Intelligent agent implementation method and system based on workflow
By dividing the work process into multiple nodes, assessing user confidence and loss mitigation capabilities, and dynamically adjusting the workflow, the problem of existing systems being unable to adapt is solved, resulting in higher task success rates and resource utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG POWER GRID CO LTD
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-12
AI Technical Summary
Existing workflow-based intelligent agent systems lack dynamic policy controllers and real-time state awareness mechanisms, which prevents workflows from being optimized or adapted, increases technical barriers and development costs, limits scalability and user flexibility in self-adjustment, and makes it difficult to achieve true intelligent adaptation in open environments or cross-domain businesses.
By dividing the workflow into multiple nodes, evaluating users' historical task information and real-time status, dynamically adjusting the node order and success rate of the workflow, and combining user confidence and loss mitigation capabilities to select the most suitable workflow, dynamic response and resource optimization are achieved.
It improves the dynamic adaptability and task success rate of workflow agents, reduces resource waste, and enhances user experience and execution efficiency.
Smart Images
Figure CN122022433A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of intelligent agent workflow, and in particular to a workflow-based intelligent agent implementation method and system. Background Technology
[0002] In the current development of workflow-based intelligent agent systems, existing methods primarily focus on improving the automation and reliability of task execution through predefined process rules and static node logic. However, existing systems generally lack effective dynamic policy controllers or real-time state awareness mechanisms, making runtime optimization or node-level adaptive refactoring of workflows impossible. Furthermore, existing solutions largely rely on manual configuration and debugging, which not only increases the technical threshold and development costs but also limits the scalability of workflows and the flexibility for user-defined adjustments. This rigid design pattern makes it difficult for workflow intelligent agents to achieve true "intelligent adaptation" in open environments or cross-domain business scenarios, ultimately leading to a degraded user experience and low execution efficiency. Summary of the Invention
[0003] The purpose of this invention is to provide a workflow-based intelligent agent implementation method and system to solve the problems mentioned in the background art.
[0004] Firstly, this application provides a workflow-based intelligent agent implementation method, which adopts the following technical solution: The process of obtaining work tasks is divided into multiple nodes, and multiple basic workflows are formed by arranging and combining these nodes. Obtain the user's historical work task information, and assess the user's basic confidence value for the work task based on the historical work task information; Obtain the basic work nodes in the basic workflow, record the nodes before the basic work nodes as predecessor nodes, and combine the basic confidence value to obtain the node confidence value of the basic work nodes. Obtain the basic success rate of the basic work nodes, combine the node confidence value to obtain the node success rate of the basic work nodes, and evaluate the workflow success rate of the basic workflow based on the node success rate. Based on historical task information, the user's workflow loss mitigation capability is assessed using the basic workflow. Combined with the workflow success rate, the user's workflow is selected from the basic workflow and pushed to the user's end.
[0005] Preferably, the step of obtaining the user's historical work task information and assessing the user's basic confidence value for the work task based on the historical work task information is as follows: Determine whether the user has performed the same task based on historical task information. If the user has performed the same task, extract the same task information from the historical task information. Obtain the average execution time of all users in the task, calculate the average execution time of each user and record it as the user execution time, and calculate the time difference between the average execution time and the user execution time. The user's proficiency confidence score is obtained by combining information on the same work task and the time difference. The information on the same work task includes the historical task success rate and the average number of historical task executions. If the same work task has not been performed before, a similar work task is obtained, and a proficiency confidence value is obtained based on the work evaluation of the similar task. Obtain the user's real-time status, assess the status confidence value based on the user's real-time status, and combine the proficiency confidence value to obtain the user's basic confidence value for the work task.
[0006] Preferably, if the same work task has not been performed before, the step of obtaining a similar work task and calculating the proficiency confidence value based on the similar task work evaluation is as follows: Historical tasks are extracted from historical work task information, and the task similarity between historical tasks and work tasks is compared. Similar work tasks are then selected based on task similarity. Obtain historical task information for similar tasks, calculate the average task similarity, and combine the historical task information to obtain a similarity confidence value; Extract the task thinking patterns of historical tasks, compare the similarity of the task thinking patterns of historical tasks with those of current tasks, and filter out the thinking tasks based on the similarity of thinking. Assess users' knowledge transfer ability score, calculate the average thinking similarity of thinking tasks, and combine the knowledge transfer ability score to calculate the thinking confidence score. The proficiency confidence score is obtained by combining similarity confidence score and thinking confidence score.
[0007] The preferred step for assessing a user's knowledge transfer ability is as follows: The number of tasks using the same thinking pattern actively by the user and the first time the user used it is counted and recorded as the number of actively executed tasks. The average time between a user's first active application of the same thought pattern and the first application is recorded as the active execution time. Statistically analyze the number of task domains included in historical tasks corresponding to the same thinking pattern, and calculate the correct application rate of transferring the same thinking pattern. The user's knowledge transfer capability value is obtained by combining the number of proactive executions, the duration of proactive executions, the number of task domains, and the correct application rate.
