A method and system for questioning and dispatching

Abstract

The application relates to a question and scheduling method and system, and belongs to the technical field of natural language processing and intelligent scheduling. The method comprises the following steps: acquiring question interaction multi-source state data, setting quantifiable monitoring indexes and three-level judgment standards in each dimension, and constructing a hierarchical process monitoring mechanism; preprocessing multi-source data, building a multi-dimensional state recognition model, judging the state level according to the standards and outputting a research and judgment result; constructing a strategy set, continuously monitoring state changes, constructing a state feature vector and a strategy parameter vector; executing a scheduling action, acquiring feedback evaluation effect, correlating and storing related data and calculating strategy matching degree, and iteratively optimizing monitoring indexes and judgment standards. The application realizes accurate perception and scientific research and judgment of multi-dimensional states in the whole process of question interaction, solves the pain points of rigidity, lack of whole-process monitoring and iteration ability of the prior art, and is suitable for various intelligent question interaction scenes.

Description

Technical Field

[0001] This invention belongs to the field of natural language processing and intelligent scheduling technology, specifically relating to a questioning and scheduling method and system. Background Technology

[0002] In scenarios such as intelligent assessment, talent selection, and skills evaluation that rely on interactive questioning, the coherence, rhythm, and controllability of the questioning process directly determine the efficiency and accuracy of the assessment results. With the development of intelligent interaction technology, existing technologies have achieved basic intelligent questioning functions, i.e., automatically outputting questions according to a preset list. However, there are significant technical shortcomings in the dynamic control of the entire questioning process, failing to meet the needs of efficient and accurate intelligent assessment. Specific deficiencies are as follows:

[0003] Existing intelligent questioning systems only focus on the "question output" stage, lacking a multi-dimensional status monitoring mechanism for the entire questioning interaction process. They cannot capture in real time the evaluation subject's response status (such as response completeness, semantic relevance, and emotional fluctuations), the environmental conditions of the evaluation scenario (such as noise interference and abnormal lighting), and the operational status of the interaction process (such as data transmission delays and program malfunctions). This lack of status awareness causes the system to proceed with questions according to fixed logic, failing to detect problems where "the question does not match the evaluation subject's state." For example, continuously asking frequently questions to an emotionally stressed evaluation subject will exacerbate their resistance, and missing information in responses cannot be detected in a timely manner, ultimately affecting the evaluation experience and data quality.

[0004] Current technologies generally employ a fixed, "one-size-fits-all" scheduling model, applying the same questioning intervals, speaking speeds, and sequences to all assessment subjects. This fails to consider differences in response speed and cognitive levels among assessment subjects, nor does it incorporate dynamic adjustments based on real-time conditions. This rigid scheduling approach has significant drawbacks: for assessment subjects who respond slowly, a fixed pace prevents them from fully considering their answers, leading to information omissions; for assessment subjects who respond quickly, an excessively slow pace reduces assessment efficiency; furthermore, it cannot mitigate the impact of unexpected events such as environmental interference or equipment malfunctions through pace adjustments, further exacerbating process chaos.

[0005] The existing system lacks a linkage mechanism between "status assessment and follow-up question triggering." When the evaluated subject's answer contains missing information, deviates from the topic, or is semantically ambiguous, it cannot accurately trigger follow-up questions based on the real-time status; it can only continue with subsequent questions according to a pre-set list. This leads to the omission of key information during the assessment process, resulting in insufficient completeness and validity of the assessment data. At the same time, the connection between follow-up questions and the original questioning process lacks planning. Even when follow-up questions are triggered manually, problems such as mismatch between the content of the follow-up questions and the missing information, and inappropriate timing of the follow-up questions can easily occur, further affecting the comprehensiveness of the assessment results.

[0006] Faced with abnormal scenarios such as assessment interruptions, equipment failures, and severe environmental interference, existing intelligent questioning systems lack mature emergency handling mechanisms. They cannot quickly locate abnormal nodes, back up assessment data, or generate targeted process recovery solutions. Once an anomaly occurs, the entire assessment process often needs to be restarted, resulting in the loss of previous assessment data and an extended assessment cycle. This not only reduces assessment efficiency but also causes resistance from the assessment subjects due to the restart of the process, affecting the continuity and reliability of the assessment.

