Intelligent office automation process management method and system

Abstract

The application relates to the technical field of intelligent office automation process management, and discloses an intelligent office automation process management method and system, wherein the method receives a sub-process request through an API interface and extracts context information associated with an original business request, the context information including a computing resource priority identifier, a target completion timestamp and resource emergency degree coding. This effectively solves the problem of context information loss in the prior art when a sub-process is processed, so that each sub-process is no longer independently run, but can inherit the global importance of the original business request.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent office automation process management, and more specifically, to an intelligent office automation process management method and system. Background Technology

[0002] In modern enterprise operating environments, intelligent office automation process management systems play a crucial role in optimizing cross-departmental business processes and improving operational efficiency. However, existing intelligent office automation process management systems have significant limitations when handling non-standardized or abnormal business scenarios. When a high-priority initial business request cannot be processed through the conventional path due to its non-standardized characteristics (e.g., procuring new materials), the system typically decomposes it or triggers a series of independent, standardized sub-processes (e.g., new supplier onboarding, special financial approvals, etc.) to address the issue.

[0003] The problem with existing systems is that they fail to effectively inherit and pass on the global business context of the initial request when executing these subsequent derived sub-processes, especially key contextual information such as its "priority" and "business objectives." Each sub-process runs according to its own preset, isolated rules and processing time limits, causing the system to break down a task that is urgent and important as a whole into multiple independent tasks that appear routine or even low-priority locally. This loss of contextual information prevents the system from recognizing the inherent logical connections and dependencies between these independent sub-processes, and further prevents it from dynamically adjusting priorities, coordinating resources, or handling exceptions based on the global importance of the original task when a sub-process node (e.g., a routine financial approval) encounters obstacles. This results in critical business being blocked by non-critical standardized procedures, causing the automated system to degenerate from an efficiency engine into a rigid barrier to the process when handling complex and abnormal scenarios, severely impacting the timeliness and efficiency of business operations.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] This invention provides an intelligent office automation process management method and system, which aims to solve the technical problem that existing intelligent office automation process management systems cannot effectively inherit and pass on the global business background of the initial request when handling non-standard or abnormal business scenarios, resulting in obstruction of sub-process priority adjustment, resource coordination or abnormal escalation processing, thereby affecting business timeliness and efficiency.

[0006] The technical solution of this application is as follows:

[0007] In a first aspect, this application discloses an intelligent office automation process management method, executed by a process engine system, comprising:

[0008] The API interface receives a subprocess request and uses a log parser to extract context information associated with the original business request that triggered the subprocess request. The context information includes the computing resource priority identifier, target completion timestamp, and resource urgency code of the original business request.

[0009] Based on the context information, combined with the current CPU utilization, memory usage and I / O wait queue length collected by the system performance monitor, the scheduling priority score of the subprocess in the queue to be processed is calculated by a weighted sorting algorithm, and the processing order is determined according to the scheduling priority score.

[0010] Based on the context information, combined with the real-time online status of the preset approval nodes, task queue length, and historical average processing time dynamically obtained from the personnel task scheduling engine, the load capacity of the target approval node is calculated through dynamic routing rules, and the task processing path of the sub-process is adjusted according to the load capacity, including adjustments such as triggering proxy nodes to take over or escalating permissions to the superior node.

[0011] Furthermore, in the method of this application, the step of calculating the load capacity of the target approval node through dynamic routing rules and adjusting the task processing path of the sub-process according to the load capacity further includes:

[0012] Obtain the context information corresponding to the approval task; the context information also includes the deviation type identifier.

[0013] Based on the deviation type identifier as an index, the corresponding risk case data hash value and rule update record timestamp are crawled from multiple heterogeneous data sources through a crawler module or database connection pool;

[0014] The hash value of risk case data and the timestamp of rule update record are input into the risk assessment model built based on decision tree or Bayesian network to calculate the risk score vector corresponding to the deviation type identifier. The risk score vector includes the risk probability value and the impact range quantification value.

[0015] Based on the risk scoring vector, a risk data package containing risk level identifiers is automatically generated;

[0016] Risk data packets are forwarded to the approval terminal via the internal data bus for automatic rendering of risk level identifiers on the terminal device.

[0017] Based on the above, the method of this application further includes the steps of calculating the load capacity of the target approval node through dynamic routing rules and adjusting the task processing path of the sub-process according to the load capacity:

[0018] Monitor whether multiple sub-processes with high computing resource priority trigger proxy node takeover or privilege escalation to the parent node simultaneously within a preset time window, and determine whether they point to the same approver identifier.

[0019] When it is determined to be the same approver identifier, the number of pending tasks corresponding to the approver identifier is obtained through the task queue monitoring interface, and the estimated total processing time of the approver is calculated as the node load value by combining the historical average processing time of each task stored in the database.

[0020] Query the backup table of the access control system for the list of candidate nodes for approvers. The list of candidate nodes includes the identifiers of subordinates or peers with the same or downgraded permissions.

[0021] Determine whether the node load value exceeds the preset processing time threshold or whether the number of pending tasks exceeds the quantity threshold;

[0022] When the time limit is exceeded, a routing operation is performed based on the computing resource priority identifier and target completion timestamp of the original business request, combined with the current online status and permission depth in the candidate node list: the message body of part of the sub-process request is cloned and the routing target field is modified to the selected candidate node identifier.

[0023] By using the temporary privilege escalation interface, time-limited approval privileges are granted to the candidate nodes corresponding to the selected candidate node identifiers, and the effective period of the privileges is recorded.

[0024] Furthermore, in the method of this application, the step of performing traffic splitting operation by combining the current online status and permission depth of the candidate node list further includes:

[0025] Continuously monitor the behavior data of the candidate personnel corresponding to the selected candidate nodes when processing the sub-process requests through the event listener. The behavior data includes the click flow frequency on the approval interface, the document viewing time, and the HTTP response delay to system notifications.

[0026] The behavioral data is dynamically time-normalized and compared with the historical behavioral baseline of the candidate personnel handling the same type of sub-process in the storage medium in order to calculate the deviation of the current processing efficiency.

[0027] When the deviation exceeds the preset deviation threshold, the context information of the sub-process request to be diverted is vector-matched with the professional domain knowledge graph of the candidate personnel to evaluate their real-time processing ability score.

[0028] When the real-time processing capability score is lower than the preset capability lower limit, the collaborative scheduler is activated: a collaborative processing request is sent to the collaborative node corresponding to another candidate with relevant professional knowledge and the shortest current task queue length, along with context information and deviation data.

[0029] Alternatively, activate the secondary dispatcher: remove the subprocess requests to be dispatched from the to-do list of the current candidate node and re-insert them into the global queue to be assigned, and select other candidate nodes with the highest overall matching degree for secondary assignment.

[0030] In some preferred embodiments, the step of sending a collaborative processing request to the collaborative node corresponding to another candidate with relevant expertise and the shortest current task queue length further includes:

[0031] Based on the context information of the diverted sub-process requests, at least one core professional domain feature tag vector is extracted using a text embedding model.

[0032] The professional skill tag vectors of multiple candidates are obtained from the collaborative ability pool, and the cognitive load index is predicted by Kalman filtering.

[0033] Calculate the cosine similarity between the core professional domain feature tag vector and the professional skill tag library vector to obtain the professional matching score;

[0034] The professional matching score and the cognitive load index are weighted and summed to generate a comprehensive matching score;

[0035] The candidate with the highest overall matching score is selected as the collaborative processing personnel, and a collaborative processing request is sent to the terminal corresponding to the collaborative processing personnel through a message queue. The collaborative processing request contains a snapshot of the message body of the sub-process request and serialized data of the context information.