[0008] Preferably, the steps of obtaining the user's real-time status and assessing the status confidence value based on the user's real-time status are as follows: The user's standard state is extracted from historical work task information, and the similarity between the user's real-time state and the user's standard state is obtained by comparison. Collect real-time task scenarios for work tasks and analyze the difficulty of presenting tasks in real-time scenarios; The user's fatigue level is obtained, and a state confidence value is obtained by combining state similarity and task difficulty.
[0009] Preferably, the steps of obtaining the basic work nodes in the basic workflow, recording the nodes preceding the basic work nodes as predecessor nodes, and combining the basic confidence value evaluation to obtain the node confidence value of the basic work nodes are as follows: Obtain the completion difficulty of basic working nodes and statistically analyze the correlation between node completion difficulty and confidence value; The confidence values corresponding to the preceding nodes are obtained by finding the relationships and summing them to get the preceding confidence values. The number of preceding nodes is counted, and the node confidence value of the basic working node is obtained by combining the preceding confidence value and the basic confidence value.
[0010] Preferably, the steps of obtaining the basic success rate of the basic work nodes, combining the node confidence value to obtain the node success rate of the basic work nodes, and evaluating the workflow success rate of the basic workflow based on the node success rate are as follows: Set a node success rate threshold, and designate basic working nodes with a success rate lower than the node success rate threshold as failed nodes. Count the number of failed nodes, calculate the average success rate of the failed nodes and record it as the average success rate; The average impact of failed nodes on the work task is obtained, and the workflow success rate of the basic workflow is evaluated by combining the number of nodes and the average success rate.
[0011] Preferably, the step of assessing the user's workflow loss mitigation capability value for the basic workflow based on historical task information is as follows: Obtain information on how users have processed historical tasks, and assess the user's loss mitigation capability based on this information. Get the average success rate of all users at the failed node and record it as the other success rate; get the percentage of the task volume of the preceding node in the basic workflow. Get the number of failed nodes before the preceding nodes, and get the number of failed nodes after the basic working node and record it as the number of failed nodes after the preceding nodes; Calculate the difference between the number of subsequent failures and the number of preceding failures, and combine it with other success rates and task volume ratios to obtain the node stop loss value of the basic work node; The user's workflow loss prevention capability value is obtained by combining the user's loss prevention capability value and the node loss prevention value.
[0012] Preferably, the step of obtaining the user's processing information for historical work tasks and evaluating the user's loss-stopping capability value based on the processing information is as follows: Use processed information to find users' failed tasks and collect the average persistence time of users at the failure nodes in the failed tasks; Failed tasks whose average persistence time reaches the threshold without a preset persistence threshold are recorded as stop-loss tasks; otherwise, they are recorded as non-stop-loss tasks. Calculate the average number of execution nodes for stop-loss tasks and record it as the number of stop-loss nodes; The average duration of non-stop-loss tasks at the failure point is statistically analyzed and recorded as the non-stop-loss duration; The user's stop-loss capability value is obtained by combining the number of stop-loss points and the duration of non-stop-loss.
[0013] Secondly, this application provides a workflow-based intelligent agent implementation system, which adopts the following technical solution: The basic sequence module obtains the workflow of a task, divides the workflow of a task into multiple nodes, and forms multiple basic workflows by arranging and combining the nodes. The basic confidence module obtains the user's historical work task information and evaluates the user's basic confidence value for the work tasks based on the historical work task information. The node confidence module obtains the basic work nodes in the basic workflow, records the nodes before the basic work nodes as predecessor nodes, and evaluates the node confidence value of the basic work nodes in combination with the basic confidence value. The work success module obtains the basic success rate of the basic work nodes, combines the node confidence value to obtain the node success rate of the basic work nodes, and evaluates the workflow success rate of the basic workflow based on the node success rate. The stop-loss selection module evaluates the user's workflow stop-loss capability value for the basic workflow based on historical task information, and selects the user's workflow from the basic workflow based on the workflow success rate and pushes it to the user's terminal.
[0014] In summary, this application includes at least one of the following beneficial technical effects: 1. The workflow of a task is divided into multiple nodes, and these nodes are arranged and combined to form multiple basic workflows. Based on the user's historical task information, a basic confidence value for the task is assessed. The confidence values of preceding nodes are then obtained by finding the correlation between the completion difficulty and confidence value of each basic workflow node and summing these values. The node confidence value of each basic workflow node is obtained by combining the number of preceding nodes and the basic confidence value. Failed nodes are identified using node success rates. The success rate of the basic workflow is assessed by evaluating the number of failed nodes, the average node success rate, and the average impact of failed nodes on the task. Finally, the user's workflow loss mitigation capability is assessed based on historical task information. Combined with the workflow success rate, the user's workflow is selected from the basic workflows and pushed to the user. By using the node confidence values in the workflow nodes, different workflow success rates are evaluated. Then, combined with the loss mitigation capability values of different workflows, the most suitable workflow for the user is selected. This enables the workflow to dynamically respond to user individuality, improving the dynamic adaptability of the workflow-based intelligent agent.
