Intelligent optimization method and device for weak body supervision, electronic equipment and storage medium

Abstract

This application relates to the field of artificial intelligence technology and can be applied to the fields of fintech and digital healthcare. It discloses a method, device, electronic device, and storage medium for intelligent optimization of supervision of individuals with weak constitutions. The method includes: fusing and processing multi-source, multi-modal data from preset channels, including behavioral, business, and external relational data of the target individual with weak constitution supervision. The fused data is then input into a weak constitution diagnostic algorithm model, where feature extraction, factor weight calculation, and outlier detection are performed to obtain negative factors and quantitative assessment results. Based on these results, a personalized supervision strategy containing training content, intensity, and scenario configuration is generated through knowledge graph matching. Subsequently, a virtual training environment is constructed for immersive training, collecting biosignal and behavioral data in real time and dynamically adjusting the environment. Finally, a difference-in-differences model is used to evaluate the training results, iteratively optimizing the diagnostic model and supervision strategy. This method improves the accuracy, relevance, and effectiveness of supervision of individuals with weak constitutions.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and can be applied to the fields of fintech / digital healthcare, particularly to a method, device, electronic device, and storage medium for weak body-supervised intelligent optimization. Background Technology

[0002] In the context of applying artificial intelligence technology to industry training, some application areas (such as finance and healthcare) still suffer from technical shortcomings and process defects in the training of supervisors with weak capabilities. For example, in the financial sector, the insurance industry's training of renewal supervisors with weak capabilities (policy management teams with low renewal rates and high surrender rates) has long relied on the subjective experience of supervisors to develop plans, lacking data-driven and precise diagnosis. Existing processes rely solely on basic policy data recorded in customer relationship management systems, making it impossible to pinpoint individual capability gaps. Furthermore, the use of generic courses without differentiating the causes of weaknesses results in limited improvement in core business indicators after training. At the same time, the evaluation of training effectiveness depends on long-term business data, leading to significant feedback lag. Traditional training platforms only support video courseware delivery, and some related training systems are still based on static knowledge bases, unable to dynamically generate personalized case studies. In the healthcare field, training for vulnerable primary care teams (such as those with insufficient chronic disease management capabilities) faces similar challenges: training programs are often based on general clinical guidelines, failing to incorporate data on primary care physicians' clinical practices and patient feedback, resulting in a disconnect between training content and actual needs; effectiveness evaluation relies on long-term assessment data, hindering real-time strategy adjustments, and lacks immersive training methods to simulate complex case emergency scenarios, leading to low training conversion rates and slow improvement in primary care service capabilities. Overall, existing technologies suffer from strong reliance on experience, severe homogenization, delayed feedback, resource misallocation, and technological limitations, failing to meet the industry's need for precise empowerment and efficient improvement of service capabilities for vulnerable groups. Summary of the Invention

[0003] The main technical problem addressed by the implementation method of this application is that existing weak body supervision training relies on historical experience, is homogeneous, and has a lagging effect evaluation.

[0004] To address the aforementioned technical problems, the first technical solution adopted in this application is: providing a method for intelligent optimization of weak entity supervision, comprising: fusing and processing multi-source multimodal data acquired from preset channels, wherein the multi-source multimodal data includes behavioral data, business data, and external related data of the target weak entity supervision object; inputting the fused multi-source multimodal data into a preset weak entity diagnosis algorithm model, and performing intelligent diagnosis of the causes of weak entities through feature extraction, factor weight calculation, and outlier detection to obtain the weak entity negative factors and quantitative evaluation results corresponding to the target weak entity supervision object; and, based on the weak entity negative factors and the quantitative evaluation results, through... A pre-defined knowledge graph is used to generate a personalized supervision strategy corresponding to the target weak subject. The personalized supervision strategy includes training content data, training intensity data, and scenario configuration data. A virtual training environment is constructed, and immersive interactive supervision training is conducted according to the personalized supervision strategy. Biosignals and behavioral data of the target weak subject are collected in real time during the training process, and the virtual training environment is dynamically adjusted based on the biosignals and behavioral data. The results of the immersive interactive supervision training are evaluated using a dual-difference model, and the weak subject diagnosis algorithm model and the personalized supervision strategy are iteratively optimized based on the evaluation results.

[0005] Optionally, the step of fusing multi-source, multimodal data acquired from a preset channel includes: acquiring raw data in real time from the preset channel through a pre-built real-time data pipeline; performing speech-to-text processing on the speech data in the raw data, extracting related information and historical frequencies from the text, and determining the weight value of the related information through a word frequency and document distribution correlation calculation method; unifying data from different sources to the same time base using a time calibration algorithm, wherein the time calibration algorithm includes time zone deviation correction and transmission delay compensation logic; filtering outliers in the physiological signal data in the raw data based on statistical rules, removing interference data that exceeds the normal fluctuation range; and combining the processed structured data with unstructured data to generate a standardized fusion feature set.

[0006] Optionally, the step of inputting the fused multi-source multimodal data into a preset weakness diagnosis algorithm model and performing intelligent diagnosis of the causes of weakness through feature extraction, factor weight calculation, and outlier detection includes: performing feature analysis on the fused multi-source multimodal data to extract multi-level features associated with the weakness state, the multi-level features including skill performance layer feature vectors, psychological state layer feature vectors, and environmental factor layer feature vectors; calculating the influence weight of each level of feature vectors using a preset feature importance assessment algorithm, and filtering out feature vectors whose weight values ​​exceed a preset weight threshold range; performing attribution calculation on each level of feature vectors using a feature contribution analysis algorithm to obtain the attribution metric value of each level of feature vectors to the weakness state; and generating the weakness negative factor and the corresponding quantitative assessment result based on the ranking result of the attribution metric value and a preset feature attribute classification, the weakness negative factor including skill level factors, psychological level factors, and environmental level factors.

[0007] Optionally, the step of generating a personalized supervision strategy corresponding to the target weak subject through matching with a preset knowledge graph based on the weak negative factors and the quantitative evaluation results includes: calling a preset knowledge graph, which contains data on the correlation between weak factors and supervision schemes, as well as training content adaptation rules for different scenarios; matching the weak negative factors with factor categories in the knowledge graph to adapt to the corresponding benchmark supervision scheme; optimizing the parameters of the benchmark supervision scheme based on the quantitative evaluation results, including allocating the proportion of training content according to the weight of the weak negative factors and setting the training intensity gradient based on the quantitative value of the weak state; and generating a personalized supervision strategy including training content priority, training cycle, and scenario switching rules based on the feature data of the target weak subject and the adjusted benchmark supervision scheme.

[0008] Optionally, the steps of constructing a virtual training environment, conducting immersive interactive supervised training according to the personalized supervision strategy, collecting biosignals and behavioral data of the target weak supervisee in real time during the training process, and dynamically adjusting the virtual training environment based on the biosignals and behavioral data include: constructing a virtual training scene containing features of multiple types of interactive objects based on scene configuration data in the personalized supervision strategy, wherein the interactive object features match the object attributes in the actual business scenario; collecting physiological state signals and training operation behavior data of the target weak supervisee in real time during the execution of the immersive interactive supervised training; performing emotional stability analysis on the collected physiological state signals to obtain emotional stability analysis results; performing response effectiveness analysis and process integrity analysis on the collected training operation behavior data to obtain behavior data analysis results; and dynamically adjusting the scene parameters of the virtual training environment based on the emotional stability analysis results and the behavior data analysis results; wherein the adjustment of the scene parameters includes: reducing the interaction difficulty set in the virtual training environment if the fluctuation range of the emotional value exceeds the preset emotional stability value range; and increasing the guidance prompt data of the virtual training environment if the effectiveness value of the behavior response is lower than the preset behavior response effectiveness threshold.

[0009] Optionally, the step of evaluating the results of the immersive interactive supervision training using a difference-in-differences model, and iteratively optimizing the weakness diagnosis algorithm model and the personalized supervision strategy based on the evaluation results, includes: dividing the target weak supervision objects into a training group and a control group, wherein the training group performs the immersive interactive supervision training, and the control group maintains the original supervision mode training; calculating the difference in business indicators and the change in weakness status between the two groups before and after training using a difference-in-differences model, wherein the difference in business indicators includes the increase in business indicator data and the decrease in complaint volume, and the change in weakness status... The training effect level of the immersive interactive supervision training is determined based on the changes in the quantitative evaluation results of the negative factors of the weakness; if the training effect level is not within the preset training effect level range, a correlation analysis is performed based on the results of the biosignals, behavioral data and intelligent diagnosis of the causes of weakness during the training process; based on the analysis results of the correlation factors, the feature weight calculation logic of the weakness diagnosis algorithm model is adjusted, and the training content ratio and scene switching rules in the personalized supervision strategy are optimized to complete the iterative update of the weakness diagnosis algorithm model and the personalized supervision strategy.

