Role tag-based agent dynamic interaction method

Abstract

The invention provides an agent dynamic interaction method based on role labels, and relates to the technical field of dynamic interaction methods, comprising the following steps: S1, role modeling and label distribution; defining a multi-dimensional role tag system; adopting an improved graph neural network algorithm; dynamically updating role tags through an incremental stochastic gradient descent algorithm; s2, dynamic interaction rule design; s3, real-time perception and decision optimization are carried out; s4, feedback and result integration; a sensor is deployed to collect environment data in real time, the environment data is input into a hybrid model fusing PPO reinforcement learning and CNN / RNN, and cross-scene feature adaptation is achieved in combination with transfer learning. The model outputs role switching probability distribution, and a decision path is dynamically optimized; an incremental learning mechanism enables the system to continuously adapt to environmental changes, and the curing defect of a traditional single model is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic interaction methods, and in particular to a dynamic interaction method of intelligent agents based on role tags. Background Art

[0002] With the rapid development of intelligent technology, intelligent systems have been widely penetrated into many fields such as medical care, education, and industrial production, playing a key role in handling complex decision-making and interactive tasks; but as application scenarios become increasingly complex, traditional intelligent systems have exposed many problems in the dynamic interaction of multiple roles and are difficult to meet actual needs, mainly reflected in two aspects: role definition and interaction rules, and decision-making optimization mechanism.

[0003] Limitations of role definition and interaction rules: Traditional intelligent systems often use static role definitions and hard-coded interaction rules. Role labeling systems are single and fixed, lacking multidimensionality and flexibility. For example, role attribute labels are simple to set. In the medical field, they can only distinguish between roles such as doctors and nurses, without being able to further segment roles by skill level (elementary / intermediate / advanced) or professional field (such as internal medicine, surgery, or diagnostic skills). In terms of constructing interaction rules, traditional systems do not adequately analyze task dependency graphs and historical interaction data. When constructing task dependency graphs, they do not clearly define task input / output and logical relationships, nor do they fully consider factors such as collaboration frequency and effectiveness, as well as task sequence, resource dependency, and information transmission. As a result, the generated role dependency relationships are inaccurate and cannot reflect the actual collaboration needs.

[0004] Deficiencies in decision-making optimization mechanisms: Existing decision-making optimization mechanisms rely on a single data source and model, lack real-time perception and feedback mechanisms, and use fixed algorithms and architectures to build decision-making models, making them difficult to adapt to different scenarios. For example, using only a single deep learning model without combining reinforcement learning and transfer learning techniques can fail to fully leverage the advantages of each model to improve decision quality. In terms of real-time perception, traditional systems lack data collection and analysis methods. During task execution, they are unable to collect or process key data such as task completion rate, resource utilization rate, and user satisfaction in real time. As a result, the system cannot timely grasp the environmental status and task execution status; In summary, traditional intelligent systems have shortcomings in role definition, interaction rules, and decision-making optimization mechanisms, making it difficult to meet the needs of complex and changing application scenarios. Therefore, there is an urgent need for a dynamic interaction method for intelligent agents based on role labels. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the existing technology and propose a dynamic interaction method that can dynamically adjust role labels and interaction rules according to actual conditions, perceive environmental changes in real time and optimize decision-making strategies, improve the system's response speed and decision-making quality in complex environments, and enhance the system's intelligence level and application scope.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for dynamic interaction of intelligent agents based on role tags, comprising the following steps: S1. Role modeling and label assignment: Through cross-disciplinary expert workshops and historical data analysis, a multi-dimensional role labeling system is defined, including basic role names and attribute labels. An improved graph neural network algorithm is used to combine task dependency graphs and historical interaction data to generate role dependency relationships. Role labels are dynamically updated using an incremental stochastic gradient descent algorithm, and an evaluation mechanism is established to verify the effectiveness of the updates. S2. Dynamic interaction rule design: Based on the role tag system, use the rule engine to configure conditional logic interaction rules; dynamically adjust interaction rules through real-time monitoring systems and stream processing technology; use distributed computing and sparse algorithms to optimize rule execution efficiency; S3, Real-time Perception and Decision Optimization: Deploy sensors to collect environmental data, pre-process it, and input it into a dynamic decision-making model. Build a decision-making model based on reinforcement learning or deep learning, and combine it with transfer learning for training to monitor the decision execution effect in real time and continuously optimize the model. S4. Integrate feedback and results; collect interaction result data, and update role labels and interaction rules through analysis; generate multimodal feedback reports, supporting multiple output methods such as email, voice, and visual interface; iterate and optimize the system based on user feedback.