[0015] 2. If a user has performed the same task before, the historical task success rate and average number of executions are extracted from historical task information. The user's proficiency confidence score is calculated by combining this with the difference between the average execution time of all users in the task and the user's average execution time. If the user has not performed the same task before, a similarity confidence score is calculated based on historical task information of similar tasks. Similar thinking tasks are selected based on task thinking patterns, and the average thinking similarity is calculated based on these tasks. The user's knowledge transfer ability score is calculated using the number of tasks performed with the same thinking pattern before the user actively applied it, the average time between the user's first application of the same thinking pattern, the number of task domains included in the historical tasks corresponding to the same thinking pattern, and the correct application rate of the same thinking pattern. A thinking confidence score is calculated by combining this with the average thinking similarity score. A proficiency confidence score is then calculated by combining the similarity confidence score and the thinking confidence score. A state confidence score is calculated by combining the similarity between the user's real-time state and standard state, the difficulty of the real-time task scenario, and the user's fatigue level. This state confidence score is then combined with the proficiency confidence score to obtain the user's basic confidence score for the task. Assessing basic confidence levels helps determine the impact of different workflows on users, thus enabling the selection of workflows that are more suitable for users. Dynamically adjusting workflows improves the success rate of tasks. This increases the success rate of tasks implemented by workflow-based agents.
[0016] 3. Failed tasks are categorized into stop-loss tasks and non-stop-loss tasks based on the average persistence time of users at failed nodes. The user's stop-loss capability value is obtained by combining the average number of execution nodes for stop-loss tasks and the average persistence time of non-stop-loss tasks at failed nodes. The node stop-loss value for the basic workflow node is obtained based on the average success rate of all users at the failed node, the proportion of the task volume of the preceding node to the total task volume of the basic workflow, and the difference between the number of subsequent failures and the number of preceding failures. Combining the user's stop-loss capability value and the node stop-loss value yields the user's overall workflow stop-loss capability value for the basic workflow. Evaluating workflow stop-loss capability helps reduce resource waste during task execution. Selecting workflows with higher resource utilization improves the resource utilization of workflow-based intelligent agents. Attached Figure Description
[0017] Figure 1 This is a schematic diagram illustrating the specific steps of an embodiment of a workflow-based intelligent agent implementation method according to the present invention.
[0018] Figure 2 This is a schematic diagram of the module connections of an embodiment of a workflow-based intelligent agent implementation system according to the present invention. Detailed Implementation
[0019] The following examples and... Figures 1-2 The present invention will be described in further detail, but the embodiments of the present invention are not limited thereto.
[0020] This invention discloses a workflow-based intelligent agent implementation method, which specifically includes the following steps: Step S1: Obtain the workflow of the work task, divide the workflow of the work task into multiple nodes, and arrange and combine the nodes to form multiple basic workflows.
[0021] Work tasks typically have multiple step nodes. Some step nodes are fixed, while others can be rearranged in order. By arranging and combining the nodes that can be rearranged in order with the nodes in the fixed order, we can obtain multiple workflows with different node sequences, which serve as the basic workflow.
[0022] Step S2: Obtain the user's historical work task information, and assess the user's basic confidence value for the work task based on the historical work task information.
[0023] Step S3: Obtain the basic work nodes in the basic workflow, record the nodes before the basic work nodes as predecessor nodes, and obtain the node confidence value of the basic work nodes by combining the basic confidence value evaluation.
[0024] Step S4: Obtain the basic success rate of the basic work nodes, combine the node confidence value to obtain the node success rate of the basic work nodes, and evaluate the workflow success rate of the basic workflow based on the node success rate.
[0025] Step S5: Evaluate the user's workflow loss mitigation capability value for the basic workflow based on historical work task information, and select the user's workflow from the basic workflow based on the workflow success rate and push it to the user's end.
[0026] The weighted average method is used to calculate the fit between the basic workflow and the user, and the basic workflow with the highest fit is selected as the user workflow and pushed to the user's end. This can balance the success rate of tasks and the resource waste caused by tasks, that is, while ensuring a certain success rate of tasks, it also minimizes resource waste as much as possible. The weight ratio can be set by the user based on experience. When the user needs to deal with the failure of the task more, the weight ratio corresponding to the workflow's success rate is set higher. Conversely, when the user needs to reduce costs and increase efficiency more, the weight ratio corresponding to the workflow's loss mitigation capability value can be set higher.
[0027] In practical applications, existing technologies typically use fixed workflows. However, each user has different habits in executing tasks, and because task scenarios change in real time, fixed workflows struggle to adapt to different users and situations. This severely impacts task execution for most users, reducing efficiency and, in some scenarios, lowering success rates and affecting the user experience. For example, User A prefers placing easier nodes first and more difficult nodes last, achieving the highest efficiency with this node order. However, with a fixed workflow placing difficult nodes first and easier nodes last, User A might abandon the task due to excessive difficulty. This not only reduces User A's efficiency but also lowers the success rate. Dynamically adjusting workflows based on different users improves overall task execution efficiency and success rate, and enhances the smoothness of the task execution process.