[0007] In summary, existing intelligent questioning technologies suffer from core defects such as lack of status monitoring, rigid scheduling, disconnect between questioning and follow-up questions, and insufficient emergency response. Essentially, they ignore the need for dynamic control of the entire questioning process and cannot achieve closed-loop management. Summary of the Invention

[0008] To address the aforementioned problems in the existing technology, this invention provides a questioning and scheduling method and system.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] include:

[0011] S1: Obtain multi-source status data of question interaction, set quantifiable monitoring indicators for each dimension; based on the multi-source status data and historical interaction data, set three-level indicator judgment criteria, and construct a hierarchical process monitoring mechanism;

[0012] S2: Preprocess the multi-source state data; construct a multi-dimensional state recognition model based on the preprocessed data; determine the state level according to the three-level indicator judgment criteria, and output the state judgment result and the corresponding state type;

[0013] S3: Based on the state assessment results, construct a set of strategies including parameterized strategies, and continuously monitor the state changes during the question-and-answer interaction process; construct the state assessment results into a state feature vector containing multiple state feature dimensions, and construct each scheduling strategy into a corresponding strategy parameter vector; based on a preset matching degree calculation model, calculate the state feature vector and the strategy parameter vector, and output the matching degree result for selecting the scheduling strategy; if the state is normal, continue to advance the question-and-answer process; if an early warning is triggered, execute the adjustment scheduling strategy; if an abnormality is determined, execute the emergency scheduling strategy.

[0014] S4: Execute scheduling actions according to the scheduling strategy, obtain interactive feedback data after scheduling in real time and evaluate the scheduling effect; associate and store the multi-source state data, scheduling strategy and scheduling effect data, and calculate the matching degree of the scheduling strategy under different state scenarios; when the matching degree is lower than the preset threshold after continuous scheduling, update the scheduling strategy parameters under the corresponding state type.

[0015] Specifically, the process of setting quantifiable monitoring indicators for each dimension is as follows: Core monitoring dimensions are divided according to response status, environment status, and operation status, and quantifiable monitoring indicators are set for each dimension; for the response status dimension, the coverage rate of key information in the response is used to set a response completeness indicator, and the semantic fit is set by comparing the response with the core semantics of the question; for the environment status dimension, the interference level indicator is set according to the degree of interference on the interaction; and for the operation status dimension, the stability indicator is set according to data transmission latency and the frequency of fault occurrence.

[0016] Specifically, the process of setting the three-level indicator judgment criteria is as follows: collect multi-source status data and historical interaction data of monitoring indicators of each dimension, and set three-level indicator ranges of normal, early warning and abnormal for each dimension of monitoring indicators; preset the indicator range corresponding to each level; in the hierarchical process monitoring mechanism, when the monitoring indicator of any layer meets the abnormal judgment condition, it triggers the recalculation or update of the corresponding status characteristics of at least another layer.

[0017] Specifically, the process of constructing the hierarchical process monitoring mechanism is as follows:

[0018] The hierarchical process monitoring mechanism includes: response status layer, environment status layer, and operation status layer;

[0019] The core monitoring scope and specific responsibilities of each layer are preset; the response status layer is responsible for continuously tracking and evaluating the completeness of the object's response content, response attitude and emotional fluctuations; the environment status layer is responsible for real-time monitoring and evaluation of the types and impact range of interference sources in the scenario; the operation status layer is responsible for dynamically controlling the smoothness of data transmission and the program operation status during the interaction process.

[0020] Filter the acquired raw data from multiple sources and establish a data sharing channel between layers; set data synchronization trigger conditions so that when any layer detects suspected abnormal data, the relevant data is automatically synchronized to other related layers.

[0021] Specifically, the preprocessing of multi-source state data involves: cleaning the collected multi-source state data to remove invalid data and redundant information; unifying the storage and expression formats of data from different sources and in different formats; extracting core feature information from unstructured data, converting it into standardized structured data, and verifying the consistency and usability of the data.

[0022] Specifically, the process of constructing the multi-dimensional state recognition model is as follows:

[0023] It includes: a feature extraction layer, a feature fusion layer, and a state classification layer;

[0024] The feature extraction layer employs corresponding feature extraction methods for different types of preprocessed data, extracting semantic features from the answer text, Mel frequency-related features from the speech data, interference features from the environmental data, and state features from the running data.

[0025] The feature fusion layer uses a feature concatenation method to integrate the features extracted from each dimension into a unified feature vector, removing redundant and overlapping parts between features; the fused feature vector is used for subsequent scheduling strategy matching degree calculation, representing the comprehensive state features of the current question interaction;

[0026] The state classification layer uses a classification algorithm to construct a classifier. The fused feature vector is input into the classifier to classify and identify the response state, environment state, and running state.

[0027] Specifically, the process of constructing a strategy set containing parameterized strategies is as follows: sorting out the processing requirements and scheduling objectives corresponding to the state judgment results; designing adaptive scheduling actions, decomposing them into quantifiable strategy parameters and defining value constraints; establishing structured associations according to state types; and forming a standardized strategy set with parameter extension interfaces and fast retrieval indexes.