[0007] Preferably, the attribute tags of the role tag system include: skill level (elementary / intermediate / advanced), professional field and available time tag, and the tag definition is customized and subdivided according to the application scenario, such as the medical field is subdivided into diagnostic skills and internal medicine / surgery professional field tags.

[0008] Preferably, the improved graph neural network algorithm includes: clarifying task input / output and logical relationships when constructing a task dependency graph; Analyze the frequency and effectiveness of collaboration in historical interaction data; The role dependency relationship is generated by integrating task sequence, resource dependence, and information transmission factors.

[0009] Preferably, the real-time monitoring system: Collect task completion rate, resource utilization rate, and user satisfaction data; Use the real-time stream processing framework to set up sliding windows to analyze data streams, trigger threshold warnings, and adjust rules.

[0010] Preferably, the dynamic decision model adopts the PPO algorithm or CNN / RNN architecture, matches the feature space of the source model and the target model through transfer learning, and uses incremental learning to adapt to environmental changes. The dynamic decision model takes the task status and resource occupancy rate in the environmental data as input and outputs the probability distribution of the intelligent agent role switching.

[0011] The multimodal feedback interface supports collaborative output of voice, email, and visual interface, and provides an adaptation strategy that prioritizes voice on the mobile side and visualization on the office side.

[0012] Preferably, the interaction rule optimization includes: Implement conditional logic configuration through the Drools engine; Use Kafka distributed scheduling framework to parallelize rule calculations; The effectiveness of the rules is double-verified based on simulation testing and actual scenario testing. When the rule execution success rate in the simulation test is ≥95% and the user satisfaction rate in the actual scenario test is ≥90%, the rule is judged to be valid.

[0013] Preferably, the dynamic updating of the role tag includes: Use incremental learning algorithms to process real-time interactive data; Regularly evaluate label accuracy and stability indicators; When the task completion rate deviation exceeds 15%, the label recalibration is triggered.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. Multi-dimensional dynamic role modeling: Through cross-disciplinary expert collaboration and historical data analysis, a multi-dimensional labeling system has been constructed, encompassing skill level, professional field, and available time. This overcomes the limitations of traditional single-dimensional role definitions (e.g., distinguishing only between doctors and nurses). Labels support scenario-specific customization (e.g., medical and surgical diagnostic skills) and are dynamically updated using an incremental stochastic gradient descent algorithm, making role descriptions more tailored to actual needs. The improved graph neural network algorithm integrates task input / output logic, historical collaboration frequency and effects (such as resource dependencies and information transmission paths) when constructing task dependency graphs to generate accurate role dependency relationships, avoiding the collaboration bias caused by traditional hard-coded rules.

[0015] 2. Adaptive interaction rule optimization: Use a rule engine (such as Drools) to configure conditional logic and combine it with real-time stream processing technology (such as Kafka sliding window analysis) to dynamically adjust rules. When task completion rates drop or user satisfaction falls below a threshold, rule reconstruction is automatically triggered to improve system response speed. The efficiency of rule execution is optimized through distributed computing and sparsification algorithms, and the effectiveness of rules is ensured through double verification (simulation test success rate ≥ 95% and actual scenario satisfaction ≥ 90%), which is significantly better than traditional static rule systems.

[0016] 3. Real-time Perception and Hybrid Decision-Making Model: Deploy sensors to collect real-time environmental data (such as task status and resource utilization). This data is then fed into a hybrid model that integrates PPO reinforcement learning with CNN / RNN, combined with transfer learning to achieve cross-scenario feature adaptation. The model outputs a probability distribution for role switching, dynamically optimizing the decision path.

[0017] The incremental learning mechanism enables the system to continuously adapt to environmental changes (such as sudden resource shortages), solving the rigid defects of traditional single models (such as pure deep learning).