[0010] Optionally, the step of fusing and processing the multi-source multimodal data already acquired from the preset channels further includes: after the iterative optimization of the weak body diagnosis algorithm model, updating the collection dimensions and filtering rules of the multi-source multimodal data according to the feature requirements of the iterative model, and adding data source collection items that match the feature requirements of the iterative model; using the same time verification algorithm and outlier filtering rules as the original data to perform data cleaning processing on the newly added data source collection items to maintain the same data format and data feature standards; and performing feature supplementation and fusion with the newly added data source collection items and the historical fusion feature set to generate an adaptive extended fusion feature set that can be accepted by the iterative weak body diagnosis algorithm model.

[0011] To address the aforementioned technical problems, the second technical solution adopted in this application is: providing a weak-body supervision intelligent optimization device, comprising: a data fusion processing module, used to fuse and process multi-source multimodal data acquired from preset channels, wherein the multi-source multimodal data includes behavioral data, business data, and external related data of the target weak-body supervision object; a weak-body cause diagnosis module, used to input the fused multi-source multimodal data into a preset weak-body diagnosis algorithm model, and perform intelligent diagnosis of weak-body causes through feature extraction, factor weight calculation, and outlier detection to obtain the weak-body negative factors and quantitative evaluation results corresponding to the target weak-body supervision object; and a supervision strategy generation module, used to generate a strategy based on the weak-body negative factors and the quantitative evaluation results. As a result, a personalized supervision strategy corresponding to the target weak subject is generated through a preset knowledge graph matching. The personalized supervision strategy includes training content data, training intensity data, and scenario configuration data. The virtual training module is used to construct a virtual training environment, conduct immersive interactive supervision training according to the personalized supervision strategy, collect biosignals and behavioral data of the target weak subject in real time during the training process, and dynamically adjust the virtual training environment based on the biosignals and behavioral data. The model and strategy optimization module is used to evaluate the results of the immersive interactive supervision training through a difference-in-differences model, and iteratively optimize the weak subject diagnosis algorithm model and the personalized supervision strategy based on the evaluation results.

[0012] To solve the above-mentioned technical problems, the third technical solution adopted in the embodiments of this application is: to provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the weak body supervised intelligent optimization method as described above.

[0013] To solve the above-mentioned technical problems, the fourth technical solution adopted in the embodiments of this application is: to provide a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by an electronic device, the electronic device performs the weak body supervised intelligent optimization method as described above.

[0014] Unlike related technologies, this application solves the problem of data heterogeneity by fusing multi-source and multimodal data, providing unified support for diagnosis; it achieves intelligent diagnosis by leveraging a weakness diagnosis algorithm model, accurately outputting negative factors and quantitative results of weaknesses, avoiding subjective bias; it generates personalized strategies based on knowledge graphs, including training content, intensity, and scenario configuration, overcoming the limitations of generalization; it constructs and dynamically adjusts a virtual training environment, enhancing training adaptability; and it evaluates results and optimizes models and strategies through a difference-in-differences model, ensuring that the effects are quantifiable and continuously improving, significantly enhancing the accuracy and effectiveness of weakness supervision. Attached Figure Description

[0015] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0016] Figure 1 This is a schematic diagram of the operating environment of the weak body supervised intelligent optimization method provided in the embodiments of this application.

[0017] Figure 2 This is a schematic diagram of the execution flow of the weak body supervision intelligent optimization method provided in the embodiments of this application.

[0018] Figure 3 This is a schematic diagram of the execution flow of intelligent diagnosis of the causes of weaknesses in the weak body supervision intelligent optimization method provided in the embodiments of this application.

[0019] Figure 4 This is a schematic diagram of the execution flow for generating personalized supervision strategies in the weak body supervision intelligent optimization method provided in the embodiments of this application.

[0020] Figure 5 This is a schematic diagram of the system structure of the weak body supervision intelligent optimization device provided in the embodiments of this application.

[0021] Figure 6 This is a schematic diagram of the hardware structure of an electronic device for implementing the weak body supervision intelligent optimization method provided in the embodiments of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Software tools, components, or servers not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0023] It should be noted that, unless otherwise specified, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device schematic diagram or the order in the flowchart.

[0024] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0025] To facilitate understanding of this embodiment, a detailed description of the weak body supervised intelligent optimization method disclosed in this application embodiment will be provided first. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a schematic diagram of the operating environment of the weak body supervised intelligent optimization method provided in the embodiments of this application, such as... Figure 1 As shown, the execution subject of the weak body supervised intelligent optimization method provided in this application embodiment is generally an electronic device with a certain computing power, such as a computer device. In some possible implementations, this weak body supervised intelligent optimization method can be implemented by the processor calling computer-readable instructions stored in the memory. Figure 1 The computer equipment mentioned can be a server. A server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. This can be understood as... Figure 1 The number of computer devices shown is merely illustrative and can be expanded in any number according to actual needs.

[0026] Please continue reading. Figure 2 , Figure 2 This is a schematic diagram of the execution flow of the weak body supervised intelligent optimization method provided in the embodiments of this application, such as... Figure 2 As shown, it includes the following steps: S1. Integrate and process multi-source, multi-modal data acquired from preset channels. The multi-source, multi-modal data includes behavioral data, business data, and external related data of the target weak supervision object.

[0027] Step S1 is the data foundation of the entire embodiment, and its core is to integrate multi-source, multi-modal data to provide unified data support for subsequent vulnerability diagnosis. This step explicitly defines the data as encompassing behavioral data, business data, and externally related data of the target vulnerability supervision object, covering three key dimensions: the object itself, core business, and external environment. Through fusion processing, it resolves the inconsistencies in data formats, dimensions, and time bases from different sources, avoiding deviations in subsequent model diagnosis, ensuring data quality, and laying the foundation for the accuracy of the entire technical solution.

[0028] As an optional implementation, the step of fusing and processing multi-source multimodal data obtained from a preset channel in step S1 above may specifically include the following steps S11 to S15.

[0029] S11. Obtain raw data in real time from preset channels through a pre-built real-time data pipeline.

[0030] Step S11 is the entry point for data fusion, focusing on addressing the real-time performance and stability of data acquisition. Through a pre-built real-time data pipeline (such as Apache Kafka), continuous, low-latency access to data from multiple preset channels is achieved, avoiding data lag caused by traditional batch acquisition. Raw data is acquired in real-time without excessive processing, fully preserving the original characteristics of each channel's data, such as original voice recordings and operation logs, providing complete material for subsequent targeted processing and ensuring the integrity and authenticity of the data source.

[0031] S12. Perform speech-to-text processing on the speech data in the original data, extract the related information and historical frequency in the text, and determine the weight value of the related information by the word frequency and document distribution correlation calculation method.

[0032] Step S12 involves processing unstructured speech data, aiming to convert the speech signal into quantifiable text features. First, speech-to-text technology is used to convert the recorded conversation between the supervisor and the client into structured text. Then, key information related to the supervisor's weaknesses is extracted from the text, and the frequency of this information is calculated. Finally, by correlating word frequency with document distribution, and combining the frequency of information occurrence in the current text with its scarcity across all texts, the weights of different information types are determined, providing accurate text features for subsequent weakness diagnosis.

[0033] S13. A time calibration algorithm is used to unify data from different sources to the same time base. The time calibration algorithm includes time zone deviation correction and transmission delay compensation logic.

[0034] Step S13 addresses the issue of inconsistent time between multiple data sources, ensuring spatiotemporal alignment of the data. Data from different channels may have different timezone settings, and transmission may be delayed due to network fluctuations. Without calibration, data misalignment can occur, affecting the accuracy of the analysis. The time calibration algorithm corrects timezone deviations by converting all data to the same timezone; simultaneously, it compensates for transmission delays by correcting the timestamps of real-time transmitted data based on historical delay data, ultimately aligning all source data to the same time reference.

[0035] S14. Based on statistical rules, filter outliers in the physiological signal data in the original data and remove interfering data that exceeds the normal fluctuation range.

[0036] Step S14 optimizes the quality of physiological signal data by removing noisy data. Physiological signal data is susceptible to equipment errors and sudden interference, producing outliers that exceed the normal physiological range. Retaining these outliers would interfere with subsequent emotion analysis and psychological state diagnosis. Based on statistical rules, the 3σ principle is typically used. Data exceeding the mean ± 3 times the standard deviation is identified as outliers and removed, ensuring that the physiological signal data entering the subsequent fusion stage conforms to real physiological patterns and guarantees the reliability of subsequent feature extraction.

[0037] S15. Combine the processed structured data with unstructured data to generate a standardized fusion feature set.