[0036] As a technical improvement, the method of this application further includes the following steps in calculating the cognitive load index obtained by Kalman filtering prediction:

[0037] The system log collector obtains in real time the operation frequency, document viewing time, system notification response time, speaking frequency in collaboration tools, and email reply speed of the terminal devices corresponding to the candidates, as a multi-dimensional observation vector;

[0038] The multidimensional observation vector is input into the real-time task engagement calculation model based on exponential weighted moving average to obtain the engagement score;

[0039] Obtain the schedule and leave records of the candidates through a calendar application programming interface, and use them as time constraints;

[0040] The engagement score is mapped to the free time in the schedule and vacation records, and combined with the preset cognitive load level mapping table, the current cognitive load level value of the candidate is output.

[0041] The cognitive load level is input as an observation into the Kalman filter. The Kalman filter outputs the cognitive load index prediction value for the current moment based on the load prediction value of the previous moment and the current observation value.

[0042] Based on this, the method of this application, after the step of outputting the current cognitive load level value of the candidate, also includes:

[0043] Obtain actual work performance data generated by candidate personnel during the processing of diverted sub-process requests. Actual work performance data includes absolute task completion time, business rollback rate after approval, and sentiment analysis score of task feedback.

[0044] The Pearson correlation coefficient was calculated between the cognitive load level values ​​and actual work performance data to identify the statistical deviation between the predicted load and the actual performance.

[0045] When there is statistical bias and the bias exceeds the preset tolerance range, the gradient descent method is used to adjust the decay factor parameter in the exponential weighted moving average calculation model or the threshold parameter in the cognitive load level mapping table, so as to improve the fit between the subsequent output cognitive load level value and the actual work performance data.

[0046] The adjusted parameters are stored in the model parameter library, and the parameter version number and timestamp are recorded.

[0047] As a further improvement, the method of this application, after adjusting the decay factor parameter in the exponentially weighted moving average calculation model or the threshold parameter in the cognitive load level mapping table using the gradient descent method, further includes:

[0048] Select a historical sub-process request dataset within a preset time range from the storage medium as the verification set;

[0049] The cognitive load assessment model with adjusted parameters was used to simulate load classification for candidates in the validation set, and the simulated load classification results were obtained.

[0050] The simulation load classification results are compared with the corresponding actual work performance data in the validation set to calculate the simulation accuracy and misjudgment rate.

[0051] Determine whether the accuracy rate is lower than the preset accuracy threshold or whether the false positive rate is higher than the preset false positive rate limit;

[0052] When the accuracy rate is lower than the accuracy rate threshold or the false judgment rate is higher than the false judgment rate limit, a parameter adjustment risk signal is generated and pushed to the operation and maintenance terminal through the system management interface.

[0053] In response to the pushed parameter adjustment risk signal, the report generator is invoked to create a parameter risk analysis data package, and the parameter risk analysis data package is sent to the operation and maintenance terminal. The parameter risk analysis data package contains model parameter identifiers and data source identifiers.

[0054] A version management interface is provided on the operation and maintenance terminal, which includes the command entry point for rolling back to historical stable model parameters.

[0055] In one embodiment, after forwarding the risk data packet to the approval terminal via the internal data bus, the method of this application further includes:

[0056] In response to the completion of the approval operation by the proxy node or the superior node, an approval result data packet is generated.

[0057] Based on the routing information carried in the approval result data packet, the approval result data packet is sent to at least one of the following target devices: digital signature device, electronic signature server, printer, work order system interface, and notification SMS gateway.

[0058] Receive the processing status code returned by the target device, and update the status record of the sub-process request according to the processing status code.

[0059] Secondly, this application also discloses an intelligent office automation process management system, comprising:

[0060] At least one processor;

[0061] Memory that is communicatively connected to at least one processor;

[0062] The memory stores instructions that can be executed by at least one processor. When the instructions are executed by at least one processor, they cause at least one processor to perform any of the above-mentioned intelligent office automation process management methods.

[0063] The system also includes at least one of the following output interfaces: printer interface, work order system application programming interface, SMS gateway interface, and digital signature device interface, for outputting approval results.

[0064] Beneficial effects

[0065] This application discloses an intelligent office automation workflow management method and system. It receives sub-process requests via an API interface and extracts context information associated with the original business request. This context information includes a computing resource priority identifier, a target completion timestamp, and a resource urgency code. This effectively solves the problem of context information loss in existing systems when processing sub-processes, ensuring that each sub-process no longer runs in isolation but inherits the global importance of the original business request. Based on this, this application calculates the scheduling priority score of the sub-processes using a weighted sorting algorithm, combining the context information with CPU utilization, memory usage, and I / O wait queue length collected by a system performance monitor, and determines the processing order. This mechanism ensures that high-priority business sub-processes receive computing resources first, preventing critical business processes from being blocked by non-critical standardized procedures. Attached Figure Description

[0066] Figure 1 This is a flowchart of an intelligent office automation process management method provided by an embodiment of the present invention;

[0067] Figure 2 This is a flowchart of a method for adjusting the task processing path of a sub-process according to the load capacity provided in an embodiment of the present invention;

[0068] Figure 3 This is a schematic diagram of the structure of an intelligent office automation process management system provided in an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Reference Figure 1 , Figure 1 This is a flowchart of an intelligent office automation process management method provided in an embodiment of the present invention, executed by a process engine system, including:

[0071] S11 receives a subprocess request through the API interface and uses the log parser to extract context information associated with the original business request that triggered the subprocess request from the subprocess request. The context information includes the computing resource priority identifier, target completion timestamp, and resource urgency code of the original business request.

[0072] S12. Based on the context information, combined with the current CPU utilization, memory usage and I / O wait queue length collected by the system performance monitor, calculate the scheduling priority score of the subprocess in the queue to be processed using a weighted sorting algorithm, and determine the processing order based on the scheduling priority score.

[0073] S13. Based on the context information, combined with the real-time online status of the preset approval node, the task queue length and the historical average processing time dynamically obtained from the personnel task scheduling engine, the load capacity of the target approval node is calculated through dynamic routing rules, and the task processing path of the sub-process is adjusted according to the load capacity, including adjusting the triggering of the proxy node to take over or the privilege escalation to the superior node.

[0074] This application aims to provide an intelligent office automation process management method that effectively solves the problems of inefficiency and slow response caused by the loss of context information when existing systems handle complex and non-standardized business processes. By introducing context information extraction, dynamic scheduling priority calculation, and a task processing path adjustment mechanism based on load capacity, this application can ensure that sub-processes always maintain consistency with the global priority and objectives of the original business request during execution, thereby significantly improving the system's adaptability and processing efficiency in complex scenarios.

[0075] To better understand the intelligent office automation process management method proposed in this application, the following will explain in detail some key terms and implementation environments involved.

[0076] "API interface" refers to Application Programming Interface, which defines the specifications for interaction between different software components. In this application, the API interface is used to receive sub-process requests from external systems or modules, serving as the portal for the system's external interaction.

[0077] A "log parser" is a software module whose function is to extract useful information from structured or unstructured log data. In this application, the log parser is used to identify and extract contextual information related to the original business request from sub-process requests, for example, through techniques such as pattern matching and keyword recognition.

[0078] "Contextual information" refers to background data related to a specific business request, which provides the necessary environment and basis for request processing. In this application, contextual information includes computing resource priority identifiers, target completion timestamps, and resource urgency codes, which are crucial for subsequent scheduling and routing decisions.

[0079] A "system performance monitor" is a tool or module used to monitor various system performance metrics in real time. It can collect data such as CPU utilization, memory usage, and I / O wait queue length, providing real-time system load information for scheduling priority calculation.