[0018] 4. Closed-loop feedback and multimodal output: Integrate interaction result data to update labels and rules in a closed loop, forming a continuous optimization cycle. When the task completion rate deviation exceeds 15%, label calibration is automatically triggered to ensure the dynamic stability of the system. The multimodal feedback interface (voice / email / visualization) is adapted according to the scenario (mobile voice is prioritized), improving the user experience and expanding the scope of system application. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flowchart of the role tag-based intelligent agent dynamic interaction method provided by the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0021] To facilitate understanding of the present invention, the present invention will be described more comprehensively below with reference to relevant references, and several embodiments of the present invention are given. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0022] It should be noted that when an element is referred to as being "fixed on" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used in this article are for illustrative purposes only.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention pertains. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items. Example

[0024] like Figure 1 As shown, the present invention provides a technical solution: a dynamic interaction method for intelligent agents based on role tags. The overall method aims to achieve efficient dynamic interaction of intelligent agents in different scenarios by building a complete role tag system and designing flexible interaction rules, combining real-time perception and decision optimization as well as feedback and result integration mechanisms. Specifically, the method includes the following steps: S1, role modeling and label assignment; Building a character labeling system: Through cross-disciplinary expert workshops and historical data analysis, we defined a multi-dimensional character labeling system. These workshops brought together senior experts from diverse fields (such as healthcare, finance, and education) to discuss and determine basic character names and attribute labels appropriate for different scenarios. Historical data analysis collected various data from past agent interactions, including task execution and user feedback, to assist in building the labeling system. The attribute tags in this role tag system include: skill level (elementary / intermediate / advanced), which is used to measure the agent's ability in specific tasks; professional field (for example, in the medical field, it is subdivided into diagnostic skills, internal medicine / surgery professional field tags, etc., and different fields can be further subdivided according to their own characteristics), which clarifies the professional direction of the agent's expertise; and available time tags, which are used to record the agent's interactive participation status in different time periods; Tag definitions are customized and segmented according to application scenarios to meet the precise needs of role descriptions in different scenarios.

[0025] Role dependency generation: An improved graph neural network algorithm is used to combine the task dependency graph and historical interaction data to generate role dependency relationships. When building a task dependency graph, clearly define the task input / output and logical relationships. For example, the output of Task A is the input of Task B, and Task C can only begin after both Task A and Task B are completed. At the same time, we analyze the frequency and effectiveness of collaboration in historical interaction data, and count indicators such as the number of collaborations between different roles in historical tasks and the quality of task completion. Role dependency relationships are generated by integrating factors such as task sequence, resource dependency (for example, some tasks require specific resources, and different roles have different resource usage), and information transfer (for example, the content and frequency of information that needs to be exchanged between roles during task execution), thereby more accurately describing the relationship and influence between roles.

[0026] Dynamic update of role labels: Role labels are dynamically updated through the incremental stochastic gradient descent algorithm; the incremental stochastic gradient descent algorithm can process newly generated interaction data in real time and quickly adjust role labels; At the same time, an evaluation mechanism will be established to verify the update effect and regularly evaluate the updated labels to ensure their accuracy and effectiveness; Evaluation indicators may include the consistency between labels and actual character performance, the degree of improvement in the agent interaction effect after label updates, etc.

[0027] S2, dynamic interaction rule design; Interaction rule configuration: Based on the role tag system, the rule engine is used to configure conditional logic interaction rules. The rule engine uses the Drools engine, which has powerful rule configuration and management capabilities and can achieve flexible conditional logic configuration; Through the Drools engine, different interaction rules can be set based on role labels (such as skill level, professional field, etc.) and other related conditions (such as task type, time requirements, etc.). For example, when a task requires the participation of roles with advanced skill levels, agents with corresponding labels are assigned first.

[0028] Dynamic adjustment of interaction rules: Dynamic adjustment of interaction rules through real-time monitoring system and stream processing technology; The real-time monitoring system collects task completion rate, resource utilization rate, and user satisfaction data; The task completion rate reflects the efficiency and quality of the agent's task execution; the resource utilization rate reflects the agent's utilization of system resources during the interaction process; and the user satisfaction directly reflects the user's recognition of the agent's interaction effect. Use a real-time stream processing framework (such as Apache Flink) to set up a sliding window to analyze the data stream. The size of the sliding window can be adjusted according to actual needs, for example, to 5 minutes or 10 minutes. When the collected data exceeds the preset threshold, the threshold warning and rule adjustment are triggered. For example, when the resource utilization rate is too high, the interaction rules are automatically adjusted to reduce the number of tasks executed simultaneously or optimize the task allocation strategy.