[0038] Step S15 is the final integration stage of data fusion, aiming to form a feature set in a unified format to connect with the subsequent vulnerability diagnosis model. After preliminary processing, the data is divided into structured and unstructured data. Unstructured data is transformed into quantifiable features through feature encoding, and then features of different magnitudes are scaled to the same numerical range through feature normalization to avoid magnitude suppression. Finally, a standardized fusion feature set is generated, covering multi-dimensional features, with a unified format that can be directly input into the algorithm model, building a crucial bridge between data processing and vulnerability diagnosis.

[0039] As an example, in the scenario of insurance supervisory weakness, the system first obtains raw information such as policy processing logs, original call recordings with customers, and customer cancellation / renewal application data from the insurance company's core business system, supervisor call recording equipment, and customer feedback platform through real-time data pipelines. Next, after converting the call recordings into text, keywords related to customer objections such as "low returns" and "slow claims processing" are extracted. The frequency of these words is counted, and the weight of high-frequency keywords like "slow claims processing," which appear across multiple calls, is determined by correlating word frequency with document distribution, highlighting the core customer concerns. Then, the time base of each channel is standardized; for example, the UTC time of the customer feedback platform and the Beijing time of the local business system are calibrated to the same time zone to compensate for the 2-second delay in recording data transmission. Abnormal values ​​in the supervisor's heart rate monitoring data are then filtered out, such as a momentary heart rate of 200 beats / minute due to equipment detachment. Finally, the structured policy processing efficiency data, text keyword weight data, and calibrated heart rate data are combined to generate a standardized fusion feature set, providing data support for diagnosing the causes of supervisory weakness (such as insufficient ability to handle customer objections).

[0040] As another example, in the scenario of supervising the physical weakness of medical and nursing staff, information such as nursing staff operation records, raw patient vital signs data, and nursing quality scores are first obtained through a real-time data pipeline from nursing operation terminals, patient monitoring equipment, and nursing assessment systems. Then, the recorded communication between nursing staff and patients' families is transcribed into text, and relevant information such as "medication dosage" and "rehabilitation guidance" is extracted and weighted. Simultaneously, the time of each device is calibrated to unify the timestamps of monitoring devices and operation terminals, compensating for data transmission delays. Abnormal blood pressure values ​​monitored by nursing staff wristbands are then filtered out. Finally, the processed nursing operation data, text features, and physiological data are combined to generate a fusion feature set, which helps diagnose the physical weakness of nursing staff (such as insufficient operational proficiency).

[0041] Through steps S11 to S15, standardized processing of multi-source, multi-modal data from acquisition to fusion can be achieved, effectively solving problems such as heterogeneous data formats, time misalignment, and noise interference from different channels. On the one hand, the combination of real-time data acquisition and specialized processing not only preserves the integrity of the original data but also transforms unstructured speech into quantifiable features and highlights key information through weight calculation. On the other hand, time calibration and outlier filtering ensure the spatiotemporal consistency and authenticity of the data. The final fused feature set covers multi-dimensional key information and can directly provide high-quality input for subsequent weak body diagnosis algorithm models, significantly reducing errors in the data preprocessing stage and laying a reliable data foundation for accurately locating the causes of weak bodies.

[0042] S2. Input the fused multi-source multimodal data into the preset weak body diagnosis algorithm model. Through feature extraction, factor weight calculation and outlier detection, perform intelligent diagnosis of the causes of weak bodies to obtain the negative factors and quantitative evaluation results of the target weak body supervision object.

[0043] Step S2 is the core diagnostic step of the weak-body supervision intelligent optimization method. Its core function is to transform the standardized fusion data obtained in the early stage into conclusions about the causes of weaknesses that can guide subsequent strategies. This step uses a pre-set weak-body diagnosis algorithm model to connect three key operations: feature extraction, factor weight calculation, and outlier detection. Feature extraction filters key information related to the weak body status from the fusion data; factor weight calculation quantifies the degree of influence of different information on the weak body status; and outlier detection eliminates the bias of interfering data on the diagnostic results. The final output of the negative weak body factor clarifies the specific type of weak body problem, and the quantitative evaluation result provides a numerical basis for the severity of the problem. Together, they constitute the core basis for generating personalized supervision strategies and are the key link between data processing and strategy implementation.

[0044] As an alternative implementation method, please continue reading. Figure 3 , Figure 3 This is a schematic diagram illustrating the execution flow of intelligent diagnosis of the causes of weaknesses in the weak-body supervised intelligent optimization method provided in this application embodiment, as shown below. Figure 3 As shown, the process of intelligent diagnosis of the causes of weakness can specifically include the following steps S21 to S24.

[0045] S21. Perform feature analysis on the fused multi-source multimodal data to extract multi-level features associated with the weak state. The multi-level features include feature vectors of skill performance layer, feature vectors of psychological state layer, and feature vectors of environmental factors layer.

[0046] Step S21 is the basic feature screening operation in the diagnostic process. Its core objective is to extract structured, multi-level weakness-related features from the fused data. This step uses feature parsing technology to decompose the fused multi-source data into three types of feature vectors directly related to the weakness state: skill performance layer feature vectors correspond to the supervisor's business capability data, such as customer objection response accuracy and business process completion efficiency; psychological state layer feature vectors relate to the supervisor's emotions and stress resistance, such as heart rate variability and voice emotional stability; and environmental factors layer feature vectors cover external influencing conditions, such as customer age distribution and regional business competition intensity. This hierarchical extraction method ensures accurate correspondence between features and the causes of weaknesses and lays a structured data foundation for subsequent targeted attribution calculations, avoiding diagnostic biases caused by feature confusion.

[0047] S22. Calculate the influence weight of feature vectors at each level using a preset feature importance evaluation algorithm, and select feature vectors whose weight values ​​exceed the preset weight threshold range.

[0048] Step S22 quantifies the influence of features, with the core being the selection of key features that dominate the weak state using an algorithm. This step employs a pre-defined feature importance assessment algorithm, such as ensemble learning algorithms like Random Forest or XGBoost, to calculate the influence weights of the three types of feature vectors extracted in step S21. Higher weight values ​​indicate a stronger driving effect of the feature on the weak state. Subsequently, by using a pre-defined weight threshold range, secondary features with lower weights are filtered out, retaining only the core features for subsequent stages. For example, if the weight of "customer objection response accuracy" far exceeds the threshold, while the weight of "regional weather data" is below the threshold, the former is retained as a core feature, while the latter is discarded. This ensures that subsequent diagnosis focuses on key influencing factors, improving diagnostic efficiency and accuracy.

[0049] S23. Use the feature contribution analysis algorithm to perform attribution calculation on the feature vectors of each level, and obtain the attribution metric value of the feature vectors of each level to the weak state.

[0050] Step S23 involves clarifying the causal relationship between each core feature and the weak state. The core of this step is to quantify the contribution of individual features to the weak state through attribution calculation. This step employs feature contribution analysis algorithms, such as SHAP value analysis based on game theory and the LIME locally interpretable model, to calculate the attribution metric value for each of the core feature vectors selected in step S22. A positive attribution metric value, and the larger the value, the stronger the feature's driving effect on the weak state; a negative value indicates that the feature has a positive effect on improving the weak state. For example, the high attribution metric value of the "high anxiety" feature indicates that it is an important factor leading to the weak state; the negative attribution metric value of "business knowledge proficiency" indicates that improving this feature can alleviate the weak state problem. This result provides direct evidence for subsequently identifying the specific causes of the weak state.

[0051] S24. Based on the ranking results of the attribution metric and the preset feature attribute classification, generate the weakness negative factor and the corresponding quantitative evaluation result. The weakness negative factor includes skill level factor, psychological level factor and environmental level factor.

[0052] Step S24 is the output operation of the diagnostic process, the core of which is to transform the attribution calculation results into a structured diagnostic conclusion of weakness. This step first sorts the attribution metrics obtained in step S23 to clarify the priority of each core feature's impact on the weakness status; then, combined with preset feature attribute classification rules, it maps the core features to specific negative factor types of weakness: skill performance features correspond to skill-level factors, such as "weak business knowledge"; psychological state features correspond to psychological-level factors, such as "high anxiety avoidance of negotiation"; and environmental factor features correspond to environmental-level factors, such as "lack of trust among elderly customers." Simultaneously, the attribution metrics are transformed into quantitative assessment results, such as "the contribution of high anxiety to the weakness status is 0.72," ultimately forming a complete diagnostic conclusion of "factor type + quantification level," providing a clear basis for subsequent generation of personalized supervision strategies.