[0028] The steps for obtaining a user's historical work task information and assessing the user's basic confidence level in their work tasks based on this information are as follows: Step S21: Determine whether the user has performed the same task based on the historical task information. If the same task has been performed, extract the same task information from the historical task information.
[0029] Historical task information includes tasks that the user has executed in the past, the execution time of different tasks, the number of times they were executed, and other information.
[0030] Step S22: Obtain the average execution time of all users in the task, calculate the average execution time of users and record it as the user execution time, and calculate the time difference between the average execution time and the user execution time.
[0031] The average execution time for all users refers to the average execution time of all users who have performed work tasks.
[0032] Step S23: Combine the information of the same task with the duration difference to obtain the user's proficiency confidence value. The information of the same task includes the success rate of historical tasks and the average number of times historical tasks are executed.
[0033] The user's proficiency confidence score is calculated using a weighted average method. A larger difference in time indicates that the user is more proficient in the task than other users. A higher historical task success rate and a higher average number of historical task executions indicate that the user is more capable of completing the task. When users are proficient and capable of completing a task, they are more confident in their ability to do so, resulting in a higher proficiency confidence score.
[0034] Step S24: If the same work task has not been performed before, obtain similar work tasks and obtain the proficiency confidence value based on the similar task work evaluation.
[0035] Step S25: Obtain the user's real-time status, evaluate the user's status confidence value based on the real-time status, and combine the proficiency confidence value to obtain the user's basic confidence value for the work task.
[0036] In practical applications, workflows with different node orders can affect the effectiveness of users completing tasks, with user confidence in the task being a crucial indicator. The order of nodes in a workflow influences fluctuations in user confidence during task execution, and is also affected by the user's initial confidence level. When users are confident and proficient in completing the task, changes in the workflow node order have little impact, as any node is easily manageable. However, when users are less confident, the workflow node order significantly affects their performance. Therefore, when evaluating the success rate of executing tasks according to a workflow, it is essential to consider both the impact of node order and the user's level of understanding of the task.
[0037] If the same work task has not been performed before, the steps to obtain a similar work task and calculate the proficiency confidence value based on the similar task work assessment are as follows: Step S241: Extract historical tasks based on historical task information, compare the task similarity between historical tasks and current tasks, and filter similar tasks based on task similarity.
[0038] The cosine similarity model is used to compare and obtain the task similarity. A task similarity threshold is set. Historical tasks whose task similarity reaches the task similarity threshold are regarded as similar work tasks.
[0039] Step S242: Obtain historical task information for similar tasks, calculate the average task similarity, and combine the historical task information to obtain a similarity confidence value.
[0040] The task information for similar tasks includes the historical task success rate and the average number of historical task executions. The average time a user takes to execute a similar task is calculated as the difference between the average time a user takes to execute the same task and the average time a user takes to execute the same task. Combined with the average task similarity, a similarity confidence value is calculated, similar to the method used to calculate the proficiency confidence value.
[0041] Step S243: Extract the task thinking patterns of historical tasks, compare the thinking similarity between the task thinking patterns of historical tasks and work tasks, and filter out the thinking work tasks based on the thinking similarity.
[0042] Task-oriented thinking patterns refer to the stable characteristics of an individual's cognitive strategies, information processing methods, and decision-making tendencies when completing a task. Users can set different task-oriented thinking patterns, such as divergent thinking, holistic thinking, and economical thinking. The similarity of these thinking patterns is obtained by comparing them using a cosine similarity model.
[0043] Step S244: Assess the user's knowledge transfer ability value, calculate the average value of the thinking similarity of the thinking task, and calculate the thinking confidence value by combining the knowledge transfer ability value.
[0044] The confidence score is calculated using a weighted average method. The higher the knowledge transfer ability score and the higher the average value of thinking similarity, the more capable the user is of completing the task, and therefore the higher their confidence score will be.
[0045] Step S245: Combine similarity confidence value and thinking confidence value to obtain proficiency confidence value.
[0046] In practical applications, after normalizing the similarity confidence score and the thinking confidence score, the summation yields the proficiency confidence score. When a user has not performed the same task before, it is difficult to determine their proficiency level. In this case, the user's proficiency confidence score can be estimated based on similar tasks. When a user has performed a task with similar content, they are more confident in completing it due to the high degree of similarity. Furthermore, when thinking patterns are similar, users do not need to rebuild a new mental framework, allowing for smoother information processing and significantly reduced cognitive load, thus increasing their confidence. Moreover, when users have stronger knowledge transfer abilities, they are better able to apply existing thinking patterns to the task, enabling them to more clearly assess their abilities and task difficulty, reducing anxiety caused by uncertainty, and generating positive expectations for the task, resulting in even higher confidence scores.