[0080] The "weighted sorting algorithm" is an algorithm that sorts tasks based on multiple indicators and their weights. In this application, it combines contextual information and system performance data to calculate the scheduling priority score of subprocesses to determine their processing order in the queue.

[0081] The "Personnel Task Scheduling Engine" is a system component responsible for managing and allocating personnel tasks. It can dynamically obtain the real-time online status of preset approval nodes, task queue length, and historical average processing time, providing personnel resource information for dynamic routing rules.

[0082] "Dynamic routing rules" are a set of rules that determine the task flow path based on real-time conditions and preset strategies. In this application, it calculates the load capacity of the target approval node based on context information and data provided by the personnel task scheduling engine, and adjusts the task processing path of the sub-process accordingly.

[0083] "Proxy node takeover" refers to the temporary takeover of approval tasks by a pre-set proxy node when an approval node is overloaded or unavailable due to personnel issues.

[0084] "Elevate permissions to a higher-level node" means that when an approval task requires higher-level approval or the current node cannot process it, the task will be automatically reported to a higher-level node with higher permissions for processing.

[0085] The intelligent office automation process management method described in this application operates in a distributed system environment, which typically includes multiple service modules, a database, a message queue, and user terminal devices. The process engine system, as the core component, is responsible for coordinating and managing the execution of the entire process. API interfaces are deployed at the system's entry layer to receive external requests. Log parsers, system performance monitors, and personnel task scheduling engines serve as internal services, providing data support and decision-making basis.

[0086] The core of the intelligent office automation process management method in this application lies in receiving sub-process requests through an API interface and using a log parser to extract context information associated with the original business request that triggered the sub-process request. The context information includes the computing resource priority identifier, target completion timestamp, and resource urgency code of the original business request.

[0087] Specifically, the API interface can be a RESTful API endpoint. When an external system or service needs to initiate a sub-process, it sends an HTTP POST request to this endpoint, with the request body containing detailed information about the sub-process. For example, a "new supplier admission" sub-process request in a procurement process might include the supplier name, the type of materials to be procured, etc. The log parser can be configured to listen for request logs received by the API interface, or it can act as a processing module of the API interface, parsing the request message body immediately upon receiving a request. During the parsing process, the log parser identifies and extracts contextual information related to the original business request from the request message according to preset rules or patterns. For example, it can obtain this information by looking up specific field names (such as "original_request_id", "priority_level", "due_timestamp", etc.) or by matching the message content using regular expressions. The calculated resource priority identifier can be an integer value (e.g., 1-5, with 1 being the highest priority), the target completion timestamp can be a date and time string in ISO 8601 format, and the resource urgency code can be an enumeration value (e.g., "low", "medium", "high", "urgent").

[0088] Furthermore, based on the context information, combined with the current system CPU utilization, memory usage, and I / O wait queue length collected by the system performance monitor, the scheduling priority score of the subprocess in the pending queue is calculated using a weighted sorting algorithm, and the processing order is determined based on the scheduling priority score.

[0089] The system performance monitor can be a standalone microservice that periodically collects metrics such as server CPU utilization, memory utilization, and disk I / O queue length, and publishes this data to a message queue or stores it in a time-series database. When a subprocess request is received and its context information is parsed, the process engine system queries the latest performance data provided by the system performance monitor. The weighted sorting algorithm can be a predefined function that takes the priority identifier from the context information, the target completion timestamp (converted to remaining time), the resource urgency encoding (mapped to a numerical value), and system performance data (such as CPU utilization, memory utilization, and I / O wait queue length) as input. Each input parameter is assigned a weight; for example, the resource priority identifier might have the highest weight, followed by the target completion timestamp, and then the system performance metrics. The algorithm sums these weighted values ​​to obtain a comprehensive scheduling priority score.

[0090] In addition, based on contextual information, combined with the real-time online status of preset approval nodes, task queue length, and historical average processing time dynamically obtained from the personnel task scheduling engine, the load capacity of the target approval node is calculated through dynamic routing rules, and the task processing path of the sub-process is adjusted according to the load capacity, including adjustments such as triggering proxy node takeover or privilege escalation to the superior node.

[0091] The personnel task scheduling engine can be a standalone module that maintains the real-time online status of all approvers (e.g., via heartbeat mechanisms or login status), the current task queue length of each approval node, and the historical average processing time for various tasks at that node. When a sub-process needs to enter the approval stage, the process engine system queries the personnel task scheduling engine for relevant information about the target approval node. Dynamic routing rules can be a set of decision trees or rule engines based on conditional logic. It comprehensively considers the contextual information of the sub-process (e.g., the urgency of the original business request, computing resource priority identifiers), as well as the real-time online status, task queue length, and historical average processing time of the target approval node. For example, if the target approval node's current task queue is too long, or the approver is offline, the dynamic routing rules will calculate that the node's load capacity has reached or exceeded a threshold. In this case, the system will adjust the task processing path of the sub-process according to a preset strategy. One adjustment method is to trigger a proxy node takeover, automatically assigning the sub-process's approval task to a preset proxy approver or proxy team. Another adjustment method is to escalate permissions to a higher-level node, automatically reporting the sub-process's approval task to a leader or management with higher permissions for approval.

[0092] The core innovation of this application lies in its immediate use of a log parser to extract context information associated with the original business request that triggered the sub-process request after receiving the sub-process request via an API interface. This context information includes the computing resource priority identifier, target completion timestamp, and resource urgency code of the original business request. This mechanism ensures that the global priority and target of the original business request can be effectively inherited and passed to each sub-process, fundamentally solving the problem of context information loss in existing technologies.

[0093] Furthermore, this application employs a weighted sorting algorithm, combined with contextual information and real-time system resource data collected by a system performance monitor, to dynamically calculate the scheduling priority score of sub-processes. This enables the system to intelligently adjust the processing order of sub-processes in the pending queue based on the actual urgency of the business and the current resource status of the system, thereby avoiding the rigid scheduling strategies of traditional systems and significantly improving resource utilization efficiency and task response speed.

[0094] More importantly, this application introduces dynamic routing rules. These rules calculate the load capacity of the target approval node based on contextual information, real-time personnel status provided by the personnel task scheduling engine, and historical processing time, and adjust the task processing path of the sub-process accordingly. This adjustment includes triggering proxy nodes to take over or escalating permissions to higher-level nodes, effectively solving the bottleneck problem that may occur at approval nodes in traditional systems, and ensuring that critical business operations can be handled promptly and flexibly when encountering obstacles.

[0095] In summary, the intelligent office automation process management method of this application, by introducing context information inheritance, dynamic scheduling priority calculation, and dynamic routing adjustment mechanism based on load capacity, realizes intelligent management of complex and non-standard business processes, significantly improves the system's adaptability, efficiency, and responsiveness, and provides strong technical support for enterprises when dealing with complex business scenarios.

[0096] refer to Figure 2 , Figure 2 This is a flowchart of a method for adjusting the task processing path of a sub-process according to load capacity, provided by an embodiment of the present invention. S13 includes:

[0097] S131, Obtain the context information corresponding to the approval task, wherein the context information further includes a deviation type identifier;

[0098] S132, based on the deviation type identifier as an index, crawl the corresponding risk case data hash value and rule update record timestamp from multiple heterogeneous data sources through a crawler module or database connection pool;

[0099] S133, input the hash value of the risk case data and the timestamp of the rule update record into the risk assessment model built based on decision tree or Bayesian network to calculate the risk score vector corresponding to the deviation type identifier. The risk score vector includes a risk probability value and an impact range quantification value.

[0100] S134, Based on the risk scoring vector, automatically generate a risk data packet containing a risk level identifier;

[0101] S135, the risk data packet is forwarded to the approval terminal via the internal data bus for automatic rendering of the risk level identifier on the terminal device.