[0028] Specifically, the process of constructing each scheduling strategy into a corresponding strategy parameter vector is as follows: based on the influence weight of the scheduling strategy on the state, quantifiable core parameters are selected; the parameter quantification standard and dimension unit are unified, and the parameters are mapped to the [0,1] interval to complete normalization; the parameters are arranged in a preset fixed dimension order and the index is retained to form a strategy parameter vector with a unified structure.

[0029] Specifically, the process of calculating the state feature vector and the strategy parameter vector is as follows: using the state improvement effect, process coherence, and question-progression efficiency as core measurement dimensions, the state differences before and after scheduling execution are compared; the improvement of the scheduling strategy on abnormal and early warning states is calculated; the percentage of uninterrupted question-progression duration during scheduling is statistically analyzed; the completion efficiency of question-progression according to plan under the corresponding state scenario is measured; and the matching degree of the scheduling strategy under different state scenarios is calculated by combining the statistical results of various calculations.

[0030] Specifically, the process of triggering an early warning and executing the adjustment scheduling strategy is as follows: matching the specific type of the early warning and executing the corresponding adjustment action according to the specific early warning type; for response-type early warnings, slowing down the questioning speed and extending the response interval; for environmental early warnings, pausing the questioning and resuming after the interference weakens; for operational early warnings, postponing the questioning order; for early warnings with missing response information, generating follow-up questions based on the missing information and pushing them.

[0031] Specifically, the process of updating the scheduling strategy parameters under the corresponding state type is as follows: when the matching degree under a single state type does not reach the preset threshold, extract the associated historical data to locate the core strategy parameters, iteratively adjust and verify them, and then update the corresponding scheduling strategy parameters; when the matching degree under a single state type still does not reach the standard after the strategy parameters are updated, analyze the adaptability of the monitoring indicators or three-level judgment criteria of the corresponding dimension, adjust the indicator settings or judgment criteria range synchronously, and optimize the corresponding strategy parameters in conjunction; when the matching degree under multiple different state scenarios does not reach the preset threshold, investigate the mapping logic matching defects between the state feature vector and the strategy parameter vector, and synchronously calibrate the scheduling strategy parameters of each relevant state scenario.

[0032] Specifically, a questioning and scheduling system includes:

[0033] Monitoring indicator construction module: acquires multi-source status data of question interaction, sets quantifiable monitoring indicators for each dimension; based on the multi-source status data and historical interaction data, sets three-level indicator judgment criteria, and constructs a hierarchical process monitoring mechanism;

[0034] State recognition and judgment module: preprocesses the multi-source state data; constructs a multi-dimensional state recognition model based on the preprocessed data; determines the state level according to the three-level indicator judgment criteria, and outputs the state judgment result and the corresponding state type;

[0035] The scheduling rule matching module constructs a set of strategies containing parameterized policies based on the state assessment results, and continuously monitors the state changes during the question-and-answer interaction process; it constructs a state feature vector containing multiple state feature dimensions from the state assessment results, and constructs a corresponding policy parameter vector for each scheduling policy; based on a preset matching degree calculation model, it calculates the matching degree of the state feature vector and the policy parameter vector, and outputs the matching degree result for selecting the scheduling policy; if the state is normal, the question-and-answer process continues; if an early warning is triggered, an adjustment-type scheduling policy is executed; if an anomaly is determined, an emergency-type scheduling policy is executed.

[0036] Scheduling Iteration Optimization Module: Executes scheduling actions according to the scheduling strategy, obtains interactive feedback data after scheduling in real time and evaluates the scheduling effect; associates and stores the multi-source state data, scheduling strategy and scheduling effect data, and calculates the matching degree of the scheduling strategy under different state scenarios; when the matching degree is lower than the preset threshold after continuous scheduling, updates the scheduling strategy parameters under the corresponding state type.

[0037] The beneficial effects of this invention are as follows:

[0038] (1) By setting up a hierarchical process monitoring mechanism, quantifiable monitoring indicators and three-level indicator judgment standards, and with a multi-dimensional status recognition model, the system can achieve accurate perception and scientific judgment of the multi-dimensional status of the entire question interaction process, and completely solve the technical pain points of the existing technology that lacks full-process status monitoring and cannot accurately capture the response, environment and operation status; the monitoring level is divided according to the answer, environment and operation dimensions and the core responsibilities of each level are clarified. A data sharing and synchronous linkage mechanism is established between the layers to ensure that there are no blind spots in status monitoring. At the same time, quantifiable indicators that fit the actual needs are set for each dimension. The three-level indicator range is preset in combination with multi-source and historical data. Then, the standardized data after preprocessing is accurately classified and identified by the recognition model with feature extraction, fusion and classification layers. It can quickly distinguish the three types of status and specific subtypes of normal, warning and abnormal, and provide a reliable basis for subsequent scheduling strategy matching, avoid blind questioning due to lack of status perception, and improve the efficiency and accuracy of status judgment, and lay a solid foundation for the smooth progress of the question process;