[0047] The steps for assessing a user's knowledge transfer ability are as follows: Step S2441: Count the number of tasks with the same thinking pattern that the user actively applies and the first application, and record them as the number of active executions.
[0048] For example, if a user passively performs five tasks using the Mindset No. 1 between their first application of it and their second proactive application, then the number of proactive executions is five. Proactive application means the user did not receive guidance or assistance from other users.
[0049] Step S2442: Calculate the average time between the user's active application of the same thinking pattern and the first application, and record it as the active execution time.
[0050] The average time between the first and second use of the same thought pattern by a user is taken as the active execution time.
[0051] Step S2443: Count the number of task domains contained in the historical tasks corresponding to the same thinking pattern, and count the correct application rate of transferring the same thinking pattern.
[0052] If the performance of a work task using the transferred mindset meets the standards or achieves the expected results, it is considered a correct application. The correct application rate is calculated as the ratio of the number of correct applications of the transferred mindset to the total number of applications.
[0053] Step S2444: Combine the number of active executions, the duration of active executions, the number of task domains, and the correct application rate to obtain the user's knowledge transfer capability value.
[0054] In practical applications, a weighted average method can be used to calculate the knowledge transfer capability score. Fewer active executions and shorter execution times indicate that the user can transfer knowledge more quickly, reflecting a stronger ability to accept and transfer knowledge. A larger number of task domains and a higher correct application rate indicate a broader scope of knowledge transfer and a more accurate application of that knowledge, resulting in a higher knowledge transfer capability score. Stronger knowledge transfer capabilities lead to greater familiarity with similar tasks and a greater sense of control over them, thus increasing confidence in completing tasks.
[0055] The steps for obtaining a user's real-time status and assessing the status confidence value based on that status are as follows: Step S251: Extract the user's standard state based on historical work task information, and compare the user's real-time state with the user's standard state to obtain the state similarity.
[0056] If a user has performed the same task before, that task is retrieved; otherwise, a similar task is retrieved as a reference task. The completion effect of the reference task is then recorded, and reference tasks whose completion effects reach a preset threshold are selected. The user status corresponding to these reference tasks is collected. Common indicators among these user statuses are collected, and a standard user status is formed based on these common indicators. Common indicators include user emotion, user health status, etc., and a cosine similarity model is used to compare and obtain the status similarity.
[0057] Step S252: Collect real-time task scenarios of work tasks and calculate the difficulty of presenting work scenarios in real-time.
[0058] Real-time task scenarios encompass workload, timeframe, and requirements. The difficulty of a real-time task can be assessed by the user or determined by the actual performance of the task within that scenario. For example, given the same task content, task 1 requires completion within ten days, while task 2 requires completion within five days. Even if tasks 1 and 2 are identical, the difference in time requirements clearly indicates that task 2 is more challenging.
[0059] Step S253: Obtain the user's fatigue level and combine it with state similarity and task presentation difficulty to obtain a state confidence value.
[0060] In practical applications, user fatigue levels can be collected through questionnaires, and a weighted average method can be used to calculate a state confidence score. A customer's confidence in completing a task is influenced not only by their familiarity with the task but also by their current state. When a user is in a better state, they are obviously more confident in completing the task. Conversely, when a user is in a less favorable state, their self-assessment will lower their confidence. The greater the user's fatigue, the lower their confidence in completing the task. This is because fatigue reduces cognitive resources, making it difficult for users to maintain attention and executive control; this decline in cognitive ability directly translates into doubt about their ability to complete the task.
[0061] The steps for obtaining the basic work nodes in the basic workflow, recording the nodes preceding the basic work nodes as predecessor nodes, and combining the basic confidence value evaluation to obtain the node confidence value of the basic work nodes are as follows: Step S31: Obtain the node completion difficulty of the basic working nodes and calculate the correlation between node completion difficulty and confidence value.
[0062] Node completion difficulty refers to the user's assessment of the difficulty of completing a basic work node. This difficulty can be obtained by evaluating the average completion time of basic work nodes or by directly collecting user-assessed node completion difficulty data. The higher the completion difficulty of a node, the greater the user's confidence in subsequent tasks upon completion. The correlation between node completion difficulty and confidence refers to the relationship between the node completion difficulty and the increase in user confidence after completing a node. This can be established by collecting data on the increase in user confidence after completing nodes with different completion difficulties and mapping them one-to-one. For example, if a user completes node A (node completion difficulty 80), their confidence increases by 8. If a user completes node B (node completion difficulty 50), their confidence increases by 5.
[0063] Step S32: Find the confidence value corresponding to the preceding node based on the association relationship and sum them to obtain the preceding confidence value.
[0064] Find the confidence values of the nodes preceding the base worker node, sum them to get the preceding confidence values of the base worker node, assuming that the preceding nodes have all completed successfully.
[0065] Step S33: Count the number of preceding nodes, and combine the preceding confidence value and the basic confidence value to obtain the node confidence value of the basic working node.