[0053] As an example, in the scenario of insurance vulnerability supervision, the first step is to perform feature analysis on the fused multi-source data, extracting three types of key feature vectors. The skill performance layer feature vector includes data such as the supervisor's policy processing efficiency and customer objection response accuracy; the psychological state layer feature vector covers indicators such as heart rate fluctuations and emotional stability of the supervisor's voice during communication with customers; and the environmental factors layer feature vector includes information such as the age distribution of customers in the supervisor's area of ​​responsibility and the promotional efforts of competing insurance products. Next, a feature importance assessment algorithm is used to calculate the weight of each feature, filtering out core features with weights exceeding the threshold, such as customer objection response accuracy and high heart rate fluctuations. Then, a feature contribution analysis algorithm is used to calculate the attribution metric values ​​of these core features, revealing that the attribution metric value of high heart rate fluctuations is significantly higher than other features. Finally, combined with feature attribute classification, high heart rate fluctuations are categorized as a psychological factor, and customer objection response accuracy is categorized as a skill-level factor, generating quantitative assessment results to clarify the contribution of high heart rate fluctuations to the vulnerability status, providing a basis for subsequently developing targeted supervision strategies.

[0054] As another example, in the scenario of monitoring the physical weakness of medical and nursing staff, multi-level feature vectors are first extracted from the fused data. The skill performance layer includes nursing operation compliance rate and emergency response speed; the psychological state layer covers blood pressure changes and emotional stability during nursing care; and the environmental factors layer includes ward patient traffic and equipment usage frequency. Next, the weights of each feature are calculated to select core features such as nursing operation compliance rate and blood pressure changes. Then, the attribution measures of these features are obtained through algorithms. Finally, combined with classification rules, nursing operation compliance rate is classified as a skill-level factor, and blood pressure changes are classified as a psychological-level factor, generating quantitative assessment results and clarifying the degree of influence of each factor on the physical weakness status.

[0055] Through steps S21 to S24, precise diagnosis and structured attribution of weaknesses can be achieved, effectively addressing the problems of "vague causes and subjective assessments" in traditional supervision. On one hand, hierarchical feature extraction ensures comprehensive coverage of information related to weaknesses, avoiding the omission of key influencing factors. On the other hand, feature weight calculation and contribution analysis quantify the impact of each factor on the weakness status, identifying core driving factors and eliminating interference from secondary information. The resulting negative factors clearly define the problem type, and the quantitative assessment results provide objective numerical evidence. The combination of these two aspects transforms the causes of weaknesses from abstract descriptions into precise, quantifiable, and categorizable conclusions, providing clear and reliable decision support for the subsequent generation of personalized supervision strategies, significantly improving the adaptability and effectiveness of the supervision plan.

[0056] S3. Based on the negative factors of the weak subjects and the quantitative assessment results, a personalized supervision strategy is generated for the target weak subject through a pre-set knowledge graph matching. The personalized supervision strategy includes training content data, training intensity data and scenario configuration data.

[0057] Step S3 is a crucial link connecting the diagnosis of weaknesses with the actual implementation of supervision. Its core is transforming the diagnostic results into a practical, personalized supervision plan. This step takes the negative factors of weaknesses and quantitative assessment results as input, relying on the association rules of a pre-set knowledge graph to achieve the transformation from "problem diagnosis" to "solution." The generated personalized supervision strategy includes three core data categories: training content, training intensity, and scenario configuration. This ensures both a precise match between the strategy and the causes of weaknesses and avoids the inefficiency of a "one-size-fits-all" approach to supervision through personalized design. It provides a clear execution framework for subsequent immersive virtual training, ensuring that supervisory resources are focused on key areas for improvement.

[0058] As an alternative implementation method, please continue reading. Figure 4 , Figure 4 This is a schematic diagram illustrating the execution flow of generating personalized supervision strategies in the weak body supervision intelligent optimization method provided in this application embodiment, such as... Figure 4 As shown, the process of generating a personalized supervision strategy can specifically include the following steps S31 to S34.

[0059] S31. Call the preset knowledge graph, which contains the correlation data between weak factors and supervision schemes, as well as training content adaptation rules for different scenarios.

[0060] Step S31 is the knowledge support stage for strategy generation, the core of which is to call upon a pre-defined knowledge graph to provide decision-making basis. The knowledge graph stores two types of key information: first, the relationship between weaknesses and supervision solutions, such as "weak customer objection handling ability" corresponding to "communication skills training"; second, scenario adaptation rules, such as "elderly customer group" adapting to "simplified process simulation scenario". This information is built based on historical supervision cases and domain expert experience, forming a structured knowledge network that can quickly match potential solutions for current weaknesses, avoiding the blindness of strategy generation and laying the knowledge foundation for subsequent accurate matching of benchmark solutions.

[0061] S32. Match the weak negative factors with the factor categories in the knowledge graph to adapt them to the corresponding benchmark supervision scheme.

[0062] Step S32 is the initial matching stage of strategy generation. Its core is aligning the diagnosed weaknesses with standard factor categories in the knowledge graph to determine the basic supervision framework. For example, if the skill-level factor "low proficiency in claims process" is diagnosed, the system will retrieve a benchmark solution corresponding to the same factor in the knowledge graph, such as "claims process simulation training + case analysis." Through factor category matching, the system ensures that the benchmark solution retrieved from the knowledge graph is directly relevant to the current weakness, avoiding mismatch between the solution and the problem, and providing a reliable basic template for subsequent parameter optimization.

[0063] S33. Optimize the parameters of the benchmark supervision scheme based on the quantitative evaluation results. The optimized parameters include allocating the proportion of training content according to the weight of the negative factors of the weak body, and setting the training intensity gradient based on the quantitative value of the weak body status.

[0064] Step S33 is the core of personalized strategy adjustment, which involves optimizing the parameters of the baseline plan based on quantitative assessment results. The proportion of training content is allocated according to the weight of the negative factors affecting the individual's condition. For example, if the weight of "communication anxiety" is higher than that of "process proficiency," then the proportion of psychological counseling content is increased to 60%. A training intensity gradient is set based on the quantitative value of the individual's condition. For instance, if the quantitative value is 0.8 (out of 1), the training intensity is set to high, including high-frequency stress scenario simulations. Through parameter optimization, the supervision plan is upgraded from a general baseline template to a personalized framework tailored to the severity of the problems, improving the strategy's relevance.

[0065] S34. Based on the adjusted benchmark supervision scheme, and according to the characteristic data of the target weak supervision object, generate a personalized supervision strategy that includes training content priority, training cycle and scenario switching rules.

[0066] Step S34 is the final output of strategy generation, the core of which is to form a complete implementation plan based on the individual characteristics of the supervisee. Building upon the adjusted baseline plan, the priority of training content is determined according to the subject's age, learning pace, and other characteristic data; for example, beginners are prioritized to master the basic processes. Differentiated training cycles are set, such as experienced supervisors shortening the cycle to two weeks. Scenario switching rules are established, such as increasing scenario complexity after three consecutive successful completions. The final generated strategy includes specific implementation details, maintaining the accuracy of the initial diagnosis while also considering individual differences, ensuring that the supervision plan can be directly implemented.

[0067] As an example, in an insurance vulnerability supervision scenario, a pre-defined knowledge graph is first invoked. This graph stores relationships such as "weak handling of customer objections" corresponding to "communication skills training," as well as adaptation rules such as "simplified scenarios for elderly customers." Next, the diagnosed vulnerability factor of "low proficiency in claims procedures" is matched with factor categories in the knowledge graph and adapted to a baseline supervision plan of "claims procedure simulation + case analysis." Then, parameters are optimized based on quantitative evaluation results. If the quantitative value of "claims procedure proficiency" is 0.3 (out of 1) and has a high weight, the proportion of process simulation content is increased to 70%, and the training intensity is set to beginner level. Finally, considering the characteristics of the supervisee, newcomers are given priority in learning the basic process, with a training cycle of 3 weeks. A switching rule of "increasing scenario complexity after two consecutive successful operations" is established, generating a complete personalized supervision strategy.

[0068] As another example, in the scenario of supervising healthcare staff with weaker skills, a knowledge graph is first invoked, which includes the correlation between "low operational compliance rate" and "compliant operation training," as well as rules for "adapting high-intensity scenarios to intensive care units." The "slow emergency response" factor is matched to the baseline scheme of "emergency procedure drills," and parameters are optimized based on a quantification value of 0.4, increasing the proportion of emergency drills to 65% and setting the intensity to medium. Furthermore, considering the nursing staff's years of service, with a 4-week cycle for new nurses, a rule of "increasing scenario difficulty after 3 successful completions" is formulated to generate personalized strategies.