[0039] (2) By setting up a set of strategies including parameterized strategies, a vector mapping relationship between states and strategies, and a calculation-driven decision-making mechanism, dynamic adaptation scheduling and precise iterative optimization of the questioning process are realized; a structured set of parameterized strategies is constructed based on the state judgment results, specific state subtypes are decomposed and the strategy parameter range is preset, and the precise and rapid matching of states and strategies is realized through a vector mapping table and a fast retrieval mechanism combined with a matching degree calculation model. When the state is normal, the questioning is steadily promoted; when an early warning is issued, adaptation actions such as adjusting the speech rate, extending the interval, pausing to avoid interference, delaying the order, and precise follow-up questions are executed according to the specific type; when an abnormality occurs, emergency handling is performed to ensure the coherence and adaptation of the questioning process; at the same time, multi-source state data, scheduling strategies, and effect data are stored together, and the strategy matching degree is calculated around the state improvement effect, process coherence, and questioning promotion efficiency. Then, based on the trigger conditions, the strategy parameters, monitoring indicators, judgment criteria, or mapping relationships are updated in a targeted manner according to the matching degree differences, forming a "perception-computation decision-evaluation-precise optimization" process. The complete closed loop not only improves the accuracy, flexibility, and emergency response capabilities of question scheduling, but also allows the entire method to be continuously optimized and upgraded with the application scenarios, highlighting non-obvious creativity. Attached Figure Description

[0040] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0041] Figure 1 This is a flowchart illustrating a questioning and scheduling method and system according to the present invention;

[0042] Figure 2 This is a data flow diagram of a questioning and scheduling method and system according to the present invention. Detailed Implementation

[0043] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0044] Please see Figure 1-2 A questioning and scheduling method and system;

[0045] include:

[0046] S1: Obtain multi-source status data of question interaction, set quantifiable monitoring indicators for each dimension; based on the multi-source status data and historical interaction data, set three-level indicator judgment criteria, and construct a hierarchical process monitoring mechanism;

[0047] S2: Preprocess the multi-source state data; construct a multi-dimensional state recognition model based on the preprocessed data; determine the state level according to the three-level indicator judgment criteria, and output the state judgment result and the corresponding state type;

[0048] S3: Based on the state assessment results, construct a set of strategies including parameterized strategies, and continuously monitor the state changes during the question-and-answer interaction process; construct the state assessment results into a state feature vector containing multiple state feature dimensions, and construct each scheduling strategy into a corresponding strategy parameter vector; based on a preset matching degree calculation model, calculate the state feature vector and the strategy parameter vector, and output the matching degree result for selecting the scheduling strategy; if the state is normal, continue to advance the question-and-answer process; if an early warning is triggered, execute the adjustment scheduling strategy; if an abnormality is determined, execute the emergency scheduling strategy.

[0049] S4: Execute scheduling actions according to the scheduling strategy, obtain interactive feedback data after scheduling in real time and evaluate the scheduling effect; associate and store the multi-source state data, scheduling strategy and scheduling effect data, and calculate the matching degree of the scheduling strategy under different state scenarios; when the matching degree is lower than the preset threshold after continuous scheduling, update the scheduling strategy parameters under the corresponding state type.

[0050] In this embodiment, the multi-source state data refers to various state-related raw data obtained from different dimensions and collection channels throughout the entire question-and-answer interaction process. It is categorized around three monitoring dimensions and specifically includes three parts: first, response state data, including the evaluation object's response text, voice characteristics, response duration, etc.; second, environmental state data, including environmental noise, interference sources, and scene environmental parameters within the question-and-answer interaction scenario; and third, operational state data, including data transmission latency, program running logs, and device operating parameters during the question-and-answer interaction process. These three types of data together constitute the core data foundation for end-to-end state perception.

[0051] In this embodiment, the monitoring indicators specifically include: first, response status dimension indicators, the core of which are response completeness indicators (measured by the coverage rate of key information in the response) and semantic fit indicators (measured by the semantic matching degree between the response and the core of the question); second, environmental status dimension indicators, the core of which is the interference level indicator (measured by the degree of impact of interference on the question interaction); and third, operational status dimension indicators, the core of which is the stability indicator (measured by data transmission latency and the frequency of failures). All indicators are quantitative bases that can directly determine the status level.

[0052] In this embodiment, the state change specifically refers to the dynamic changes of each dimension of the state in the time dimension throughout the entire question-and-answer interaction process. The core is based on the preset three-level states of normal, warning, and abnormal, and specific sub-types. Specifically, it includes three aspects of change: First, the answer state change, such as the response from smooth to delayed, the response completeness from meeting the standard to missing, and the semantic fit from fit to deviation; second, the environmental state change, such as the scene from no interference to slight interference, and slight interference to severe interference; third, the running state change, such as the running from stable to slow transmission, and slow transmission to program failure. These changes are the core triggering basis for the dynamic matching and scheduling strategy.