[0066] In practical applications, the confidence value of each preceding node is calculated using a weighted average method based on the number of preceding nodes and their confidence values. The node confidence value is then obtained by summing the preceding node confidence values and the baseline confidence value after normalization. The more preceding nodes and the higher the baseline confidence value, the more likely the user has successfully completed most of the work nodes in the workflow, leading to greater confidence in subsequent tasks. Therefore, whether the user is confident in successfully completing the current basic work nodes depends not only on their proficiency with the task but also on their completion of the preceding tasks.
[0067] The steps to obtain the basic success rate of the basic work nodes, combine the node confidence value to obtain the node success rate of the basic work nodes, and evaluate the workflow success rate of the basic workflow based on the node success rate are as follows: Step S41: Set a node success rate threshold and designate basic working nodes with a success rate lower than the node success rate threshold as failed nodes.
[0068] Whether a basic task node can be successfully completed depends on two factors: its difficulty and the user's mindset. The stronger the user's confidence in the basic task node, the more likely they are to complete it. The basic success rate is calculated by dividing the number of users who successfully complete a basic task node by the total number of users. Multiplying this basic success rate by the node confidence value yields the node success rate.
[0069] Step S42: Count the number of failed nodes, calculate the average success rate of the failed nodes and record it as the average success rate.
[0070] Step S43: Obtain the average impact of failed nodes on the work task, and evaluate the workflow success rate of the basic workflow by combining the number of nodes and the average success rate.
[0071] In practical applications, the average impact of failed nodes on a task can be assessed by users based on the task's completion effect after a failed node. The workflow success rate is then calculated using a weighted average method. Some nodes have low success rates, indicating a high probability of failure during execution. The success rate of a node is related to user confidence; higher user confidence increases the success rate. This leads to varying success probabilities across workflows with different node sequences, due to the different positions of the nodes. A higher number of failed nodes in a workflow results in a higher overall failure rate. Conversely, a lower average success rate for failed nodes increases the likelihood of task failure, thus lowering the overall workflow success rate. When a failed node has a significant impact on the task, its failure severely affects the overall task completion, making users more likely to abandon the task and less inclined to actively attempt remediation, further lowering the overall task success rate and consequently, the workflow success rate.
[0072] The steps for assessing a user's workflow loss mitigation capability based on historical task information are as follows: Step S51: Obtain the user's processing information for historical work tasks, and evaluate the user's loss prevention capability value based on the processing information.
[0073] Step S52: Obtain the average success rate of all users of the failed node and record it as the other success rate; obtain the percentage of the task volume of the preceding node in the basic workflow.
[0074] Task volume ratio refers to the ratio of the task volume of the preceding node to the total tasks of the basic workflow.
[0075] Step S53: Obtain the number of failed nodes in the preceding nodes and the number of failed nodes after the basic working nodes and record it as the number of failed nodes after the preceding nodes.
[0076] Step S54: Calculate the difference between the number of subsequent failures and the number of preceding failures, and combine it with other success rates and task volume ratios to obtain the node stop loss value of the basic working node.
[0077] The weighted average method can be used to calculate the stop-loss value for each node. A larger difference in quantity indicates that subsequent work is more difficult than the already completed work. When users struggle to complete the current basic work node, they are more likely to abandon the task and cut their losses due to the even greater difficulty of subsequent work, resulting in a larger stop-loss value. Conversely, a higher success rate reflects the feasibility of the node, giving users hope of completion and making them less likely to abandon it, thus resulting in a smaller stop-loss value. Furthermore, a larger task volume percentage indicates greater time, effort, and cost invested in the initial stages, making it more difficult to abandon the task at the basic work node. Users are less likely to abandon the invested costs, thereby lowering the stop-loss value.
[0078] Step S55: Combine the user's stop-loss capability value and the node stop-loss value to obtain the user's workflow stop-loss capability value for the basic workflow.
[0079] In practical applications, the workflow loss mitigation capability value is obtained by normalizing the user's loss mitigation capability value and the node loss mitigation capability value, and then summing them. When completing a task through a workflow, it is necessary to consider not only the impact of the order of work nodes on the success rate of the task, but also whether loss mitigation can be achieved in a timely manner when the task fails. For example, if a task is difficult to succeed, placing the difficult step in the first step allows the user to abandon the task promptly if it fails, reducing the consumption of costs and resources. However, if the difficult step is placed in the last step, because a large amount of manpower and resources have already been invested in the early stages, the user will be reluctant to abandon the task and will continue to invest time and energy at the difficult point. In the case of failure in both cases, it is clear that the latter wastes a lot of resources at the difficult point. Considering the loss mitigation capability value of different workflows can significantly reduce unnecessary costs. The workflow loss mitigation capability value is not only affected by the order of nodes, but also by the individual characteristics of the user. By evaluating users and workflows from multiple perspectives, a more accurate workflow loss mitigation capability value can be obtained.
[0080] The steps for obtaining user processing information on historical work tasks and assessing the user's stop-loss capability value based on this information are as follows: Step S511: Use the processed information to find the user's failed tasks and collect the average persistence time of the user at the failed nodes in the failed tasks.
