Element Extraction: Business Overview: Extract relevant information about XX product, project necessity assessment, project economic assessment, project benefits, contract counterparty, contract amount, core contract content, etc., and summarize it into a structured description of no more than 100 words (represented in the interface as "Contract name is payment of XX expenses in XX year XX quarter, contract counterparty is XX Co., Ltd., payment amount is XX yuan"); Project Stage: According to the rules, only the current actual progress field of the business is extracted. Since no relevant information was obtained, it is marked as "missing"; Current Payment Milestone: Extract XX yuan from the "Payment Amount" field and extract the milestone name "XX expenses in XX year XX quarter" from the "Payment Remarks" field. Since the approval form clearly marks it as "initial payment", only the above two pieces of information are displayed; Acceptance Conclusion: According to the rules, business progress information on the achievement of payment conditions should be extracted. Since no relevant valid content was obtained, it is marked as "missing"; Subsequent Whether milestone times deviate from the plan: Based on the original data, it is clearly stated that "subsequent milestone times have not deviated from the plan and are proceeding according to schedule," without any additional inferences; Preceding key approvals: Approval records are extracted, and the approvers "Wang XX from XX Office, Zhang XX from XX Department" and their respective department abbreviations are output. Since both are only marked "Agreed" without additional approval comments or separate descriptions of the comments, they are summarized into a concise statement that meets the requirements; Other supplementary information: The key information related to the decision "Please select XX to bear all bank handling fees for this payment" is extracted; Completeness verification: After comparing the essential elements of payment-related tasks in the approval element library, it was found that "Project stage" and "Conclusion of this acceptance" did not extract valid information, and were marked as "Core information missing," and the missing items were highlighted in red; Conflict detection results are displayed synchronously: The conflict types and conflict descriptions detected for this payment amount, contract amount, payee entity, process nodes, etc., are visualized; Results Display: Summary Page: The summary page displays the current task and another pending task, sorted by submission time. Tasks with missing elements or conflicts are highlighted with different color labels. Approval can be completed directly on the summary page. View Details: The approving user can click on the task title to access the details. Figure 5 The complete display interface shown has the following information: the top of the interface displays the number of pending tasks and the conflict / missing status; the core area displays the task type and name; the extracted elements are displayed in a structured manner according to fixed dimensions; missing elements are highlighted in red to indicate "missing"; if there is a logical conflict, the conflict type and conflict description are displayed separately on the interface; the bottom of the interface has two-way approval interaction buttons for "reject" and "agree", and it also supports jumping to the target system 1 to view the complete document details.

[0127] The interface features a clean and efficient layout, tailored to the work habits of approval users. Its core presentation logic is as follows: The top displays an overall prompt for pending approval tasks, clearly indicating the number of pending tasks (1 pending approval item) and the status of the element issues (1 item lacking core content). The core area prominently displays the approval task type [Payment] and its name, such as "XX Year XX Quarter XX Expense Payment," clearly identifying the task's subject. The extracted approval elements are structured and displayed according to fixed dimensions such as business overview, project stage, current payment milestone, current acceptance conclusion, whether subsequent milestones have deviated, prior key approvals, and other supplementary information. Missing elements such as "project stage" and "current acceptance conclusion" are marked with "[Missing]" and highlighted in red, allowing approval users to quickly locate the missing core information. At the same time, the results of multi-dimensional logical conflict detection are displayed intuitively, making it easier for approval personnel to quickly identify data contradictions and process anomalies. The interface features two-way approval interaction buttons at the bottom: "Reject" and "Agree". Approving users can directly click to complete the operation. It also supports auxiliary functions such as "Collapse / Expand Details" and "Jump to Original Document Details".

[0128] This interface enables structured presentation of approval element information, precise labeling of element completeness issues, intuitive display of logical conflicts, and one-stop execution of approval operations. It eliminates the need for users to switch platforms or manually sort out information, significantly improving the convenience and efficiency of approval operations. It is a complete application of the publicly disclosed approval task processing method from data collection to result implementation.