[0053] Specifically, the process of setting quantifiable monitoring indicators for each dimension is as follows: Core monitoring dimensions are divided according to response status, environment status, and operation status, and quantifiable monitoring indicators are set for each dimension; for the response status dimension, the coverage rate of key information in the response is used to set a response completeness indicator, and the semantic fit is set by comparing the response with the core semantics of the question; for the environment status dimension, the interference level indicator is set according to the degree of interference on the interaction; and for the operation status dimension, the stability indicator is set according to data transmission latency and the frequency of fault occurrence.

[0054] Specifically, the process of setting the three-level indicator judgment criteria is as follows: collect multi-source status data and historical interaction data of monitoring indicators of each dimension, and set three-level indicator ranges of normal, early warning and abnormal for each dimension of monitoring indicators; preset the indicator range corresponding to each level; in the hierarchical process monitoring mechanism, when the monitoring indicator of any layer meets the abnormal judgment condition, it triggers the recalculation or update of the corresponding status characteristics of at least another layer.

[0055] Specifically, the process of constructing the hierarchical process monitoring mechanism is as follows:

[0056] The hierarchical process monitoring mechanism includes: response status layer, environment status layer, and operation status layer;

[0057] The core monitoring scope and specific responsibilities of each layer are preset; the response status layer is responsible for continuously tracking and evaluating the completeness of the object's response content, response attitude and emotional fluctuations; the environment status layer is responsible for real-time monitoring and evaluation of the types and impact range of interference sources in the scenario; the operation status layer is responsible for dynamically controlling the smoothness of data transmission and the program operation status during the interaction process.

[0058] In this embodiment, the specific implementation of building a layered process monitoring mechanism is as follows: Three independent monitoring service modules—response status, environment status, and runtime status—are built based on a microservice architecture. Each module is deployed using containerization to ensure high availability. A dedicated data processing thread is configured for each module using the Flink real-time processing engine, and the monitoring frequency is set using the Quartz scheduled task framework (adapting to the scenario as needed). A data sharing channel between layers is established using the Kafka message queue, and a partitioning strategy is used to achieve accurate routing of data from different dimensions. A data fluctuation threshold is set using a sliding window anomaly detection algorithm. When any module detects data exceeding the threshold, a Kafka message push is automatically triggered, synchronizing the abnormal data and associated context to other related modules. Simultaneously, the log and alarm components are linked to record the abnormal trajectory, achieving real-time collaborative monitoring and anomaly tracing of the entire process status.

[0059] Filter the acquired raw data from multiple sources and establish a data sharing channel between layers; set data synchronization trigger conditions so that when any layer detects suspected abnormal data, the relevant data is automatically synchronized to other related layers.

[0060] Specifically, the preprocessing of multi-source state data involves: cleaning the collected multi-source state data to remove invalid data and redundant information; unifying the storage and expression formats of data from different sources and in different formats; extracting core feature information from unstructured data, converting it into standardized structured data, and verifying the consistency and usability of the data.

[0061] Specifically, the process of constructing the multi-dimensional state recognition model is as follows:

[0062] It includes: a feature extraction layer, a feature fusion layer, and a state classification layer;

[0063] The feature extraction layer employs corresponding feature extraction methods for different types of preprocessed data, extracting semantic features from the answer text, Mel frequency-related features from the speech data, interference features from the environmental data, and state features from the running data.

[0064] The feature fusion layer uses a feature concatenation method to integrate the features extracted from each dimension into a unified feature vector, removing redundant and overlapping parts between features; the fused feature vector is used for subsequent scheduling strategy matching degree calculation, representing the comprehensive state features of the current question interaction;

[0065] The state classification layer uses a classification algorithm to construct a classifier. The fused feature vector is input into the classifier to classify and identify the response state, environment state, and running state.

[0066] In this embodiment, the specific implementation of constructing a multi-dimensional state recognition model is as follows: An end-to-end model architecture is built based on the TensorFlow deep learning framework; the feature extraction layer calls the BERT pre-trained model to embed words into the response text, generating semantic feature vectors; the Mel frequency cepstral coefficients of the speech data are extracted using the Librosa library as speech features; wavelet transform algorithm is used to extract interference features from environmental data; and statistical methods are used to extract temporal state features from the running logs; the feature fusion layer introduces an attention mechanism, assigns dynamic weights to features of each dimension, concatenates them, and then normalizes them using the BatchNorm layer to remove redundancy; the state classification layer uses a lightweight CNN classifier with cross-entropy loss function as the optimization objective, iteratively trains the model using the Adam optimizer, employs an early stopping strategy during training to prevent overfitting, accelerates the model using TensorRT after training, and deploys it to edge computing nodes to achieve real-time state recognition and output.