[0069] Through steps S31 to S34, the diagnostic results of weaknesses can be efficiently transformed into personalized supervision strategies tailored to actual needs, addressing the problem of traditional supervision solutions being "highly generalizable but poorly adaptable." On one hand, relying on the association rules and scenario adaptation logic of the knowledge graph ensures that the generated baseline solution directly corresponds to the causes of weaknesses, preventing the strategy from deviating from the core issue. On the other hand, by optimizing parameters through quantitative evaluation results and refining execution details based on object characteristics, the strategy matches both the severity of the problem and individual differences. The final output strategy encompasses key information such as training content, cycle, and scenario rules, directly guiding subsequent training execution, significantly improving the utilization efficiency of supervision resources, and providing a clear and implementable action framework for accurately improving the state of weaknesses.

[0070] S4. Construct a virtual training environment, conduct immersive interactive supervision training based on personalized supervision strategies, collect biosignals and behavioral data of the target weak subject in real time during the training process, and dynamically adjust the virtual training environment based on the biosignals and behavioral data.

[0071] Step S4 is the execution phase of transforming personalized supervision strategies into practical applications. Its core is immersive supervision training and dynamic optimization through a virtual environment. This step first constructs a virtual scenario matching actual business needs based on the strategy. During training, physiological and behavioral data are collected in real time. Scenario parameters are adjusted by analyzing the data, forming a closed loop of "training-feedback-optimization." This dynamic adjustment mechanism ensures that the training difficulty matches the state of the supervised individual, avoiding inefficient training or frustration caused by fixed scenarios. Simultaneously, immersive interaction enhances the training experience, making the supervision results closer to actual business scenarios and providing real improvement data for subsequent evaluation.

[0072] As an optional implementation, step S4 may specifically include steps S41 to S45.

[0073] S41. Based on the scenario configuration data in the personalized supervision strategy, construct a virtual training scenario containing features of multiple types of interactive objects, wherein the interactive object features match the object attributes in the actual business scenario.

[0074] Step S41 is the scenario construction stage for virtual training. Its core is to build a highly realistic training environment based on the scenario configuration data in the personalized strategy. The virtual scenario includes multiple types of interactive objects, whose characteristics are consistent with the object attributes in actual business. For example, the "virtual avatar of elderly customers" in the insurance scenario needs to match the communication habits of real elderly people, and the "virtual model of emergency equipment" in the medical scenario needs to replicate the operating logic of the actual equipment. This high degree of matching design ensures that the training scenario is highly similar to the real business environment, allowing the experience gained by the supervised trainees to be directly transferred to actual work, thus enhancing the practical value of the training.

[0075] S42. During the execution of the immersive interactive supervised training, the physiological state signals and training operation behavior data of the target weak supervised object are collected in real time.

[0076] Step S42 is the data acquisition stage of the training process, the core of which is to acquire key data reflecting the state of the supervised individual in real time. Physiological state signals include heart rate, blood pressure, and skin conductance, reflecting emotional fluctuations and stress levels; training operational behavior data covers operational steps, response times, decision paths, etc., reflecting the proficiency and accuracy of business skills. Real-time linkage between wearable devices (such as smart bracelets and motion capture devices) and the virtual environment ensures the continuity and timeliness of data acquisition, providing first-hand data support for subsequent state analysis and scenario adjustments, which is a prerequisite for achieving dynamic optimization.

[0077] S43. Perform emotional stability analysis on the collected physiological state signals to obtain the emotional stability analysis results.

[0078] Step S43 involves analyzing physiological data, the core of which is assessing the emotional stability of the supervised individual through physiological signals. Based on collected data such as heart rate variability and respiratory rate, emotion recognition algorithms (e.g., anxiety detection models based on heart rate variability) are used to analyze emotional fluctuation trends. For example, a sudden and frequent increase in heart rate may indicate anxiety, while a stable heart rate suggests emotional stability. The analysis results provide a basis for quantifying the adjustment of the difficulty of the virtual scenario, ensuring that the training intensity is appropriate for the supervised individual's psychological capacity and avoiding excessive pressure that could negatively impact training effectiveness.

[0079] S44. Perform response effectiveness analysis and process integrity analysis on the collected training operation behavior data to obtain the behavior data analysis results.

[0080] Step S44 involves analyzing behavioral data, with the core objective of evaluating the quality of the supervised participant's actions during training. Response effectiveness analysis focuses on whether the action achieved the objective, such as "whether customer objections were successfully resolved and the customer was appeased" in an insurance scenario. Process integrity analysis focuses on whether the operational steps conform to standard specifications, such as "whether key steps were omitted in the emergency procedure" in a medical scenario. By comparing behavioral data with preset standards, effectiveness and integrity scores are quantified, providing an objective basis for identifying weaknesses in business capabilities and adjusting guidance prompts, ensuring that training focuses on addressing specific weaknesses.

[0081] S45. Based on the emotional stability analysis results and the behavioral data analysis results, dynamically adjust the scene parameters of the virtual training environment; wherein, the adjustment of the scene parameters includes: if the fluctuation range of the emotional value exceeds the preset emotional stability value range, then reduce the interaction difficulty already set in the virtual training environment; if the validity value of the behavioral response is lower than the preset behavioral response validity threshold, then increase the guidance prompt data of the virtual training environment.

[0082] Step S45 is the dynamic adjustment step of the virtual environment, the core of which is to optimize the real-time scene parameters based on the state analysis results. If emotional fluctuations exceed the stable range, the system automatically reduces the difficulty of interaction, such as simplifying the content of customer objections; if the effectiveness of behavioral responses is below the threshold, guidance prompts are added, such as highlighting key steps. This adjustment mechanism ensures that the virtual environment always matches the real-time state and ability level of the supervisee, avoiding frustration and abandonment due to excessive difficulty, and preventing ineffective training due to insufficient prompts, thus maximizing training efficiency and effectiveness.

[0083] As an example, in the scenario of insurance supervision for vulnerable clients, a training scenario is first constructed based on the scenario configuration data of the personalized supervision strategy. This scenario includes virtual avatars of different types of clients, such as simulating the communication habits of elderly clients and the insurance needs of middle-aged clients. The characteristics of the interaction objects are fully matched with the attributes of clients in actual business. After training begins, physiological signals such as heart rate and skin conductance are collected in real time by a smart bracelet. At the same time, behavioral data such as the supervisor's operation steps and response time in handling client objections are recorded. Then, the physiological signals are analyzed. If the heart rate is found to be consistently higher than the normal range and fluctuates frequently, it is determined that the emotional stability is insufficient. The behavioral data is analyzed. If the handling of client objections fails to achieve the goal of appeasement and there are omissions in the steps, it is determined that the effectiveness of the response and the completeness of the process are not up to standard. Finally, the scenario is adjusted according to the analysis results. When there are large emotional fluctuations, the content of the client objection is simplified. When the effectiveness is low, the correct handling steps are highlighted to ensure that the training is adapted to the real-time state of the supervisor.

[0084] As another example, in the scenario of physical supervision for healthcare workers, a training scenario with a virtual patient and emergency equipment was first constructed according to a strategy, with the equipment operation logic and patient symptoms consistent with reality. During training, physiological signals such as blood pressure and respiratory rate of the nurses were collected, as well as behavioral data such as emergency operation steps and decision-making time. Analysis revealed that a sudden rise in blood pressure indicated emotional instability, and omission of key steps indicated poor process completeness. Subsequently, the scenario was adjusted: the complexity of the patient's condition was reduced when there were emotional fluctuations, and operational guidance prompts were added when the process was incomplete, improving the training's relevance.

[0085] Through steps S41 to S45, a highly realistic virtual training environment can be constructed, enabling dynamically adapted immersive supervised training. On one hand, the highly matched virtual scenario ensures seamless integration between training and actual business operations, improving the efficiency of experience transfer. On the other hand, real-time collected physiological and behavioral data provides accurate evidence for state assessment. Combined with the analysis results, scenario parameters are dynamically adjusted to ensure that the training difficulty and guidance intensity are always adapted to the real-time state of the supervisee. This dynamic closed-loop mechanism avoids the inefficiency of fixed scenarios while maintaining the effectiveness and continuity of training through personalized adjustments. It significantly improves the supervisee's skill mastery speed and emotional adaptability, providing reliable improvement data for subsequent effect evaluation.

[0086] S5. The results of immersive interactive supervision training are evaluated using a dual difference model. Based on the evaluation results, the weak body diagnosis algorithm model and personalized supervision strategy are iteratively optimized.

[0087] Step S5 is the closed-loop optimization stage of the entire weak-body supervision intelligent optimization method. Its core is to evaluate training effectiveness through technology and iteratively upgrade the core model and strategy in reverse. This step uses a difference-in-differences model as the evaluation tool to quantify the actual value of immersive training. Simultaneously, it combines training process data to pinpoint the root causes of problems and adjust the weak-body diagnosis algorithm model and personalized supervision strategies accordingly. This closed-loop logic of "evaluation-analysis-optimization" continuously improves diagnostic accuracy and strategy adaptability, preventing the system from falling into "static inefficiency" and ensuring that the entire supervision system dynamically evolves with actual application scenarios, maintaining a high level of efficiency in improving weak-body performance in the long term.