[0129] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an approval task processing device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0130] like Figure 6As shown, the approval task processing device 600 of this embodiment may include: a data acquisition module 601, a data processing module 602, a conflict detection module 603, a data desensitization module 604, an element verification module 605, an opinion generation module 606, and an information generation module 607. The data acquisition module 601 is configured to acquire raw approval task data from multiple target systems; the data processing module 602 is configured to perform data cleaning, format standardization, and task classification on the raw approval task data to generate structured approval task data; the conflict detection module 603 is configured to perform multi-dimensional logical conflict detection across elements and fields on the structured approval task data to obtain conflict detection results. Multi-dimensional logical conflict detection includes amount conflict detection, time conflict detection, subject conflict detection, clause conflict detection, and process node conflict detection; the data desensitization module 604 is configured to identify sensitive information in the structured approval task data based on a preset sensitive element library and perform differentiated automatic processing on different categories of sensitive information according to the approval user's permission level. The data is anonymized to obtain anonymized approval task data. Sensitive information includes monetary data, customer information, project confidentiality, qualification certificates, bank accounts, and core contract terms. Element verification module 605 is configured to extract elements and verify element completeness from the anonymized approval task data, generating element completeness verification results. Opinion generation module 606 is configured to generate reference approval opinions based on the element completeness verification results. Information generation module 607 is configured to generate centralized display information based on the original approval task data, conflict detection results, element completeness verification results, and reference approval opinions. The centralized display information includes summary information of approval tasks, element completeness verification results, and reference approval opinions. If the conflict detection result indicates a conflict, the corresponding approval task is automatically marked with a conflict type.

[0131] In this embodiment, the specific processing and technical effects of the following modules in the approval task processing device 600—data acquisition module 601, data processing module 602, conflict detection module 603, data desensitization module 604, element verification module 605, opinion generation module 606, and information generation module 607—can be found in reference to [reference needed]. Figure 2 The relevant descriptions of steps 201-207 in the corresponding embodiments will not be repeated here.

[0132] In some optional implementations of this embodiment, the data acquisition module 601 is further configured to: call the target system's approval task query application programming interface (API) to send encrypted approval user identity information to the target system, and receive the original approval task data sent by the target system after the account password and dynamic token in the approval user identity information have been successfully authenticated; or obtain the original approval task data through a dedicated data interface established with the target system; or obtain the original approval task data pre-uploaded by the target system from a blockchain node.

[0133] In some optional implementations of this embodiment, the data processing module 602 is further configured to: perform redundant data removal, erroneous data correction, and valid data filtering on the original approval task data; redundant data removal involves deleting invalid characters and duplicate fields; erroneous data correction involves identifying and correcting garbled text and fields with inconsistent formats; and valid data filtering involves removing irrelevant information and retaining core data related to the approval process. The original approval task data is also subjected to basic information standardization and text data structuring. Basic information standardization includes converting time fields into date-time combination formats and standardizing applicant information and document names into string concatenation formats. Text data structuring involves splitting task details into string concatenation formats and extracting structured information. Finally, the original approval task data is first classified using keyword matching, and then classified using machine learning.

[0134] In some optional implementations of this embodiment, the data processing module 602 is further configured to: match the original approval task data with the feature keywords corresponding to multiple task type labels respectively to determine the preliminary task type of the original approval task data; input the preliminary task type and the original approval task data into the task classification model, and output the final task type of the original approval task data. The task classification model is obtained by training a Naive Bayes classification model using historical approval task data and corresponding historical task types.

[0135] In some optional implementations of this embodiment, the element verification module 605 is further configured to: extract element information corresponding to the approval element library from the de-identified approval task data; the approval element library stores the corresponding required elements and optional elements according to task type tags; the approval element library supports manual and automatic updates; manual updates include manually adding, modifying, and deleting elements; and automatic updates involve identifying new elements in historical approval task data through an element recognition model, which is trained using historical approval task data and corresponding high-frequency elements; and compare the element information with the required elements corresponding to the task type tags in the approval element library to generate an element completeness verification result.

[0136] In some optional implementations of this embodiment, the element verification module 605 is further configured to: extract textual elements from the task titles and core descriptions in the desensitized approval task data using a keyword extraction algorithm and a semantic analysis model; the semantic analysis model is obtained by training a bidirectional encoder representation model using historical approval task data and corresponding textual elements, or by using an industry-specific pre-trained model corresponding to the task type of the original approval task data to extract textual elements from the task titles and core descriptions in the structured approval task data; different task types correspond to different industry-specific pre-trained models, and the industry-specific pre-trained models are obtained by training a dedicated recognition model using historical approval task data of the corresponding task type and corresponding textual elements; and store the textual elements in a string concatenation format to generate element information.

[0137] In some optional implementations of this embodiment, the element verification module 605 is further configured to: compare the element information with the essential elements of the corresponding task type label one by one; if there are any necessary elements that have not been extracted, mark them as missing elements; perform information integrity checks on the extracted necessary elements; if there are any elements with missing information, mark them as incomplete elements; record the names and problem descriptions of missing and incomplete elements, and generate element completeness verification results.