[0067] Specifically, the process of constructing a strategy set containing parameterized strategies is as follows: sorting out the processing requirements and scheduling objectives corresponding to the state judgment results; designing adaptive scheduling actions, decomposing them into quantifiable strategy parameters and defining value constraints; establishing structured associations according to state types; and forming a standardized strategy set with parameter extension interfaces and fast retrieval indexes.

[0068] Specifically, the process of constructing each scheduling strategy into a corresponding strategy parameter vector is as follows: based on the influence weight of the scheduling strategy on the state, quantifiable core parameters are selected; the parameter quantification standard and dimension unit are unified, and the parameters are mapped to the [0,1] interval to complete normalization; the parameters are arranged in a preset fixed dimension order and the index is retained to form a strategy parameter vector with a unified structure.

[0069] Specifically, the process of calculating the state feature vector and the strategy parameter vector is as follows: using the state improvement effect, process coherence, and question-progression efficiency as core measurement dimensions, the state differences before and after scheduling execution are compared; the improvement of the scheduling strategy on abnormal and early warning states is calculated; the percentage of uninterrupted question-progression duration during scheduling is statistically analyzed; the completion efficiency of question-progression according to plan under the corresponding state scenario is measured; and the matching degree of the scheduling strategy under different state scenarios is calculated by combining the statistical results of various calculations.

[0070] Specifically, the process of triggering an early warning and executing the adjustment scheduling strategy is as follows: matching the specific type of the early warning and executing the corresponding adjustment action according to the specific early warning type; for response-type early warnings, slowing down the questioning speed and extending the response interval; for environmental early warnings, pausing the questioning and resuming after the interference weakens; for operational early warnings, postponing the questioning order; for early warnings with missing response information, generating follow-up questions based on the missing information and pushing them.

[0071] In this embodiment, the specific practical steps for calculating the matching degree of scheduling strategies under different state scenarios are as follows:

[0072] The first step is to quantify the core metrics: The cosine similarity algorithm is used to calculate the similarity between the state feature vectors before and after scheduling, which is taken as the state improvement effect A (A∈[0,1], the higher the similarity, the larger the value of A); the sliding window algorithm is used to statistically analyze the proportion of the uninterrupted duration of the question-asking process to the total process time, which is taken as the process coherence B (B∈[0,1], the higher the proportion, the larger the value of B); a linear regression model is used to fit the question-asking plan progress curve, and the deviation rate between the actual progress and the planned progress is calculated. The question-asking progress efficiency C is obtained through normalization (C∈[0,1], the smaller the deviation, the larger the value of C).

[0073] The second step is to determine the indicator weights: use the entropy weight method to calculate the objective weights w1, w2, and w3 of A, B, and C (satisfying w1+w2+w3=1) to avoid bias in setting subjective weights;

[0074] The third step is to calculate the matching degree: The matching degree is calculated using a weighted summation formula, the specific formula of which is as follows: Where M is the matching degree of the scheduling strategy (M∈[0,1]), the closer the value of M is to 1, the stronger the adaptability of the scheduling strategy to the current state scenario; finally, the matching degree M and the corresponding A, B, C values ​​and weight parameters are synchronously stored in the time series database to form a complete quantitative analysis link, providing a traceable and verifiable quantitative basis for subsequent iterative optimization.

[0075] Specifically, the process of updating the scheduling strategy parameters under the corresponding state type is as follows: when the matching degree under a single state type does not reach the preset threshold, extract the associated historical data to locate the core strategy parameters, iteratively adjust and verify them, and then update the corresponding scheduling strategy parameters; when the matching degree under a single state type still does not reach the standard after the strategy parameters are updated, analyze the adaptability of the monitoring indicators or three-level judgment criteria of the corresponding dimension, adjust the indicator settings or judgment criteria range synchronously, and optimize the corresponding strategy parameters in conjunction; when the matching degree under multiple different state scenarios does not reach the preset threshold, investigate the mapping logic matching defects between the state feature vector and the strategy parameter vector, and synchronously calibrate the scheduling strategy parameters of each relevant state scenario.

[0076] Specifically, a questioning and scheduling system includes:

[0077] Monitoring indicator construction module: acquires multi-source status data of question interaction, sets quantifiable monitoring indicators for each dimension; based on the multi-source status data and historical interaction data, sets three-level indicator judgment criteria, and constructs a hierarchical process monitoring mechanism;

[0078] State recognition and judgment module: preprocesses the multi-source state data; constructs a multi-dimensional state recognition model based on the preprocessed data; determines the state level according to the three-level indicator judgment criteria, and outputs the state judgment result and the corresponding state type;

[0079] The scheduling rule matching module constructs a set of strategies containing parameterized policies based on the state assessment results, and continuously monitors the state changes during the question-and-answer interaction process; it constructs a state feature vector containing multiple state feature dimensions from the state assessment results, and constructs a corresponding policy parameter vector for each scheduling policy; based on a preset matching degree calculation model, it calculates the matching degree of the state feature vector and the policy parameter vector, and outputs the matching degree result for selecting the scheduling policy; if the state is normal, the question-and-answer process continues; if an early warning is triggered, an adjustment-type scheduling policy is executed; if an anomaly is determined, an emergency-type scheduling policy is executed.