[0102] Specifically, upon receiving a sub-process request, the system first retrieves the context information related to the approval task. This context information includes, in addition to the computational resource priority identifier, target completion timestamp, and resource urgency code of the original business request, a deviation type identifier. This deviation type identifier indicates the category of potential business deviations or anomalies in the current approval task, such as excessive approval, special transaction types, or attempts to exceed authorized limits. This identifier can be automatically appended by the business system when generating the sub-process request, or identified and generated by the log parser according to specific rules when extracting context information.

[0103] Based on the acquired deviation type identifier, the system uses it as a query index to retrieve relevant risk case data hash values ​​and rule update timestamps from multiple heterogeneous data sources using a crawler module or a database connection pool. These heterogeneous data sources can include, but are not limited to, historical risk databases, compliance policy document repositories, external risk intelligence platforms, and audit reporting systems. The crawler module is responsible for extracting information from unstructured or semi-structured data sources, while the database connection pool is used for efficient access to structured databases. Risk case data hash values ​​are used to quickly locate and verify historical risk events, and the rule update timestamps ensure that the risk assessment rules used are up-to-date, avoiding judgments based on outdated rules.

[0104] Subsequently, the hash values ​​of the captured risk case data and the timestamps of rule update records are used as inputs and fed into a risk assessment model pre-built based on decision trees or Bayesian networks. This model analyzes this data to calculate a risk score vector corresponding to the deviation type identifier. The risk score vector is a multi-dimensional numerical value, typically including a risk probability value and an impact scope quantification value. The risk probability value represents the likelihood of a specific deviation type event occurring, while the impact scope quantification value represents the degree of business, financial, or compliance impact that the event may cause if it occurs. Decision tree models can classify risks through a series of conditional judgments, while Bayesian networks can handle uncertainty and perform probabilistic reasoning.

[0105] After obtaining the risk score vector, the system automatically generates a risk data package containing a risk level identifier based on the vector. The risk level identifier can be a classification label such as "high risk," "medium risk," or "low risk," or it can be a specific risk score range. This risk data package encapsulates the risk assessment results and is prepared for subsequent risk alerts and processing.

[0106] Finally, the risk data packet is forwarded to the corresponding approval terminal via the internal data bus. The internal data bus ensures efficient and secure data transmission. On the approval terminal, the risk data packet is received and parsed, and the risk level indicator it contains is automatically rendered and displayed on the approval interface, for example, presented to the approver in a prominent color, icon, or text prompt.

[0107] This application's solution addresses the problem of traditional solutions that rely solely on load capacity for routing, neglecting potential business risks, by incorporating a risk assessment mechanism into dynamic routing rules. Through this technical solution, the application enables early identification and quantitative assessment of potential risks in intelligent office automation processes. By combining deviation type identifiers with historical risk data and rule update records, and utilizing advanced risk assessment models, the system can generate precise risk score vectors and risk level identifiers for each approval task. Consequently, approvers can intuitively obtain task risk information on their terminal devices, enabling them to make more informed and prudent approval decisions.

[0108] In the above method, the step of calculating the load capacity of the target approval node through dynamic routing rules and adjusting the task processing path of the sub-process according to the load capacity further includes:

[0109] Monitor whether multiple sub-processes with high computing resource priority trigger proxy node takeover or privilege escalation to the parent node simultaneously within a preset time window, and determine whether they point to the same approver identifier.

[0110] When the same approver identifier is identified, the number of pending tasks corresponding to the approver identifier is obtained through the task queue monitoring interface, and the estimated total processing time of the approver is calculated as the node load value by combining the historical average processing time of each task stored in the database.

[0111] The backup table of the access control system is used to query the list of candidate nodes for the approver. The list of candidate nodes contains the identifiers of subordinates or peers with the same or downgraded permissions.

[0112] Determine whether the node load value exceeds a preset processing time threshold or whether the number of pending tasks exceeds a quantity threshold;

[0113] When the time limit is exceeded, a diversion operation is performed based on the computing resource priority identifier and target completion timestamp of the original business request, combined with the current online status and permission depth in the candidate node list: the message body of part of the sub-process request is cloned and the routing target field is modified to the selected candidate node identifier.

[0114] The temporary privilege escalation interface grants time-limited approval privileges to the candidate node corresponding to the selected candidate node identifier, and records the effective period of the privileges.

[0115] Specifically, this application first monitors whether multiple sub-processes with high computing resource priority are simultaneously routed to the same approver for proxy takeover or privilege escalation within a specific time window. This monitoring aims to identify potential approval bottlenecks. High computing resource priority refers to sub-processes that have high system resource requirements or are of extremely high business importance, such as processes involving large-scale fund approvals or urgent project decisions. The preset time window can be configured according to actual business needs, such as 5 minutes or 10 minutes, to capture instantaneous or short-term load spikes.

[0116] Furthermore, once a potential load concentration is identified, the system obtains the number of tasks currently pending for the approver through the task queue monitoring interface, and calculates an estimated total processing time based on their historical average processing time. This estimated total processing time serves as the approver's node load value. This node load value quantifies the current workload of the approver. For example, if an approver has 10 pending tasks, and the historical average processing time for each task is 5 minutes, then their estimated total processing time is 50 minutes.

[0117] Based on this, the system will query the backup table of the approval personnel for a list of alternative nodes. This list of alternative nodes includes identifiers of subordinates or peers with the same or downgraded permissions, who are qualified to handle the relevant approval tasks. For example, when a department manager is overloaded, senior subordinates or other department managers at the same level can serve as alternative nodes.

[0118] The system then determines whether the calculated node load exceeds a preset processing time threshold or whether the number of pending tasks exceeds a quantity threshold. These thresholds are preset based on system performance, business requirements, and personnel capacity, and are used to define whether the approver is overloaded. For example, if the estimated total processing time exceeds 2 hours, or the number of pending tasks exceeds 20, the approver is considered overloaded.

[0119] When overload is detected, the system will perform a traffic splitting operation. Specifically, based on the computing resource priority identifier and target completion timestamp of the original business request, combined with the current online status and permission depth of the candidate node list, one or more suitable candidate nodes will be intelligently selected for traffic splitting. The traffic splitting operation involves cloning the message body of a portion of the sub-process request and modifying its routing target field to the identifier of the selected candidate node, thereby assigning some tasks to other personnel for processing. For example, for sub-processes with relatively low priority but still requiring timely processing, they can be split to subordinates with lower permission depths who are currently online and idle.

[0120] To ensure the smooth completion of approval processes for the decentralized sub-processes, the system grants time-limited approval permissions to selected candidate node identifiers via a temporary permission escalation interface, recording the effective period of the permissions. This means that even if candidate personnel do not normally possess permissions to handle such tasks, they can obtain temporary authorization during the decentralization period, thus ensuring business continuity. The temporary nature of the permissions prevents abuse and they are automatically revoked upon task completion.

[0121] This application's solution effectively addresses the overload problem of single-approval personnel that may result from traditional routing adjustments by introducing dynamic monitoring and intelligent traffic distribution mechanisms for the load on approval nodes. Through the aforementioned technical solution, this application can significantly improve the resilience and efficiency of intelligent office automation processes.

[0122] In this regard, this application further proposes that the steps for performing traffic splitting operations based on the current online status and permission depth of the candidate node list further include:

[0123] The system continuously monitors the behavior data of the candidate personnel corresponding to the selected candidate nodes when processing the sub-process requests that have been diverted, through the event tracking device. The behavior data includes the click flow frequency on the approval interface, the document viewing time, and the HTTP response delay to system notifications.