[0138] In some optional implementations of this embodiment, the opinion generation module 606 is further configured to: match the element completeness verification result with multiple basic templates to determine the target basic template that matches the element completeness verification result, the basic templates including normal approval templates, element missing templates and element incomplete templates; fill the target basic template with content based on the element completeness verification result to generate initial approval reference opinions; adjust the initial approval reference opinions based on the task type and approval user profile of the original approval task data to generate reference approval opinions, the approval user profile being constructed based on the approval user's historical approval preferences, decision-making habits and focus.

[0139] In some optional implementations of this embodiment, the information generation module 607 is further configured to: generate summary display information for approval tasks, the display fields of which include task name, approval type, applicant user, submission time, element completeness status, and processing priority; the sorting method of the summary display information includes sorting by submission time, sorting by approval type, sorting by processing priority, and sorting by keyword search; the status identifier of the summary display information is to use different color labels to distinguish different element statuses; generate complete display information and verification results for a single approval task, the complete display information and verification results are displayed in sections according to task basic information, approval element list, verification result details, and approval reference opinions; the approval element list is categorized and displayed according to required elements and optional elements, and the element value and extraction source of each element are marked; the verification result details include the element name and problem description of missing or incomplete elements, and support jumping to the corresponding target system to view the original approval task data.

[0140] According to embodiments of this disclosure, this disclosure also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to implement the approval task processing method described in any of the above embodiments.

[0141] According to embodiments of this disclosure, this disclosure also provides a readable storage medium storing computer instructions that enable a computer to execute the approval task processing method described in any of the above embodiments.

[0142] According to embodiments of this disclosure, this disclosure also provides a computer program product that, when executed by a processor, can implement the approval task processing method described in any of the above embodiments.

[0143] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0144] like Figure 7As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded into random access memory (RAM) 703 from storage unit 708. The RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0145] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0146] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as the approval task processing method. For example, in some embodiments, the approval task processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the approval task processing method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the approval task processing method by any other suitable means (e.g., by means of firmware).

[0147] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0148] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0149] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0150] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0151] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0152] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0153] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0154] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for processing approval tasks, characterized in that, include: Collect raw approval task data from multiple target systems; The original approval task data is cleaned, formatted, and classified to generate structured approval task data. The task classification includes first using keyword matching to perform initial task classification on the original approval task data, and then using a dynamic weight classification model that integrates the business association features of approval tasks and historical classification feedback data to perform final task classification on the original approval task data. The structured approval task data is subjected to multi-dimensional logical conflict detection across elements and fields to obtain conflict detection results. The multi-dimensional logical conflict detection includes: amount conflict detection for contract amount, payment amount, and invoice amount; time conflict detection for application time, contract signing time, acceptance time, and payment time; subject conflict detection for applicant, contract counterparty, and payee; clause conflict detection for text description and attachment clauses and approval opinions; and process node conflict detection for prior approval nodes and process flow sequence. Sensitive information is identified in the structured approval task data based on a preset sensitive element database, and different automatic desensitization processing is performed on different categories of sensitive information according to the approval user's permission level to obtain desensitized approval task data. The sensitive information includes monetary data, customer information, project secrets, qualification certificates, bank accounts and core contract terms. The anonymized approval task data undergoes element extraction and element completeness verification to generate element completeness verification results. Element extraction includes cross-validating the extraction results from a general BERT model with those from an industry-specific pre-trained model. For ambiguous elements, semantic completion is performed by combining the task context and historical expression habits of similar task elements. Based on the element completeness verification results, reference approval opinions are generated. These reference approval opinions are a four-dimensional opinion system that integrates element verification results, user profiles, business background data of related tasks, and real-time enterprise compliance requirements, including problem identification, risk level, handling suggestions, and reference cases. Based on the original approval task data, the conflict detection results, the element completeness verification results, and the reference approval opinions, centralized display information is generated. This centralized display information includes approval task summary information, the element completeness verification results, and the reference approval opinions. If the conflict detection results indicate a conflict, the corresponding approval task is automatically labeled with a conflict type.

2. The method according to claim 1, characterized in that, The collection of raw approval task data from multiple target systems includes: Call the target system's approval task query application programming interface (API) to send encrypted approval user identity information to the target system, and receive the original approval task data sent by the target system after successful dual authentication of the account password and dynamic token in the approval user identity information; or The original approval task data is obtained through a dedicated data interface established with the target system; or Obtain the original approval task data that the target system has pre-uploaded from the blockchain node.