[0088] As an optional implementation, step S5 may specifically include steps S51 to S55.

[0089] S51. Divide the target weak object supervision object into a training group and a control group, wherein the training group performs the immersive interactive supervision training, and the control group maintains the original supervision mode training.

[0090] Step S51 is a crucial operation in constructing the evaluation benchmark. Its core principle is to eliminate interfering factors through group design, ensuring the objectivity of the training effect evaluation. The target supervised subjects are divided into a training group and a control group. The training group uses immersive interactive supervised training, while the control group follows the original model. Both groups maintain consistency in core attributes such as business foundation and initial weaknesses. This grouping method effectively isolates the influence of irrelevant variables such as "time passage" and "external policy changes," for example, avoiding misjudging "overall industry business improvement" as the effect of immersive training. This lays the foundation for subsequent comparison by accurately calculating the net effect of training using a difference-in-differences model.

[0091] S52. Calculate the difference in business indicators and the change in the weak state of the two groups of objects before and after training using a difference-in-differences model. The difference in business indicators includes the increase in business indicator data and the decrease in the number of complaints. The change in the weak state is determined based on the change in the quantitative evaluation result of the negative factors of the weak state.

[0092] Step S52 is the core operation for quantifying the training effect. The core is to calculate the differences in key indicators between the two groups using a difference-in-differences model, thus separating the true impact of the training. This model compares the differences before and after training in the training group with those in the control group to derive the net effect of immersive training, specifically reflected in the difference values ​​of business indicators and the change values ​​of the weak state: the difference values ​​of business indicators reflect actual business results, such as a 5% higher increase in renewal rate and an 8% higher decrease in complaints in the training group compared to the control group; the change values ​​of the weak state are based on the quantitative assessment results of the negative factors of the weak state, such as a 0.3 decrease in the quantitative value of the "weak handling of customer objections" factor in the training group, while only a 0.1 decrease in the control group.

[0093] S53. Determine the training effectiveness level of immersive interactive supervision training based on the difference values ​​of the business indicators.

[0094] Step S53 involves determining the training effectiveness level, with the core being the establishment of standardized effectiveness evaluation criteria based on the difference values ​​of business indicators. The system presets a range for training effectiveness levels, such as "Excellent" corresponding to an improvement of over 10% in business indicators, "Pass" corresponding to an improvement of 5%-10%, and "Fail" corresponding to an improvement of less than 5%. The effectiveness level of the current immersive training is determined by comparing the difference values ​​of business indicators calculated in step S52 with the preset level range. For example, if the renewal rate of the training group is 12% higher than that of the control group, it is judged as "Excellent"; if it is only 3% higher, it is judged as "Fail".

[0095] S54. If the training effect level is not within the preset training effect level range, then a correlation analysis is performed based on the biosignals, behavioral data, and the results of intelligent diagnosis of the causes of weakness during the training process.

[0096] Step S54 involves identifying the reasons for poor training effectiveness, with the core being the use of correlation analysis to find key factors affecting training results. When the training effectiveness level falls below the preset range, the system retrieves biosignals and behavioral data collected during training, along with previous diagnostic results for weaknesses, and performs multi-dimensional correlation analysis. For example, if the system finds that the biosignals of "high anxiety" persist in the training group, and the corresponding behavioral data on "handling customer objections" has low validity, while the previous diagnosis of "psychological weakness factors" for this group has a high weight, it can be concluded that "insufficient psychological counseling training content has led to unresolved anxiety, thus affecting operational effectiveness," providing direction for subsequent optimization.

[0097] S55. Based on the analysis results of the related factors, adjust the feature weight calculation logic of the weakness diagnosis algorithm model, and optimize the training content allocation and scenario switching rules in the personalized supervision strategy to complete the iterative update of the weakness diagnosis algorithm model and the personalized supervision strategy.

[0098] Step S55 involves iterative optimization of the system, primarily adjusting the model and strategies based on the root cause analysis results to improve subsequent supervision effectiveness. Addressing the issues identified in step S54, optimizations are made in two aspects: For the weakness diagnosis algorithm model, the feature weight calculation logic is adjusted; for example, if the impact of "psychological state characteristics" on weakness diagnosis is underestimated, the weight of such characteristics is increased. For personalized supervision strategies, the training content allocation and scenario switching rules are optimized; for instance, the proportion of psychological counseling content is increased to 40%, and the scenario switching rule is adjusted to "increasing the difficulty only after three consecutive achievement of a stable emotional value." Through these two adjustments, the model and strategies are iteratively updated, ensuring that subsequent supervision can specifically address early-stage issues and improve overall optimization effectiveness.

[0099] As an example, in the scenario of insurance vulnerable supervision, 40 vulnerable supervision subjects were first divided into a training group and a control group, with both groups having similar initial business capabilities and vulnerable status. The training group adopted immersive interactive supervision training, while the control group followed the traditional offline lecture model. After the training period, calculations using a difference-in-differences model revealed that the policy renewal rate of the training group increased by 15%, while that of the control group increased by 6%, with a business indicator difference of 9%. The quantitative value of the "weak handling of customer objections" factor decreased by 0.4 in the training group and by 0.1 in the control group, showing a significant difference in the change value of vulnerable status. According to the preset standard, a 9% improvement in business indicators reached the "good" level. If the standard was not met, further analysis was conducted: it was found that subjects in the training group who frequently exhibited anxiety biological signals had lower validity in their customer communication behavior data. Combined with the result of "high weight of psychological factors" in the previous diagnosis, it was determined that the psychological counseling content was insufficient. Subsequently, the diagnostic model was adjusted to increase the weight of psychological state characteristics; the strategy was optimized by increasing the proportion of psychological counseling content from 20% to 35%, and the scenario switching rule was changed to "increasing the difficulty only after two days of stable emotions," completing the iterative optimization.

[0100] Through steps S51 to S55, a scientific evaluation and dynamic iteration of the immersive supervised training effect can be achieved. On the one hand, the combination of group comparison and the difference-in-differences model accurately isolates the true effect of training, avoids interference from external factors, and makes the effect evaluation more objective. On the other hand, for cases where the effect does not meet the standard, multi-dimensional data correlation analysis is used to locate the core problem, and then the diagnostic model and supervision strategy are adjusted accordingly. This mechanism ensures that the training effect is quantifiable and verifiable, and continuously improves the accuracy of diagnosis and the adaptability of strategies, so that the entire weak body supervision system can continuously evolve with actual application and maintain a high efficiency in improving weak bodies in the long term.

[0101] As a preferred embodiment, the above-mentioned steps for fusion processing of multi-source multimodal data obtained from preset channels may further include the following steps S61 to S63.

[0102] S61. After iterative optimization of the weak body diagnosis algorithm model, update the collection dimensions and filtering rules of multi-source multimodal data according to the feature requirements of the iterated model, and add data source collection items that match the feature requirements of the iterated model.

[0103] Step S61 is the basic data adaptation operation after model iteration. The core is to adjust the data collection dimensions and filtering rules according to the new feature requirements of the weak body diagnosis algorithm model after iteration. For example, when the model adds the analysis requirement of "customer communication tone features", it is necessary to collect tone and pitch data in supervisory calls, and at the same time update the filtering rules to ensure that the new data meets the model input standards, so as to provide a new data source for subsequent data processing.

[0104] S62. Use the same time verification algorithm and outlier filtering rules as the original data to perform data cleaning on the newly added data source collection items in order to maintain the same data format and data feature standards.

[0105] Step S62 is a crucial operation to ensure the quality and consistency of the newly added data. The new data source is processed using the same time verification algorithm and outlier filtering rules as the original data. This avoids inconsistencies in data format or feature standards due to differences in processing rules, ensuring seamless integration of the new data with historical data and not affecting the accuracy of subsequent fusion processing.

[0106] S63. The newly added data source collection items are combined with the historical fusion feature set to supplement and fuse features, so as to generate an adapted extended fusion feature set that can be accepted by the iterative weak body diagnosis algorithm model.

[0107] Step S63 is the final operation to complete data adaptation. Its core is to supplement and fuse the cleaned new data source with the historical fusion feature set. This generates an adapted extended fusion feature set through feature combination. This set directly meets the input requirements of the iteratively iterative weak body diagnostic algorithm model, ensuring that the model still has high-quality data support after iteration, maintaining the continuity and accuracy of the diagnostic function.