[0081] The processed information includes the results of users' processing of historical tasks, processing time, and other information.
[0082] Step S512: Record failed tasks whose average persistence time reaches the threshold without a preset persistence threshold as stop-loss tasks; otherwise, record them as non-stop-loss tasks.
[0083] Step S513: Calculate the average number of execution nodes for the stop-loss task and record it as the number of stop-loss nodes.
[0084] Step S514: Calculate the average duration of non-stop-loss tasks at the failure node and record it as the non-stop-loss duration.
[0085] Step S515: Combine the number of stop-loss nodes and the duration of non-stop-loss to obtain the user's stop-loss capability value.
[0086] In practical applications, the weighted average method used in the examples calculates the user's stop-loss capability value. A larger number of stop-loss points indicates that the user is more likely to disregard initial investments and cut losses promptly, meaning they are more likely to adhere to the principle of "sunk costs not being involved in major decisions," thus demonstrating stronger stop-loss capability. Conversely, a longer non-stop-loss period reflects a greater inability for the user to make timely stop-loss decisions, resulting in a lower stop-loss capability value. When the success rate of a task is low, selecting a workflow with a higher stop-loss capability value helps improve resource utilization. If a user has strong stop-loss capability, they can make correct decisions at failure points, and the order of workflow nodes has less impact on their stop-loss. Conversely, when a user's stop-loss capability is weak, the order of workflow nodes significantly affects resource waste during the actual execution of the task.
[0087] A workflow-based intelligent agent implementation system, by applying the workflow-based intelligent agent implementation method described above, includes: The basic sequence module obtains the workflow of the work task, divides the workflow of the work task into multiple nodes, and forms multiple basic workflows by arranging and combining the nodes.
[0088] The basic confidence module obtains the user's historical work task information and evaluates the user's basic confidence value for the work tasks based on the historical work task information.
[0089] The node confidence module obtains the basic work nodes in the basic workflow, records the nodes before the basic work nodes as predecessor nodes, and evaluates the node confidence value of the basic work nodes by combining the basic confidence value.
[0090] The work success module obtains the basic success rate of basic work nodes, combines the node confidence value to obtain the node success rate of basic work nodes, and evaluates the workflow success rate of the basic workflow based on the node success rate.
[0091] The stop-loss selection module evaluates the user's workflow stop-loss capability value for the basic workflow based on historical task information, and selects the user's workflow from the basic workflow based on the workflow success rate and pushes it to the user's terminal.
[0092] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A workflow-based intelligent agent implementation method, characterized in that, Includes the following steps: The process of obtaining work tasks is divided into multiple nodes, and multiple basic workflows are formed by arranging and combining these nodes. Obtain the user's historical work task information, and assess the user's basic confidence value for the work task based on the historical work task information; Obtain the basic work nodes in the basic workflow, record the nodes before the basic work nodes as predecessor nodes, and combine the basic confidence value to obtain the node confidence value of the basic work nodes. Obtain the basic success rate of the basic work nodes, combine the node confidence value to obtain the node success rate of the basic work nodes, and evaluate the workflow success rate of the basic workflow based on the node success rate. Based on historical task information, the user's workflow loss mitigation capability is assessed using the basic workflow. Combined with the workflow success rate, the user's workflow is selected from the basic workflow and pushed to the user's end.
2. The workflow-based intelligent agent implementation method according to claim 1, characterized in that, The steps for obtaining a user's historical work task information and assessing the user's basic confidence level in their work tasks based on this information are as follows: Determine whether the user has performed the same task based on historical task information. If the user has performed the same task, extract the same task information from the historical task information. Obtain the average execution time of all users in the work task, calculate the average execution time of each user and record it as the user execution time, and calculate the time difference between the average execution time and the user execution time; The user's proficiency confidence score is obtained by combining information on the same work task and the time difference. The information on the same work task includes the historical task success rate and the average number of historical task executions. If the same work task has not been performed before, a similar work task is obtained, and a proficiency confidence value is obtained based on the work evaluation of the similar task. Obtain the user's real-time status, assess the status confidence value based on the user's real-time status, and combine the proficiency confidence value to obtain the user's basic confidence value for the work task.
3. The workflow-based intelligent agent implementation method according to claim 2, characterized in that, If the same work task has not been performed before, the steps to obtain a similar work task and calculate the proficiency confidence value based on the similar task work assessment are as follows: Historical tasks are extracted from historical work task information, and the task similarity between historical tasks and work tasks is compared. Similar work tasks are then selected based on task similarity. Obtain historical task information for similar tasks, calculate the average task similarity, and combine the historical task information to obtain a similarity confidence value; Extract the task thinking patterns of historical tasks, compare the similarity of the task thinking patterns of historical tasks with those of current tasks, and filter out the thinking tasks based on the similarity of thinking. Assess users' knowledge transfer ability score, calculate the average thinking similarity of thinking tasks, and combine the knowledge transfer ability score to calculate the thinking confidence score. The proficiency confidence score is obtained by combining similarity confidence score and thinking confidence score.