3. The method according to claim 1, characterized in that, The process of cleaning and standardizing the original approval task data includes: The original approval task data is subjected to redundant data removal, error data correction, and effective data filtering. Redundant data removal involves deleting invalid characters and duplicate fields. Error data correction involves identifying and correcting garbled text and fields with inconsistent formats. Effective data filtering involves removing irrelevant information and retaining core data related to the approval process. The original approval task data is standardized in terms of basic information and structured in terms of text data. The standardization of basic information includes converting the time field into a date and time combination format and standardizing the applicant information and file name into a string concatenation format. The structured text data involves splitting the task details into a string concatenation format and extracting the structured information.

4. The method according to claim 3, characterized in that, The process involves first performing initial task classification on the original approval task data using keyword matching, and then performing final task classification on the original approval task data using a dynamic weighted classification model that integrates the business association features of the approval tasks and historical classification feedback data. This includes: The original approval task data is matched with the feature keywords corresponding to multiple task type tags to determine the preliminary task type of the original approval task data. The preliminary task type and the original approval task data are input into the task classification model, and the final task type of the original approval task data is output. The task classification model is trained on the Naive Bayes classification model using historical approval task data and corresponding historical task types.

5. The method according to claim 1, characterized in that, The step of extracting elements and verifying element completeness from the de-identified approval task data, and generating element completeness verification results, includes: Extract element information corresponding to the approval element library from the de-identified approval task data. The approval element library stores the corresponding essential elements and optional elements according to the task type label. The approval element library supports manual and automatic updates. The manual update includes manually adding, modifying and deleting elements. The automatic update is to identify new elements in the historical approval task data through an element recognition model. The element recognition model is trained using historical approval task data and corresponding high-frequency elements. The element information is compared with the required elements of the corresponding task type tag in the approval element library to generate the element completeness verification result.

6. The method according to claim 5, characterized in that, The step of extracting element information corresponding to the approval element database from the de-identified approval task data includes: Using keyword extraction algorithms and semantic analysis models, textual elements are extracted from the task titles and core expressions in the de-identified approval task data. The semantic analysis model is obtained by training a bidirectional encoder representation model using historical approval task data and corresponding textual elements, or by using an industry-specific pre-trained model corresponding to the task type of the original approval task data to extract textual elements from the task titles and core expressions in the structured approval task data. Different task types correspond to different industry-specific pre-trained models, which are obtained by training a dedicated recognition model using historical approval task data and corresponding textual elements of the corresponding task type. The text elements are stored in a string concatenation format to generate the element information.

7. The method according to claim 5, characterized in that, The step of comparing the element information with the required elements of the corresponding task type tag in the approval element database to generate the element completeness verification result includes: The element information is compared one by one with the essential elements of the corresponding task type label. If there are any necessary elements that have not been extracted, they are marked as missing elements. Perform an information integrity check on the extracted necessary elements. If any elements are missing information, mark them as incomplete. Record the names and problem descriptions of missing or incomplete elements, and generate the element completeness verification results.

8. The method according to claim 1, characterized in that, The step of generating reference approval opinions based on the completeness verification results of the elements includes: The element completeness verification result is matched with multiple basic templates to determine the target basic template that matches the element completeness verification result. The basic templates include normal approval templates, element missing templates, and element incomplete templates. Based on the completeness verification results of the aforementioned elements, the target basic template is populated with content to generate initial approval reference opinions; Based on the task type and approval user profile of the original approval task data, the initial approval reference opinion is adjusted to generate the reference approval opinion. The approval user profile is constructed based on the approval user's historical approval preferences, decision-making habits, and focus.

9. The method according to claim 1, characterized in that, The process of generating centralized display information based on the original approval task data, the conflict detection results, the element completeness verification results, and the reference approval opinions includes: Generate summary display information for approval tasks. The display fields of the summary display information include task name, approval type, applicant user, submission time, element completeness status, and processing priority. The sorting methods of the summary display information include sorting by submission time, sorting by approval type, sorting by processing priority, and sorting by keyword search. The status identifier of the summary display information uses different color labels to distinguish the status of different elements. Generate complete display information and verification results for a single approval task. The complete display information and verification results are displayed in sections according to basic task information, approval element list, verification result details, and approval reference opinions. The approval element list is categorized into required elements and optional elements, and the element value and extraction source of each element are marked. The verification result details include the element name and problem description for missing or incomplete elements, and support jumping to the corresponding target system to view the original approval task data.

10. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the approval task processing method according to any one of claims 1-9.

Images