[0108] The weak component supervision intelligent optimization method provided in this application achieves efficient and accurate weak component optimization through multi-stage collaboration. At the data level, standardized fusion processing of multi-source, multi-modal data solves problems such as format heterogeneity and temporal misalignment, providing high-quality data support for diagnosis. In the diagnosis stage, hierarchical feature extraction and quantitative attribution accurately locate negative factors of weak components, avoiding the subjective ambiguity of traditional supervision. Strategy generation relies on knowledge graphs to achieve accurate matching between diagnostic results and solutions, taking into account both problem priority and individual differences. The dynamic adaptation mechanism in the virtual training stage enhances training immersion and efficiency, ensuring experience transferability. The evaluation and optimization closed loop continuously improves model diagnostic accuracy and strategy adaptability through scientific evaluation and iterative adjustments, significantly enhancing the intelligence and effectiveness of weak component supervision.

[0109] Please continue reading. Figure 5 , Figure 5 This is a schematic diagram of the system structure of the weak body-supervised intelligent optimization device provided in the embodiments of this application, as shown below. Figure 5 As shown, the weak body supervision intelligent optimization device 50 includes: a data fusion processing module 51, a weak body cause diagnosis module 52, a supervision strategy generation module 53, a virtual training module 54, and a model and strategy optimization module 55.

[0110] The data fusion processing module 51 is specifically used to fuse and process multi-source multimodal data obtained from preset channels. The multi-source multimodal data includes behavioral data, business data, and external related data of the target weak supervision object.

[0111] The weakness cause diagnosis module 52 is specifically used to input the fused multi-source multimodal data into the preset weakness diagnosis algorithm model, and to perform intelligent diagnosis of weakness causes through feature extraction, factor weight calculation and outlier detection, so as to obtain the weakness negative factors and quantitative evaluation results corresponding to the target weakness supervision object.

[0112] The supervision strategy generation module 53 is specifically used to generate a personalized supervision strategy corresponding to the target weak object through a preset knowledge graph matching based on the weak object's negative factors and the quantitative evaluation results. The personalized supervision strategy includes training content data, training intensity data, and scenario configuration data.

[0113] The virtual training module 54 is specifically used to construct a virtual training environment, conduct immersive interactive supervised training according to the personalized supervision strategy, collect biosignals and behavioral data of the target weak supervisee in real time during the training process, and dynamically adjust the virtual training environment based on the biosignals and behavioral data.

[0114] The model and strategy optimization module 55 is specifically used to evaluate the results of the immersive interactive supervision training through a dual difference model, and iteratively optimize the weak body diagnosis algorithm model and the personalized supervision strategy based on the evaluation results.

[0115] As an optional implementation, the data fusion processing module 51 is further configured to: acquire raw data in real time from the preset channel through a pre-built real-time data pipeline; perform speech-to-text processing on the speech data in the raw data, extract related information and historical frequency from the text, and determine the weight value of the related information through a word frequency and document distribution correlation calculation method; use a time calibration algorithm to unify data from different sources to the same time base, the time calibration algorithm including time zone deviation correction and transmission delay compensation logic; perform outlier filtering on the physiological signal data in the raw data based on statistical rules, removing interference data that exceeds the normal fluctuation range; and combine the processed structured data and unstructured data to generate a standardized fusion feature set.

[0116] As an optional implementation, the weakness cause diagnosis module 52 is further specifically used to perform feature analysis on the fused multi-source multimodal data, extract multi-level features associated with the weakness state, the multi-level features including skill performance layer feature vectors, psychological state layer feature vectors, and environmental factor layer feature vectors; calculate the influence weight of each level of feature vectors using a preset feature importance evaluation algorithm, and filter out feature vectors whose weight values ​​exceed a preset weight threshold range; use a feature contribution analysis algorithm to perform attribution calculation on each level of feature vectors to obtain the attribution metric value of each level of feature vectors to the weakness state; based on the ranking result of the attribution metric value and the preset feature attribute classification, generate the weakness negative factor and the corresponding quantitative evaluation result, the weakness negative factor including skill level factors, psychological level factors, and environmental level factors.

[0117] As an optional implementation, the supervision strategy generation module 53 is further configured to invoke a preset knowledge graph, which contains data on the correlation between weak factors and supervision schemes, as well as training content adaptation rules for different scenarios; match the weak negative factors with factor categories in the knowledge graph to adapt to the corresponding benchmark supervision scheme; optimize the parameters of the benchmark supervision scheme based on the quantitative evaluation results, including allocating the proportion of training content according to the weight of the weak negative factors and setting the training intensity gradient based on the quantitative value of the weak state; and generate a personalized supervision strategy containing training content priority, training cycle, and scenario switching rules based on the adjusted benchmark supervision scheme and the feature data of the target weak supervision object.

[0118] As an optional implementation, the virtual training module 54 is further configured to construct a virtual training scenario containing features of multiple types of interactive objects based on the scenario configuration data in the personalized supervision strategy, wherein the interactive object features match the object attributes in the actual business scenario; during the execution of the immersive interactive supervision training, the physiological state signals and training operation behavior data of the target weak supervisee are collected in real time; the collected physiological state signals are subjected to emotional stability analysis to obtain emotional stability analysis results; the collected training operation behavior data is subjected to response effectiveness analysis and process integrity analysis to obtain behavior data analysis results; and the scenario parameters of the virtual training environment are dynamically adjusted based on the emotional stability analysis results and the behavior data analysis results; wherein the adjustment of the scenario parameters includes: if the fluctuation range of the emotional value exceeds the preset emotional stability value range, the interaction difficulty set in the virtual training environment is reduced; if the effectiveness value of the behavior response is lower than the preset behavior response effectiveness threshold, the guidance prompt data of the virtual training environment is increased.

[0119] As an optional implementation, the model and strategy optimization module 55 is further used to divide the target weak supervision object into a training group and a control group, wherein the training group performs the immersive interactive supervision training, and the control group maintains the original supervision mode training; calculate the difference value of business indicators and the change value of weak status of the two groups before and after training through a difference-in-differences model, wherein the difference value of business indicators includes the improvement of business indicator data and the decrease of complaint volume, and the change value of weak status is determined based on the change of the quantitative evaluation result of the weak negative factors; determine the training effect level of immersive interactive supervision training based on the difference value of business indicators; if the training effect level is not within the preset training effect level range, perform correlation analysis based on the results of the biosignals, behavioral data and intelligent diagnosis of weak causes during the training process; adjust the feature weight calculation logic of the weak diagnosis algorithm model based on the analysis results of the correlation factors, and optimize the training content ratio and scene switching rules in the personalized supervision strategy to complete the iterative update of the weak diagnosis algorithm model and the personalized supervision strategy.

[0120] As an optional implementation, the data fusion processing module 51 is further configured to, after the iterative optimization of the weak body diagnosis algorithm model, update the collection dimensions and filtering rules of the multi-source multimodal data according to the feature requirements of the iterative model, add data source collection items that match the feature requirements of the iterative model; perform data cleaning processing on the newly added data source collection items using the same time verification algorithm and outlier filtering rules as the original data to maintain the same data format and data feature standards; and perform feature supplementation and fusion with the newly added data source collection items and the historical fusion feature set to generate an adaptive extended fusion feature set that can be accepted by the iterative weak body diagnosis algorithm model.

[0121] It should be noted that the aforementioned weak component supervision intelligent optimization device can execute the weak component supervision intelligent optimization method provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the embodiments of the weak component supervision intelligent optimization device can be found in the weak component supervision intelligent optimization method provided in the embodiments of this application.

[0122] Figure 6 This is a schematic diagram of the hardware structure of the electronic device for implementing the weak body supervised intelligent optimization method provided in the embodiments of this application, as shown below. Figure 6 As shown, the electronic device 600 includes: One or more processors 610 and memory 620, Figure 6 Take the 610 processor as an example.

[0123] The processor 610 and the memory 620 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0124] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the weak body-supervised intelligent optimization method in the embodiments of this application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, thereby implementing the weak body-supervised intelligent optimization method in the above-described method embodiments.

[0125] The memory 620 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the weak body supervision intelligent optimization device. Furthermore, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory remotely located relative to the processor 610, and these remote memories can be connected to the weak body supervision intelligent optimization device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0126] The one or more modules are stored in the memory 620. When executed by the one or more processors 610, they perform the weak body supervised intelligent optimization method in any of the above method embodiments, for example, performing the above-described... Figure 2 Method steps S1 to S5, Figure 3 Method steps S21 to S24, Figure 4 Steps S31 to S34 in the method are implemented. Figure 5 The functions of modules 51-55 in the document.

[0127] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0128] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 6 One of the processors 610 can enable the one or more processors to execute the weak body supervised intelligent optimization method in any of the above method embodiments, for example, to perform the above-described... Figure 2 Method steps S1 to S5, Figure 3 Method steps S21 to S24, Figure 4 Steps S31 to S34 in the method are implemented. Figure 5 The functions of modules 51-55 in the document.