4. The workflow-based intelligent agent implementation method according to claim 3, characterized in that, The steps for assessing a user's knowledge transfer ability are as follows: The number of tasks using the same thinking pattern actively by the user and the first time the user used it is counted and recorded as the number of actively executed tasks. The average time between a user's first active application of the same thought pattern and the first application is recorded as the active execution time. Statistically analyze the number of task domains included in historical tasks corresponding to the same thinking pattern, and calculate the correct application rate of transferring the same thinking pattern. The user's knowledge transfer capability value is obtained by combining the number of proactive executions, the duration of proactive executions, the number of task domains, and the correct application rate.
5. The workflow-based intelligent agent implementation method according to claim 2, characterized in that, The steps for obtaining a user's real-time status and assessing the status confidence value based on that status are as follows: The user's standard state is extracted from historical work task information, and the similarity between the user's real-time state and the user's standard state is obtained by comparison. Collect real-time task scenarios for work tasks and analyze the difficulty of presenting tasks in real-time scenarios; The user's fatigue level is obtained, and a state confidence value is obtained by combining state similarity and task difficulty.
6. The workflow-based intelligent agent implementation method according to claim 1, characterized in that, The steps for obtaining the basic work nodes in the basic workflow, recording the nodes preceding the basic work nodes as predecessor nodes, and combining the basic confidence value evaluation to obtain the node confidence value of the basic work nodes are as follows: Obtain the completion difficulty of basic working nodes and statistically analyze the correlation between node completion difficulty and confidence value; The confidence values corresponding to the preceding nodes are obtained by finding the relationships and summing them to get the preceding confidence values. The number of preceding nodes is counted, and the node confidence value of the basic working node is obtained by combining the preceding confidence value and the basic confidence value.
7. The workflow-based intelligent agent implementation method according to claim 1, characterized in that, The steps to obtain the basic success rate of the basic work nodes, combine the node confidence value to obtain the node success rate of the basic work nodes, and evaluate the workflow success rate of the basic workflow based on the node success rate are as follows: Set a node success rate threshold, and designate basic working nodes with a success rate lower than the node success rate threshold as failed nodes. Count the number of failed nodes, calculate the average success rate of the failed nodes and record it as the average success rate; The average impact of failed nodes on the work task is obtained, and the workflow success rate of the basic workflow is evaluated by combining the number of nodes and the average success rate.
8. The workflow-based intelligent agent implementation method according to claim 7, characterized in that, The steps for assessing a user's workflow loss mitigation capability based on historical task information are as follows: Obtain information on how users have processed historical tasks, and assess the user's loss mitigation capability based on this information. Get the average success rate of all users at the failed node and record it as the other success rate; get the percentage of the task volume of the preceding node in the basic workflow. Get the number of failed nodes before the preceding nodes, and get the number of failed nodes after the basic working node and record it as the number of failed nodes after the preceding nodes; Calculate the difference between the number of subsequent failures and the number of preceding failures, and combine it with other success rates and task volume ratios to obtain the node stop loss value of the basic work node; The user's workflow loss prevention capability value is obtained by combining the user's loss prevention capability value and the node loss prevention value.
9. The workflow-based intelligent agent implementation method according to claim 8, characterized in that, The steps for obtaining user processing information on historical work tasks and assessing the user's stop-loss capability value based on this information are as follows: Use processed information to find users' failed tasks and collect the average persistence time of users at the failure nodes in the failed tasks; Failed tasks whose average persistence time reaches the threshold without a preset persistence threshold are recorded as stop-loss tasks; otherwise, they are recorded as non-stop-loss tasks. Calculate the average number of execution nodes for stop-loss tasks and record it as the number of stop-loss nodes; The average duration of non-stop-loss tasks at the failure point is statistically analyzed and recorded as the non-stop-loss duration; The user's stop-loss capability value is obtained by combining the number of stop-loss points and the duration of non-stop-loss.
10. A workflow-based intelligent agent implementation system, characterized in that, The application of a workflow-based agent implementation method as described in any one of claims 1-9 includes: The basic sequence module obtains the workflow of a task, divides the workflow of a task into multiple nodes, and forms multiple basic workflows by arranging and combining the nodes. The basic confidence module obtains the user's historical work task information and evaluates the user's basic confidence value for the work tasks based on the historical work task information. The node confidence module obtains the basic work nodes in the basic workflow, records the nodes before the basic work nodes as predecessor nodes, and evaluates the node confidence value of the basic work nodes in combination with the basic confidence value. The work success module obtains the basic success rate of the basic work nodes, combines the node confidence value to obtain the node success rate of the basic work nodes, and evaluates the workflow success rate of the basic workflow based on the node success rate. The stop-loss selection module evaluates the user's workflow stop-loss capability value for the basic workflow based on historical task information, and selects the user's workflow from the basic workflow based on the workflow success rate and pushes it to the user's end.