[0029] Optimizing rule execution efficiency: Using distributed computing and sparsification algorithms to optimize rule execution efficiency; Distributed computing utilizes multiple servers or computing nodes to process rule calculation tasks in parallel, accelerating rule execution. Sparsification algorithms optimize rules, removing redundant and unnecessary calculations, reducing the amount of computation required, and further improving rule execution efficiency. For example, when processing a large number of rules, sparsification algorithms can filter out the rules relevant to the current task for calculation, avoiding the need to traverse all rules.

[0030] S3, real-time perception and decision optimization; Environmental data collection and preprocessing: Deploy sensors to collect environmental data. Sensors can include temperature sensors, humidity sensors, cameras, etc. The appropriate sensor type should be selected based on the specific application scenario. The collected environmental data are preprocessed and then input into the dynamic decision-making model. The preprocessing process includes data cleaning (removing noise and outliers), data normalization (unifying the data into the same dimensional range) and other operations to improve the quality and availability of the data.

[0031] Decision model construction and training: Build decision models based on reinforcement learning or deep learning, and combine them with transfer learning for training; The dynamic decision model uses the PPO algorithm or CNN / RNN architecture. The PPO algorithm is an efficient reinforcement learning algorithm that can improve learning efficiency while ensuring training stability. CNN (convolutional neural network) is suitable for processing data with spatial structure, such as image data; RNN (recurrent neural network) is good at processing sequence data, such as time series data; Transfer learning is used to match the feature spaces of the source model and the target model, transferring the knowledge of the source model (a model that has been trained in a related field or task) to the target model (the model that needs to be trained for the current task), thereby accelerating the training process of the target model. At the same time, incremental learning is used to adapt to environmental changes. Incremental learning can continuously absorb new data and knowledge, continuously optimize the model, and enable the model to better adapt to the dynamically changing environment. The dynamic decision-making model takes the task status (such as task progress, task difficulty, etc.) and resource occupancy rate in the environmental data as input, and outputs the probability distribution of the agent's role switching. That is, based on the current environmental status, it predicts the probability of the agent switching to different roles, providing a decision-making basis for the agent's role switching.

[0032] Decision execution effect monitoring and optimization: Real-time monitoring of decision execution results and continuous model optimization. By monitoring the performance of the agent in the actual interaction process, such as task completion time, user feedback and other indicators, the effectiveness of the decision model is evaluated. When it is found that the decision-making model is not effective, the parameters or structure of the model are adjusted in time and the model is optimized to improve the interactive performance of the intelligent agent.

[0033] S4, integration of feedback and results; Update role labels and interaction rules: Collect interaction result data and update role labels and interaction rules through analysis; The interaction result data includes information such as task completion status, user evaluation, and collaboration effects between intelligent agents. By analyzing these data, problems and shortcomings in role labels and interaction rules are discovered. For example, the definition of some role labels is not accurate enough, or some interaction rules are not effective in actual applications. Based on the analysis results, the role labels and interaction rules are updated and optimized to improve the quality and efficiency of intelligent agent interaction.

[0034] Multimodal feedback report generation: Generate multimodal feedback reports, supporting multiple output methods such as email, voice, and visual interface; The multimodal feedback interface supports collaborative output of voice, email, and visual interfaces, providing an adaptation strategy that prioritizes voice on mobile devices and visualization on office devices. On the mobile side, users can quickly obtain feedback information through voice, which is convenient and quick; on the office side, users can view detailed feedback reports through a visual interface, which facilitates in-depth analysis and decision-making; Multimodal feedback reports can meet the needs of different users in different scenarios and improve user experience.

[0035] System iterative optimization: iteratively optimize the system based on user feedback; Collect user feedback on the system, including evaluation and suggestions on the interaction effects of intelligent agents and the form of feedback reports; Based on user feedback, we conduct targeted optimization and improvement on each module of the system to continuously enhance the system performance and user experience.

[0036] Dynamic updating of role labels also includes: using incremental learning algorithms to process real-time interaction data. The incremental learning algorithm can process newly generated interaction data in real time and quickly update role labels; Regularly evaluate label accuracy and stability indicators. The evaluation cycle can be set according to actual conditions, such as weekly or monthly; Evaluation metrics can include the consistency between labels and actual character performance, and the changes in labels over different time periods. When the task completion rate deviation exceeds 15%, label recalibration is triggered. The task completion rate deviation refers to the gap between the actual task completion rate and the expected task completion rate. When the gap exceeds 15%, it means that the current role label may no longer accurately reflect the actual ability of the intelligent agent and needs to be recalibrated to ensure the accuracy and effectiveness of the role label.