[0124] The behavioral data is dynamically time-normalized and compared with the historical behavioral baseline of the candidate personnel handling the same type of sub-process in the storage medium to calculate the deviation of the current processing efficiency.

[0125] When the deviation exceeds a preset deviation threshold, the context information of the diverted sub-process request is vector-matched with the professional domain knowledge graph of the candidate personnel to evaluate their real-time processing ability score.

[0126] When the real-time processing capability score is lower than the preset capability lower limit, the collaborative scheduler is activated: a collaborative processing request is sent to the collaborative node corresponding to another candidate with relevant professional knowledge and the shortest current task queue length, along with the context information and deviation data.

[0127] Alternatively, activate the secondary distribution mechanism: remove the distributed subprocess request from the to-do list of the current candidate node and reinsert it into the global allocation queue, then select the other candidate node with the highest overall matching degree for secondary allocation.

[0128] Specifically, a data tracking listener can be understood as a software module embedded in the approval terminal or related applications. Its purpose is to collect, in real-time and non-intrusive, operational data of candidates when processing requests from sub-processes. This behavioral data, such as clickstream frequency, refers to the intensity of operations performed by candidates on the approval interface, reflecting their activity level and focus; document review time indicates the depth of their understanding of relevant materials and the time required; and HTTP response latency to system notifications reflects their response speed and priority judgment to system events. These data collectively constitute the basis for evaluating the current work status and engagement of candidates.

[0129] Dynamic time warping is an algorithm used to measure the similarity between two time series. Its purpose is to align and compare the currently collected behavioral data sequence with the historical behavioral baseline of candidate personnel handling similar sub-processes in the storage medium, thereby calculating the deviation of the current processing efficiency. The historical behavioral baseline can be a reference model established through statistical analysis of the average or optimal behavioral patterns of candidate personnel in handling similar tasks in the past. When the deviation exceeds a preset deviation threshold, it indicates that the candidate personnel's current performance may be lower than their normal level or expected efficiency.

[0130] In practical applications, a professional domain knowledge graph can specifically be structured data stored in a database or knowledge base, containing the candidate's areas of expertise, skill tags, and related knowledge systems. Vector matching refers to converting the contextual information of the sub-process request (such as task type, involved departments, key terms, etc.) into a vector representation using natural language processing technology, and then calculating the similarity between this vector representation and the candidate's professional domain knowledge graph to assess the candidate's professional matching degree. This yields a real-time processing ability score, which comprehensively reflects the candidate's professional competence in the current task.

[0131] When the real-time processing capability score falls below the preset lower limit, the system determines that the current candidate personnel may not be able to complete the task efficiently or accurately. At this point, the collaborative scheduler or secondary dispatcher can be activated. Activating the collaborative scheduler means the system automatically identifies another candidate personnel with relevant expertise and the shortest current task queue length, and sends a collaborative processing request to their corresponding collaborative node. This request includes the context information of the original sub-process request and the deviation data of the current candidate personnel, enabling the collaborating personnel to quickly understand the situation and provide assistance. A collaborative node can be understood as the approval terminal or workstation of the collaborating personnel.

[0132] As an alternative, a secondary dispatcher can be activated. The secondary dispatcher removes the dispatched sub-process requests from the current candidate node's to-do list and re-inserts them into the global allocation queue. Subsequently, the system uses a more comprehensive matching algorithm (e.g., considering factors such as professional matching degree, cognitive load, and task queue length) to select the other candidate node with the highest overall matching degree from all available candidate nodes for secondary allocation, ensuring that tasks are processed with the most appropriate resources. The global allocation queue can be understood as a central buffer pool for all pending sub-process requests.

[0133] This application's solution effectively addresses the potential problem of poor processing efficiency after task allocation by introducing a mechanism for continuous monitoring and dynamic evaluation of the processing behavior of candidate personnel. Through this technical solution, this application can significantly improve the intelligence level and processing efficiency of task allocation in intelligent office automation process management. Compared to basic solutions that rely solely on static information for allocation, this application avoids inefficiency or task backlog caused by blind allocation by dynamically evaluating the processing efficiency and professional capabilities of candidate personnel through real-time monitoring of their behavioral data.

[0134] The aforementioned step of sending a collaborative processing request to the collaborative node corresponding to another candidate with relevant expertise and the shortest current task queue length further includes:

[0135] Based on the context information of the diverted sub-process requests, at least one core professional domain feature tag vector is extracted using a text embedding model.

[0136] The professional skill tag vectors of multiple candidates are obtained from the collaborative ability pool, and the cognitive load index is predicted by Kalman filtering.

[0137] Calculate the cosine similarity between the core professional domain feature tag vector and the professional skill tag library vector to obtain the professional matching score;

[0138] The professional matching score and the cognitive load index are weighted and summed to generate a comprehensive matching score;

[0139] The candidate with the highest overall matching score is selected as the collaborative processing personnel, and the collaborative processing request is sent to the terminal corresponding to the collaborative processing personnel through a message queue. The collaborative processing request contains the message body snapshot and serialized data of the diverted sub-process request.

[0140] Specifically, the context information may include the business type of the sub-process request, the product modules involved, key fields, etc., and this information is input into a text embedding model. The text embedding model, for example, can be a Transformer-based model such as BERT, RoBERTa, or Word2Vec. Its function is to transform unstructured text information into a high-dimensional numerical vector, i.e., a core professional domain feature label vector. This vector can quantitatively represent the core requirements of the sub-process request in the professional domain.

[0141] The collaborative capability pool is a database or service that stores the professional skills and real-time status of all candidates. The professional skill tag vector is a pre-constructed numerical vector for each candidate, describing their areas of expertise and skill sets. The cognitive load index is a dynamic indicator predicted using methods such as Kalman filtering, used to reflect the candidate's current workload and cognitive stress, avoiding the assignment of tasks to personnel under high workload.

[0142] In practical applications, the cosine similarity calculation between the core professional domain feature tag vector and the professional skill tag library vector aims to measure the degree of matching between the professional requirements of the sub-process request and the professional skills of the candidate. A higher cosine similarity value indicates a better match. The weighted sum of the professional matching score and the cognitive load index can be weighted differently according to actual business needs. For example, in urgent tasks, the weight of cognitive load can be appropriately reduced, prioritizing professional matching; while in non-urgent tasks, the two can be balanced to ensure the rationality of task allocation. The resulting comprehensive matching score can fully reflect the fit between the candidate's professional capabilities and their current state.

[0143] Ultimately, the system selects the candidate with the highest overall matching score as the collaborator. The collaborating request is sent to the collaborator's terminal via a message queue, ensuring reliable delivery. The request includes a snapshot of the sub-process request message body and serialized data of context information, enabling the collaborator to quickly understand the task details and proceed with processing.

[0144] The solution proposed in this application addresses the problem of insufficient accuracy in personnel selection in traditional collaborative scheduling by introducing a text embedding model, a collaborative capability pool, a professional skill tag library vector, and a cognitive load index predicted by Kalman filtering.

[0145] The calculation steps for the cognitive load index obtained through Kalman filtering prediction further include:

[0146] The system log collector obtains in real time the operation frequency, document viewing time, system notification response time, speaking frequency in collaboration tools, and email reply speed of the terminal devices corresponding to the candidates, as a multi-dimensional observation vector.

[0147] The multidimensional observation vector is input into a real-time task engagement calculation model based on exponential weighted moving average to obtain an engagement score.

[0148] The schedule and leave records of the candidates are obtained through a calendar application programming interface and used as time constraints.

[0149] The engagement score is mapped to the free time periods in the schedule information and vacation records, and combined with the preset cognitive load level mapping table, the current cognitive load level value of the candidate is output.