[0080] Scheduling Iteration Optimization Module: Executes scheduling actions according to the scheduling strategy, obtains interactive feedback data after scheduling in real time and evaluates the scheduling effect; associates and stores the multi-source state data, scheduling strategy and scheduling effect data, and calculates the matching degree of the scheduling strategy under different state scenarios; when the matching degree is lower than the preset threshold after continuous scheduling, updates the scheduling strategy parameters under the corresponding state type.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A questioning and scheduling method, characterized in that, include: S1: Obtain multi-source state data of question interaction and set quantifiable monitoring indicators for each dimension; Based on the multi-source status data and historical interaction data, a three-level indicator judgment standard is set, and a hierarchical process monitoring mechanism is constructed. The hierarchical process monitoring mechanism includes: response status layer, environment status layer, and operation status layer; The core monitoring scope and specific responsibilities of each layer are preset; the response status layer is responsible for continuously tracking and evaluating the completeness of the object's response content, response attitude and emotional fluctuations; the environment status layer is responsible for real-time monitoring and evaluation of the types and impact range of interference sources in the scenario; the operation status layer is responsible for dynamically controlling the smoothness of data transmission and the program operation status during the interaction process. Filter the acquired raw data from multiple sources and establish a data sharing channel between layers; set data synchronization trigger conditions so that when any layer detects suspected abnormal data, the relevant data will be automatically synchronized to other related layers; S2: Preprocess the multi-source state data; construct a multi-dimensional state recognition model based on the preprocessed data; determine the state level according to the three-level indicator judgment criteria, and output the state judgment result and the corresponding state type; The specific process for constructing the multi-dimensional state recognition model is as follows: It includes: a feature extraction layer, a feature fusion layer, and a state classification layer; The feature extraction layer employs corresponding feature extraction methods for different types of preprocessed data, extracting semantic features from the answer text, Mel frequency-related features from the speech data, interference features from the environmental data, and state features from the running data. The feature fusion layer uses a feature concatenation method to integrate the features extracted from each dimension into a unified feature vector, removing redundant and overlapping parts between features; the feature vector is used for subsequent scheduling strategy matching degree calculation, representing the comprehensive state features of the current question interaction; The state classification layer uses a classification algorithm to construct a classifier. The fused feature vector is input into the classifier to classify and identify the response state, environment state, and running state. S3: Based on the state assessment results, construct a set of strategies including parameterized strategies, and continuously monitor the state changes during the question-and-answer interaction process; construct the state assessment results into a state feature vector containing multiple state feature dimensions, and construct each scheduling strategy into a corresponding strategy parameter vector; based on a preset matching degree calculation model, calculate the state feature vector and the strategy parameter vector, and output the matching degree result for selecting the scheduling strategy; if the state is normal, continue to advance the question-and-answer process; if an early warning is triggered, execute the adjustment scheduling strategy; if an abnormality is determined, execute the emergency scheduling strategy. S4: Execute scheduling actions according to the scheduling strategy, obtain interactive feedback data after scheduling in real time and evaluate the scheduling effect; associate and store the multi-source state data, scheduling strategy and scheduling effect data, and calculate the matching degree of the scheduling strategy under different state scenarios; when the matching degree is lower than the preset threshold after continuous scheduling, update the scheduling strategy parameters under the corresponding state type.

2. The method according to claim 1, characterized in that, The specific process for setting quantifiable monitoring indicators for each dimension is as follows: Core monitoring dimensions are divided according to response status, environment status, and operational status, and quantifiable monitoring indicators are set for each dimension; for the response status dimension, the coverage rate of key information in the response is used to set a response completeness indicator, and the semantic fit is set by comparing the response with the core semantics of the question; for the environment status dimension, the interference level indicator is set according to the degree of interference on the interaction; and for the operational status dimension, the stability indicator is set according to data transmission latency and the frequency of fault occurrence.

3. The method according to claim 1, characterized in that, The specific process of setting the three-level indicator judgment criteria is as follows: collect multi-source status data and historical interaction data of monitoring indicators of each dimension, and set three-level indicator ranges of normal, early warning and abnormal for each dimension of monitoring indicators; preset the indicator range corresponding to each level. In the hierarchical process monitoring mechanism, when the monitoring indicator of any layer meets the abnormal judgment condition, it triggers the recalculation or update of the corresponding status characteristics of at least another layer.