[0129] This application provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, which, when executed by an electronic device, enable the electronic device to perform the weak body supervised intelligent optimization method in any of the above method embodiments, for example, to perform the above-described... Figure 2 Method steps S1 to S5, Figure 3 Method steps S21 to S24, Figure 4 Steps S31 to S34 in the method are implemented. Figure 5 The functions of modules 51-55 in the document.

[0130] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A weak-body supervised intelligent optimization method, characterized in that, include: The system integrates and processes multi-source, multi-modal data acquired from preset channels, including behavioral data, business data, and external related data of the target weak supervision object. The fused multi-source multimodal data is input into a preset weak body diagnosis algorithm model. Through feature extraction, factor weight calculation and outlier detection, intelligent diagnosis of the causes of weak bodies is performed to obtain the negative factors and quantitative evaluation results of the weak body corresponding to the target weak body supervision object. Based on the negative factors of the weak body and the quantitative evaluation results, a personalized supervision strategy corresponding to the target weak body supervision object is generated by matching with a preset knowledge graph. The personalized supervision strategy includes training content data, training intensity data and scenario configuration data. A virtual training environment is constructed, and immersive interactive supervised training is conducted according to the personalized supervision strategy. The biological signals and behavioral data of the target weak supervisee are collected in real time during the training process, and the virtual training environment is dynamically adjusted according to the biological signals and behavioral data. The results of the immersive interactive supervision training are evaluated using a dual difference model, and the weak body diagnosis algorithm model and the personalized supervision strategy are iteratively optimized based on the evaluation results.

2. The weak-body supervised intelligent optimization method according to claim 1, characterized in that, The steps of fusing and processing multi-source, multi-modal data acquired from preset channels include: Raw data is acquired in real time from the preset channels through a pre-built real-time data pipeline; Speech-to-text processing is performed on the speech data in the original data to extract the associated information and historical frequency in the text. The weight value of the associated information is determined by the correlation calculation method between word frequency and document distribution. A time calibration algorithm is used to unify data from different sources to the same time base. The time calibration algorithm includes time zone deviation correction and transmission delay compensation logic. Outlier filtering is performed on physiological signal data in the original data based on statistical rules to remove interfering data that exceeds the normal fluctuation range; The processed structured data and unstructured data are combined to generate a standardized fusion feature set.

3. The weak-body supervised intelligent optimization method according to claim 1, characterized in that, The steps of inputting the fused multi-source, multi-modal data into a preset weak body diagnosis algorithm model, and performing intelligent diagnosis of the causes of weak bodies through feature extraction, factor weight calculation, and outlier detection, include: Feature analysis is performed on the fused multi-source multimodal data to extract multi-level features associated with the weak state. The multi-level features include feature vectors of skill performance layer, feature vectors of psychological state layer, and feature vectors of environmental factors layer. The influence weights of feature vectors at each level are calculated using a preset feature importance evaluation algorithm, and feature vectors whose weight values ​​exceed the preset weight threshold range are selected. The feature contribution analysis algorithm is used to perform attribution calculation on the feature vectors of each level, and the attribution metric of the feature vectors of each level to the weak body state is obtained. Based on the ranking results of the attribution metric and the preset feature attribute classification, the weakness negative factors and the corresponding quantitative evaluation results are generated. The weakness negative factors include skill-level factors, psychological-level factors and environmental-level factors.

4. The weak-body supervised intelligent optimization method according to claim 1, characterized in that, The step of generating a personalized supervision strategy for the target weak subject through matching with a preset knowledge graph based on the weak subject's negative factors and the quantitative evaluation results includes: The system invokes a pre-defined knowledge graph, which contains data on the relationship between weak factors and supervision schemes, as well as training content adaptation rules for different scenarios. The weak negative factors are matched with factor categories in the knowledge graph to adapt to the corresponding benchmark supervision scheme. The parameters of the benchmark supervision program are optimized based on the quantitative evaluation results. The optimized parameters include allocating the proportion of training content according to the weight of the negative factors of the weak body, and setting the training intensity gradient based on the quantitative value of the weak body status. Based on the adjusted benchmark supervision scheme, a personalized supervision strategy is generated according to the characteristic data of the target weak supervision object, which includes the priority of training content, training cycle and scenario switching rules.

5. The weak-body supervised intelligent optimization method according to claim 1, characterized in that, The steps of constructing a virtual training environment, conducting immersive interactive supervised training based on the personalized supervision strategy, collecting biosignals and behavioral data of the target weak supervisee in real time during the training process, and dynamically adjusting the virtual training environment based on the biosignals and behavioral data include: Based on the scenario configuration data in the personalized supervision strategy, a virtual training scenario containing features of multiple types of interactive objects is constructed, wherein the interactive object features match the object attributes in the actual business scenario; During the execution of the immersive interactive supervised training, the physiological state signals and training operation behavior data of the target weak supervisee are collected in real time. Emotional stability analysis was performed on the collected physiological state signals to obtain the emotional stability analysis results; The collected training operation behavior data is subjected to response effectiveness analysis and process integrity analysis to obtain behavior data analysis results; Based on the results of the emotion stability analysis and the results of the behavioral data analysis, the scene parameters of the virtual training environment are dynamically adjusted. The adjustment of the scene parameters includes: if the fluctuation range of the emotion value exceeds the preset emotion stability value range, then the interaction difficulty set in the virtual training environment is reduced; if the validity value of the behavior response is lower than the preset behavior response validity threshold, then the guidance prompt data of the virtual training environment is increased.

6. The weak-body supervised intelligent optimization method according to claim 1, characterized in that, The steps of evaluating the results of the immersive interactive supervision training using a dual-difference model, and iteratively optimizing the weak component diagnosis algorithm model and the personalized supervision strategy based on the evaluation results, include: The target weak subjects are divided into a training group and a control group. The training group performs the immersive interactive supervision training, while the control group maintains the original supervision mode training. The difference in business indicators and the change in the weakness status of the two groups of objects before and after training are calculated by using a difference-in-differences model. The difference in business indicators includes the increase in business indicator data and the decrease in the number of complaints. The change in the weakness status is determined based on the change in the quantitative evaluation result of the weakness negative factors. The training effectiveness level of the immersive interactive supervision training is determined based on the difference values ​​of the aforementioned business indicators. If the training effect level is not within the preset training effect level range, then a correlation analysis is performed based on the biosignals, behavioral data, and the results of intelligent diagnosis of the causes of weakness during the training process. Based on the analysis results of related factors, the feature weight calculation logic of the weakness diagnosis algorithm model is adjusted, and the training content allocation and scenario switching rules in the personalized supervision strategy are optimized to complete the iterative update of the weakness diagnosis algorithm model and the personalized supervision strategy.

7. The weak-body supervised intelligent optimization method according to claim 1, characterized in that, The step of fusing and processing multi-source, multi-modal data acquired from preset channels further includes: After the weak body diagnosis algorithm model is iteratively optimized, the collection dimensions and filtering rules of the multi-source multimodal data are updated according to the feature requirements of the iterative model, and data source collection items that match the feature requirements of the iterative model are added. The newly added data source collection items are cleaned using the same time verification algorithm and outlier filtering rules as the original data to maintain the same data format and data feature standards. The newly added data source collection items are combined with the historical fusion feature set to supplement and fuse features, so as to generate an adapted and extended fusion feature set that can be accepted by the iterative weak body diagnosis algorithm model.

8. A weak-body supervised intelligent optimization device, characterized in that, include: The data fusion processing module is used to fuse and process multi-source, multi-modal data acquired from preset channels. The multi-source, multi-modal data includes behavioral data, business data, and external related data of the target weak supervision object. The weak body cause diagnosis module is used to input the fused multi-source multimodal data into the preset weak body diagnosis algorithm model, and perform intelligent diagnosis of weak body causes through feature extraction, factor weight calculation and outlier detection to obtain the weak body negative factors and quantitative evaluation results corresponding to the target weak body supervision object; The supervision strategy generation module is used to generate a personalized supervision strategy corresponding to the target weak object based on the negative factors of the weak object and the quantitative evaluation results through a preset knowledge graph matching. The personalized supervision strategy includes training content data, training intensity data and scenario configuration data. The virtual training module is used to construct a virtual training environment, conduct immersive interactive supervised training according to the personalized supervision strategy, collect biosignals and behavioral data of the target weak supervisee in real time during the training process, and dynamically adjust the virtual training environment according to the biosignals and behavioral data. The model and strategy optimization module is used to evaluate the results of the immersive interactive supervision training through a dual difference model, and to iteratively optimize the weak body diagnosis algorithm model and the personalized supervision strategy based on the evaluation results.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the weak body supervised intelligent optimization method according to any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by an electronic device, cause the electronic device to perform the weak body supervised intelligent optimization method according to any one of claims 1-7.

Images