[0150] The cognitive load level value is input into the Kalman filter as the observation value. The Kalman filter outputs the cognitive load index prediction value at the current moment based on the load prediction value at the previous moment and the current observation value.

[0151] Specifically, the system log collector is configured to continuously monitor various operational behaviors of candidates on their terminal devices, such as keyboard input frequency, mouse click count, active duration of specific applications, document opening and viewing time, HTTP response latency to system pop-ups or notifications, message sending frequency in enterprise collaboration tools (such as instant messaging and project management platforms), and email sending, receiving, and reply speed. These discrete or continuous behavioral data are integrated into a multi-dimensional observation vector to reflect the candidate's current activity level and task engagement. Operation frequency can be understood as the number of times a user interacts with the device per unit of time; document viewing time refers to the time a user spends viewing a document; system notification response time refers to the time interval between the system issuing a notification and the user's response; and speaking frequency and email reply speed reflect the user's investment in communication and collaboration.

[0152] The multidimensional observation vectors are then input into a real-time task engagement calculation model based on an exponentially weighted moving average. This model dynamically calculates the candidate's current real-time task engagement score by assigning different weights to historical observation data, with a greater emphasis on the impact of recent data. A higher engagement score generally indicates that the candidate is currently under high workload or focused attention. Simultaneously, the candidate's schedule information, including meetings, appointments, deadlines, and vacation records, can be obtained through a calendar application programming interface (API). This information is used as a time constraint to identify the candidate's actual free time. The engagement score is mapped to these free time periods and, combined with a pre-defined cognitive load level mapping table, transforms the abstract engagement score and time constraints into specific cognitive load level values. For example, when the engagement score is high and there are few free time periods, the cognitive load level value may be high.

[0153] Finally, the cognitive load level value is input as an observation into the Kalman filter. The Kalman filter is a highly efficient recursive filter that, based on the system's previous load prediction (i.e., state estimate) and the current observation (i.e., cognitive load level value), fuses prediction and observation information to output a more accurate and robust cognitive load index prediction for the current moment. Its purpose is to effectively filter out measurement noise by dynamically fusing multi-source data and historical predictions, providing a smooth and accurate cognitive load assessment and avoiding misjudgments caused by fluctuations from a single data source.

[0154] The proposed solution constructs a comprehensive task engagement assessment system by collecting multi-dimensional behavioral data of candidates on their terminal devices in real time, combined with their schedules and vacation records. This technical solution enables dynamic, real-time, and high-precision prediction of candidates' cognitive load.

[0155] This application further proposes that, after the step of outputting the current cognitive load level value of the candidate, it also includes:

[0156] Obtain the actual work performance data generated by the candidate personnel during the processing of the diverted sub-process requests. The actual work performance data includes the absolute time of task completion, the business rollback rate after approval, and the sentiment analysis score of task feedback.

[0157] The Pearson correlation coefficient was calculated between the cognitive load level values ​​and the actual work performance data to identify the statistical deviation between the predicted load and the actual performance.

[0158] When there is a statistical bias and the bias exceeds the preset tolerance range, the gradient descent method is used to adjust the decay factor parameter in the exponentially weighted moving average calculation model or the threshold parameter in the cognitive load level mapping table, so as to improve the fit between the subsequent output cognitive load level value and the actual work performance data.

[0159] The adjusted parameters are stored in the model parameter library, and the parameter version number and timestamp are recorded.

[0160] Specifically, the actual work performance data refers to data generated by candidates during the actual processing of sub-process requests, which objectively reflects their work status and efficiency. Among these, the absolute task completion time refers to the actual time spent from task assignment to completion, used to measure processing efficiency; the post-approval business rollback rate refers to the proportion of tasks approved by candidates that are subsequently withdrawn or require reprocessing, used to assess approval quality and accuracy; and the sentiment analysis score for task feedback refers to analyzing the sentiment tendencies of task-related communication records, comments, or feedback through natural language processing, indirectly reflecting the candidate's psychological state and satisfaction when handling tasks. These data comprehensively reflect the candidate's true workload and performance from multiple dimensions.

[0161] The purpose of calculating the Pearson correlation coefficient between the cognitive load level and the actual work performance data is to quantify the degree of linear correlation between the predicted and actual values. The Pearson correlation coefficient effectively reveals whether there is a statistical association between the predicted cognitive load and actual work performance, as well as the strength and direction of this association. By calculating this correlation coefficient, it is possible to identify areas where the prediction model may be biased. For example, when the predicted load is high but the actual completion time is short, it may indicate that the model underestimates the processing capacity of the candidates, and vice versa.

[0162] In practical applications, when a statistical deviation is detected between the predicted workload and the actual performance, and this deviation exceeds a preset tolerance range, the system will activate an adaptive adjustment mechanism. The preset tolerance range can be set according to actual business needs and the system's requirements for prediction accuracy. For example, it can be set when the Pearson correlation coefficient is below a certain threshold, or when the mean absolute error between the predicted and actual values ​​exceeds a certain percentage. When the adjustment conditions are met, the model parameters will be optimized using gradient descent. Gradient descent is a commonly used optimization algorithm that iteratively adjusts parameters to minimize the objective function (e.g., prediction error). Here, the objective function can be the difference between the predicted cognitive load level and the actual work performance data.

[0163] Specifically, the decay factor parameter in the exponentially weighted moving average calculation model can be adjusted. This parameter affects the weight of historical data in calculating the current engagement score, thus impacting the real-time assessment of cognitive load. Alternatively, the threshold parameter in the cognitive load level mapping table can be adjusted. This mapping table defines the conversion rules from engagement score to cognitive load level. These adjustments aim to ensure that the subsequently output cognitive load level values ​​more accurately reflect the actual work performance of the candidates, thereby improving the model's fit.

[0164] Furthermore, to ensure the effectiveness and traceability of model parameter adjustments, the adjusted parameters will be stored in the model parameter library, along with their version number and timestamp. This helps in tracing historical parameter changes and performing version management and rollback operations during subsequent model iterations or problem troubleshooting.

[0165] This application's solution calibrates and verifies previously predicted cognitive load levels by incorporating actual work performance data generated by alternative personnel during the processing of diverted sub-process requests. Through this technical solution, the application effectively addresses the problem of discrepancies between predicted and actual performance in traditional cognitive load prediction models. By introducing actual work performance data and performing statistical bias analysis, the system can promptly identify and quantify the shortcomings of the prediction model.

[0166] In response, this application further proposes that after adjusting the decay factor parameter in the exponentially weighted moving average calculation model or the threshold parameter in the cognitive load level mapping table using the gradient descent method, it also includes:

[0167] Select a historical sub-process request dataset within a preset time range from the storage medium as the verification set;

[0168] The cognitive load assessment model with adjusted parameters was used to simulate the load classification of candidates in the validation set, and the simulated load classification results were obtained.

[0169] The simulated load classification results are compared with the corresponding actual work performance data in the validation set to calculate the simulation accuracy and misjudgment rate.

[0170] Determine whether the accuracy rate is lower than a preset accuracy threshold or whether the false positive rate is higher than a preset false positive rate upper limit;

[0171] When the accuracy rate is lower than the accuracy rate threshold or the false judgment rate is higher than the false judgment rate upper limit, a parameter adjustment risk signal is generated and the parameter adjustment risk signal is pushed to the operation and maintenance terminal through the system management interface.

[0172] In response to the pushed parameter adjustment risk signal, the report generator is invoked to create a parameter risk analysis data package, and the parameter risk analysis data package is sent to the operation and maintenance terminal. The parameter risk analysis data package contains model parameter identifier and data source identifier.