[0037] Interaction rule optimization also includes: conditional logic configuration through the Drools engine, which can be flexibly configured and managed according to different conditions and rules; The Kafka distributed scheduling framework is used to parallelize rule calculations. The Kafka distributed scheduling framework features high throughput and low latency, enabling rapid processing of large numbers of rule calculation tasks and improving the efficiency of rule execution. The effectiveness of rules is verified through both simulation and real-world scenario testing. In simulation testing, a virtual interaction scenario is constructed to test and verify the rules. A rule is considered to have performed well in the simulation scenario when the rule execution success rate is ≥95%. At the same time, in the actual scenario test, user feedback and actual task completion status are collected. When the user satisfaction rate of the actual scenario test is ≥90%, the judgment rule is effective in actual application; Only when the rules meet the corresponding indicator requirements in both simulation tests and actual scenario tests, can the rules be considered valid and can be applied in the actual intelligent agent interaction process.

[0038] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The method of dynamic interaction between intelligent agents based on role labels is characterized by: The following steps are involved: S1, role modeling and label assignment; Through cross-disciplinary expert workshops and historical data analysis, a multi-dimensional role labeling system was defined, including basic role names and attribute labels; An improved graph neural network algorithm is used to generate role dependency relationships by combining task dependency graphs and historical interaction data. Dynamically update role labels using an incremental stochastic gradient descent algorithm, and establish an evaluation mechanism to verify the update effect; S2, dynamic interaction rule design; Based on the role tag system, use the rule engine to configure conditional logic interaction rules; Dynamically adjust interaction rules through real-time monitoring systems and stream processing technology; Use distributed computing and sparsification algorithms to optimize rule execution efficiency; S3, real-time perception and decision optimization; Deploy sensors to collect environmental data, which is then pre-processed and fed into a dynamic decision-making model; Build decision-making models based on reinforcement learning or deep learning, and combine them with transfer learning for training Monitor decision execution results in real time and continuously optimize the model; S4, integration of feedback and results; Collect interaction result data and update role labels and interaction rules through analysis; Generate multimodal feedback reports, supporting multiple output methods such as email, voice, and visual interface; Iteratively optimize the system based on user feedback.

2. The method according to claim 1, characterized in that The attribute tags of the role tag system include: skill level (elementary / intermediate / advanced), professional field and available time tags. The tag definitions are customized and segmented according to the application scenario. For example, the medical field is segmented into diagnostic skills and internal medicine / surgery professional field tags.

3. The method according to claim 1, characterized in that The improved graph neural network algorithm includes: clarifying task input / output and logical relationships when constructing a task dependency graph; Analyze the frequency and effectiveness of collaboration in historical interaction data; The role dependency relationship is generated by integrating task sequence, resource dependence, and information transmission factors.

4. The method according to claim 1, wherein The real-time monitoring system: Collect task completion rate, resource utilization rate, and user satisfaction data; Use the real-time stream processing framework to set up sliding windows to analyze data streams, trigger threshold warnings, and adjust rules.

5. The method according to claim 1, wherein The dynamic decision model adopts the PPO algorithm or CNN / RNN architecture, matches the feature space of the source model and the target model through transfer learning, and uses incremental learning to adapt to environmental changes. The dynamic decision model takes the task status and resource occupancy rate in the environmental data as input and outputs the probability distribution of the intelligent agent role switching.

6. The method according to claim 1, wherein The multimodal feedback interface supports collaborative output of voice, email, and visual interface, and provides an adaptation strategy that prioritizes voice on the mobile side and visualization on the office side.

7. The method according to claim 1, characterized in that The interaction rule optimization includes: Implement conditional logic configuration through the Drools engine; Use Kafka distributed scheduling framework to parallelize rule calculations; The effectiveness of the rules is double-verified based on simulation testing and actual scenario testing. When the rule execution success rate in the simulation test is ≥95% and the user satisfaction rate in the actual scenario test is ≥90%, the rule is judged to be valid.

8. The method according to claim 1, characterized in that The dynamic update of the role tag includes: Use incremental learning algorithms to process real-time interactive data; Regularly evaluate label accuracy and stability indicators; When the task completion rate deviation exceeds 15%, the label recalibration is triggered.

Images