[0173] The operation and maintenance terminal provides a version management interface, which includes an instruction entry point for rolling back to historical stable model parameters.

[0174] Specifically, after the model parameters are adjusted, to ensure the effectiveness and security of the adjustments, a historical sub-process request dataset within a preset time range needs to be selected from the storage medium. This dataset is used as a validation set, the purpose of which is to simulate the model's performance in a real-world operating environment, thereby objectively evaluating the adjusted parameters. This validation set should be representative, reflecting the actual working scenarios of different types of sub-process requests and candidate personnel.

[0175] Furthermore, the cognitive load assessment model with adjusted parameters is used to simulate load grading for candidates in the validation set. This process simulates the model's prediction of cognitive load for candidates in practical applications and outputs the corresponding simulated load grading results.

[0176] Subsequently, the simulated load classification results are compared with the corresponding actual work performance data in the validation set. This comparison allows for the calculation of the accuracy and misclassification rate of the adjusted model in the simulated scenario. Accuracy measures the proportion of times the model correctly predicts the cognitive load level, while the misclassification rate reflects the proportion of times the model makes incorrect predictions, such as misclassifying a high load as a low load or vice versa.

[0177] Based on this, the system will determine whether the calculated accuracy rate is lower than a preset accuracy threshold, or whether the false positive rate is higher than a preset false positive rate upper limit. These thresholds and upper limits are preset based on business needs, system stability, and acceptable risk levels.

[0178] When the judgment result shows that the accuracy rate is lower than the preset accuracy threshold or the false positive rate is higher than the preset false positive rate upper limit, it indicates that the parameter adjustment may not have achieved the expected effect, or may even have introduced new risks. At this time, the system will automatically generate a parameter adjustment risk signal and push the risk signal to the operation and maintenance terminal through the system management interface, so that the operation and maintenance personnel can perceive and intervene in a timely manner.

[0179] In response to the received parameter adjustment risk signal, the system automatically invokes the report generator to create a parameter risk analysis data package. This data package contains model parameter identifiers and data source identifiers, aiming to provide operations and maintenance personnel with key information to trace the root cause of problems, such as which model, which parameter, and which data source caused the performance issue. Subsequently, this parameter risk analysis data package is sent to the operations and maintenance terminal for detailed analysis and decision-making by operations and maintenance personnel.

[0180] To provide risk response and rapid recovery capabilities, a version management interface is provided on the operations and maintenance terminal. This interface includes command entries for rolling back to historical stable model parameters. This means that if operations and maintenance personnel confirm through analysis that the adjusted parameters have serious problems or introduce unacceptable risks, they can quickly restore the model parameters to a previously verified stable version through this interface, thereby minimizing potential risks and system downtime.

[0181] This application's solution effectively addresses the model instability and potential risks that may arise from adjusting parameters solely using the gradient descent method by introducing a verification and risk management mechanism after parameter adjustment. Through this technical solution, the application effectively avoids performance degradation or prediction errors in the cognitive load assessment model caused by improper parameter adjustment, significantly improving the robustness and safety of the model parameter adjustment process. This solution not only provides a means to quantitatively evaluate the effect of parameter adjustment, but more importantly, it establishes a comprehensive risk warning and rapid recovery mechanism. When model performance becomes abnormal, the system can promptly issue risk signals and provide detailed analytical data, greatly shortening troubleshooting time. Furthermore, the rollback function provided by the version management interface ensures that the system can quickly recover to a stable operating state when facing unforeseen parameter adjustment risks, thereby guaranteeing the continuity and reliability of intelligent office automation processes and reducing potential business risks and operational costs.

[0182] This application further proposes that after forwarding the risk data packet to the approval terminal via the internal data bus, it also includes:

[0183] In response to the completion of the approval operation by the proxy node or the superior node, an approval result data packet is generated.

[0184] Based on the routing information carried in the approval result data packet, the approval result data packet is sent to at least one of the following target devices: digital signature device, electronic signature server, printer, work order system interface, and notification SMS gateway;

[0185] Receive the processing status code returned by the target device, and update the status record of the sub-process request according to the processing status code.

[0186] Specifically, once the proxy node or parent node completes the approval process for a sub-process request, the system automatically generates an approval result data packet. This approval result data packet is a digital representation of the approval decision, containing the approval conclusion (e.g., approved, rejected, pending modification), the approver's identifier, the approval timestamp, and routing information for subsequent processing. The routing information can be understood as indicating which target devices or systems the approval result should be sent to; for example, depending on the business type or the nature of the approval result, it determines whether digital signatures, printing paper documents, updating the work order system status, or sending notification SMS messages are required.

[0187] The target equipment may include a digital signature device, an electronic signature server, a printer, a work order system interface, and a notification SMS gateway. The digital signature device or electronic signature server is used to electronically sign or seal the approval results to ensure their legal validity or authenticity. The printer is used to output paper approval documents. The work order system interface is used to synchronize the approval results to the relevant work order management system to drive the execution of subsequent business processes. The notification SMS gateway is used to send approval result notifications to relevant personnel.

[0188] In practical applications, after the approval result data packet is sent to the corresponding target device, the system receives a processing status code returned by the target device. This processing status code is feedback information returned by the target device after processing the received approval result data packet, such as successful processing, processing failure, or processing in progress. The system updates the status record of the sub-process request in real time based on these processing status codes to ensure the accuracy and traceability of the process status. For example, if the digital signature is successful, the sub-process status may be updated to "signed"; if the work order system update is successful, it will be updated to "work order updated".

[0189] This application's solution effectively solves the problems of lag and inconsistency in the distribution and status updates of approval results in traditional solutions by introducing an automated result processing mechanism after the approval process is completed. Through the above technical solution, this application can significantly improve the overall efficiency and reliability of intelligent office automation processes.

[0190] refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of an intelligent office automation process management system provided in an embodiment of the present invention, including:

[0191] At least one processor;

[0192] A memory that is communicatively connected to the at least one processor;

[0193] The memory stores instructions that can be executed by the at least one processor. When the instructions are executed by the at least one processor, the at least one processor performs the above-described intelligent office automation process management method.

[0194] The system also includes at least one of the following output interfaces: a printer interface, a work order system application programming interface, an SMS gateway interface, and a digital signature device interface, for outputting approval results.

[0195] This system aims to provide a hardware and software integrated solution to effectively address the inefficiencies and sluggish responses caused by the loss of contextual information in existing intelligent office automation process management systems when handling complex and non-standardized business processes. By configuring the processor, memory, and executing specific instructions, the system can achieve intelligent scheduling, dynamic routing, and result output of sub-processes, ensuring that the sub-processes always maintain consistency with the global priority and objectives of the original business request during execution, thereby significantly improving the system's adaptability and processing efficiency in complex scenarios.

[0196] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent office automation process management, characterized in that, Executed by the process engine system, including: A subprocess request is received through the API interface, and the log parser is used to extract context information associated with the original business request that triggered the subprocess request from the subprocess request. The context information includes the computing resource priority identifier, target completion timestamp, and resource urgency code of the original business request. Based on the context information, combined with the current system CPU utilization, memory usage, and I / O wait queue length collected by the system performance monitor, the scheduling priority score of the subprocess in the waiting queue is calculated using a weighted sorting algorithm, and the processing order is determined based on the scheduling priority score. Based on the context information, combined with the real-time online status, task queue length and historical average processing time of the preset approval node dynamically obtained from the personnel task scheduling engine, the load capacity of the target approval node is calculated through dynamic routing rules, and the task processing path of the sub-process is adjusted according to the load capacity. The adjustment includes triggering the proxy node to take over or escalate the permission to the parent node, monitoring whether multiple sub-processes with high computing resource priority trigger the proxy node to take over or escalate the permission to the parent node at the same time within a preset time window, and determining whether they point to the same approval personnel identifier. When the same approver identifier is identified, the number of pending tasks corresponding to the approver identifier is obtained through the task queue monitoring interface, and the estimated total processing time of the approver is calculated as the node load value by combining the historical average processing time of each task stored in the database. The backup table of the access control system is used to query the list of candidate nodes for the approver. The list of candidate nodes contains the identifiers of subordinates or peers with the same or downgraded permissions. Determine whether the node load value exceeds a preset processing time threshold, or whether the number of pending tasks exceeds a quantity threshold; When the time limit is exceeded, a diversion operation is performed based on the computing resource priority identifier and target completion timestamp of the original business request, combined with the current online status and permission depth in the candidate node list: the message body of part of the sub-process request is cloned and the routing target field is modified to the selected candidate node identifier. The temporary privilege escalation interface grants time-limited approval privileges to the candidate node corresponding to the selected candidate node identifier, and records the effective period of the privileges.

2. The method according to claim 1, characterized in that, The step of calculating the load capacity of the target approval node through dynamic routing rules and adjusting the task processing path of the sub-process according to the load capacity further includes: Obtain the context information corresponding to the approval task, wherein the context information also includes an deviation type identifier; Based on the deviation type identifier as an index, the corresponding risk case data hash value and rule update record timestamp are crawled from multiple heterogeneous data sources through a crawler module or database connection pool; The hash value of the risk case data and the timestamp of the rule update record are input into the risk assessment model built based on decision tree or Bayesian network to calculate the risk score vector corresponding to the deviation type identifier. The risk score vector includes a risk probability value and an impact range quantification value. Based on the risk scoring vector, a risk data package containing risk level identifiers is automatically generated; The risk data packet is forwarded to the approval terminal via the internal data bus for automatic rendering of the risk level identifier on the terminal device.

3. The method according to claim 1, characterized in that, The step of performing traffic splitting based on the current online status and permission depth of the candidate node list further includes: The system continuously monitors the behavior data of the candidate personnel corresponding to the selected candidate nodes when processing the sub-process requests that have been diverted, through the event tracking device. The behavior data includes the click flow frequency on the approval interface, the document viewing time, and the HTTP response delay to system notifications. The behavioral data is dynamically time-normalized and compared with the historical behavioral baseline of the candidate personnel handling the same type of sub-process in the storage medium to calculate the deviation of the current processing efficiency. When the deviation exceeds a preset deviation threshold, the context information of the diverted sub-process request is vector-matched with the professional domain knowledge graph of the candidate personnel to evaluate their real-time processing ability score. When the real-time processing capability score is lower than the preset capability lower limit, the collaborative scheduler is activated: a collaborative processing request is sent to the collaborative node corresponding to another candidate with relevant professional knowledge and the shortest current task queue length, along with the context information and deviation data. Alternatively, activate the secondary distribution mechanism: remove the distributed subprocess request from the to-do list of the current candidate node and reinsert it into the global allocation queue, then select the other candidate node with the highest overall matching degree for secondary allocation.

4. The method according to claim 3, characterized in that, The step of sending a collaborative processing request to the collaborative node corresponding to another candidate with relevant expertise and the shortest current task queue length further includes: Based on the context information of the diverted sub-process requests, at least one core professional domain feature tag vector is extracted using a text embedding model. The professional skill tag vectors of multiple candidates are obtained from the collaborative ability pool, and the cognitive load index is predicted by Kalman filtering. Calculate the cosine similarity between the core professional domain feature tag vector and the professional skill tag library vector to obtain the professional matching score; The professional matching score and the cognitive load index are weighted and summed to generate a comprehensive matching score; The candidate with the highest overall matching score is selected as the collaborative processing personnel, and the collaborative processing request is sent to the terminal corresponding to the collaborative processing personnel through a message queue. The collaborative processing request contains the message body snapshot and serialized data of the diverted sub-process request.

5. The method according to claim 4, characterized in that, The calculation steps for the cognitive load index obtained through Kalman filtering prediction further include: The system log collector obtains in real time the operation frequency, document viewing time, system notification response time, speaking frequency in collaboration tools, and email reply speed of the terminal devices corresponding to the candidates, as a multi-dimensional observation vector. The multidimensional observation vector is input into a real-time task engagement calculation model based on exponential weighted moving average to obtain an engagement score. The schedule and leave records of the candidates are obtained through a calendar application programming interface and used as time constraints. The engagement score is mapped to the free time periods in the schedule information and vacation records, and combined with the preset cognitive load level mapping table, the current cognitive load level value of the candidate is output. The cognitive load level value is input into the Kalman filter as the observation value. The Kalman filter outputs the cognitive load index prediction value at the current moment based on the load prediction value at the previous moment and the current observation value.

6. The method according to claim 5, characterized in that, After the step of outputting the current cognitive load level value of the candidate, the method further includes: Obtain the actual work performance data generated by the candidate personnel during the processing of the diverted sub-process requests. The actual work performance data includes the absolute time of task completion, the business rollback rate after approval, and the sentiment analysis score of task feedback. The Pearson correlation coefficient was calculated between the cognitive load level values ​​and the actual work performance data to identify the statistical deviation between the predicted load and the actual performance. When there is a statistical bias and the bias exceeds the preset tolerance range, the gradient descent method is used to adjust the decay factor parameter in the exponentially weighted moving average calculation model or the threshold parameter in the cognitive load level mapping table, so as to improve the fit between the subsequent output cognitive load level value and the actual work performance data. The adjusted parameters are stored in the model parameter library, and the parameter version number and timestamp are recorded.

7. The method according to claim 6, characterized in that, After the step of adjusting the decay factor parameter in the exponentially weighted moving average calculation model or the threshold parameter in the cognitive load level mapping table using the gradient descent method, the method further includes: Select a historical sub-process request dataset within a preset time range from the storage medium as the verification set; The cognitive load assessment model with adjusted parameters was used to simulate the load classification of candidates in the validation set, and the simulated load classification results were obtained. The simulated load classification results are compared with the corresponding actual work performance data in the validation set to calculate the simulation accuracy and misjudgment rate. Determine whether the accuracy rate is lower than a preset accuracy threshold or whether the false positive rate is higher than a preset false positive rate upper limit; When the accuracy rate is lower than the accuracy rate threshold or the false judgment rate is higher than the false judgment rate upper limit, a parameter adjustment risk signal is generated and the parameter adjustment risk signal is pushed to the operation and maintenance terminal through the system management interface. In response to the pushed parameter adjustment risk signal, the report generator is invoked to create a parameter risk analysis data package, and the parameter risk analysis data package is sent to the operation and maintenance terminal. The parameter risk analysis data package contains model parameter identifier and data source identifier. The operation and maintenance terminal provides a version management interface, which includes an instruction entry point for rolling back to historical stable model parameters.

8. The method according to claim 2, characterized in that, After the risk data packet is forwarded to the approval terminal via the internal data bus, it also includes: In response to the completion of the approval operation by the proxy node or the superior node, an approval result data packet is generated. Based on the routing information carried in the approval result data packet, the approval result data packet is sent to at least one of the following target devices: digital signature device, electronic signature server, printer, work order system interface, and notification SMS gateway; Receive the processing status code returned by the target device, and update the status record of the sub-process request according to the processing status code.

9. An intelligent office automation process management system, characterized in that, include: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor. When the instructions are executed by the at least one processor, the at least one processor performs the intelligent office automation process management method according to any one of claims 1 to 8. The system also includes at least one of the following output interfaces: a printer interface, a work order system application programming interface, an SMS gateway interface, and a digital signature device interface, for outputting approval results.

Images