4. The method according to claim 1, characterized in that, The specific process for preprocessing multi-source state data is as follows: cleaning the collected multi-source state data to remove invalid data and redundant information; unifying the storage and expression formats of data from different sources and in different formats; extracting core feature information from unstructured data, converting it into standardized structured data, and verifying the consistency and usability of the data.

5. The method according to claim 1, characterized in that, The specific process of constructing a strategy set containing parameterized strategies is as follows: sorting out the processing requirements and scheduling objectives corresponding to the state judgment results; designing adaptive scheduling actions, decomposing them into quantifiable strategy parameters and defining value constraints; establishing structured associations according to state types; and forming a standardized strategy set with parameter extension interfaces and fast retrieval indexes.

6. The method according to claim 1, characterized in that, The specific process of constructing each scheduling strategy into a corresponding strategy parameter vector is as follows: based on the influence weight of the scheduling strategy on the state, select quantifiable core parameters; unify the parameter quantification standard and dimension unit, map the parameters to the [0,1] interval to complete normalization; arrange the parameters in a preset fixed dimension order and retain the index to form a strategy parameter vector with a unified structure.

7. The method according to claim 1, characterized in that, The specific process for calculating the state feature vector and strategy parameter vector is as follows: using state improvement effect, process coherence, and questioning efficiency as core measurement dimensions, compare the state differences before and after scheduling execution; calculate the improvement of the scheduling strategy on abnormal and early warning states; The statistical analysis measures the percentage of uninterrupted question-asking processes during the scheduling process; it also measures the efficiency of question-asking progress according to plan under corresponding state scenarios; and by combining the statistical results of various calculations, it calculates the matching degree of scheduling strategies under different state scenarios.

8. The method according to claim 1, characterized in that, The specific process of triggering an early warning and executing the adjustment scheduling strategy is as follows: match the specific type of the early warning and execute the corresponding adjustment action according to the specific early warning type; for response-type early warnings, slow down the questioning speed and extend the response interval; for environmental early warnings, suspend questioning and resume after the interference weakens; for operational early warnings, postpone the questioning order; for early warnings with missing response information, generate follow-up questions based on the missing information and push them.

9. The method according to claim 1, characterized in that, The specific process for updating the scheduling strategy parameters under the corresponding state type is as follows: When the matching degree under a single state type does not reach the preset threshold, extract the associated historical data to locate the core strategy parameters, iteratively adjust and verify them, and then update the corresponding scheduling strategy parameters; when the matching degree under a single state type still does not reach the standard after the strategy parameters are updated, analyze the adaptability of the monitoring indicators or three-level judgment criteria of the corresponding dimension, adjust the indicator settings or judgment criteria range synchronously, and optimize the corresponding strategy parameters in conjunction; when the matching degree under multiple different state scenarios does not reach the preset threshold, investigate the mapping logic matching defects between the state feature vector and the strategy parameter vector, and synchronously calibrate the scheduling strategy parameters of each relevant state scenario.

10. A questioning and scheduling system, characterized in that, include: Monitoring metrics construction module: Acquire multi-source status data of question interaction and set quantifiable monitoring metrics for each dimension; Based on the multi-source status data and historical interaction data, a three-level indicator judgment standard is set, and a hierarchical process monitoring mechanism is constructed. The hierarchical process monitoring mechanism includes: response status layer, environment status layer, and operation status layer; The core monitoring scope and specific responsibilities of each layer are preset; the response status layer is responsible for continuously tracking and evaluating the completeness of the object's response content, response attitude and emotional fluctuations; the environment status layer is responsible for real-time monitoring and evaluation of the types and impact range of interference sources in the scenario; the operation status layer is responsible for dynamically controlling the smoothness of data transmission and the program operation status during the interaction process. Filter the acquired raw data from multiple sources and establish a data sharing channel between layers; set data synchronization trigger conditions so that when any layer detects suspected abnormal data, the relevant data will be automatically synchronized to other related layers; State recognition and judgment module: preprocesses the multi-source state data; constructs a multi-dimensional state recognition model based on the preprocessed data; determines the state level according to the three-level indicator judgment criteria, and outputs the state judgment result and the corresponding state type; The specific process for constructing the multi-dimensional state recognition model is as follows: It includes: a feature extraction layer, a feature fusion layer, and a state classification layer; The feature extraction layer employs corresponding feature extraction methods for different types of preprocessed data, extracting semantic features from the answer text, Mel frequency-related features from the speech data, interference features from the environmental data, and state features from the running data. The feature fusion layer uses a feature concatenation method to integrate the features extracted from each dimension into a unified feature vector, removing redundant and overlapping parts between features; the feature vector is used for subsequent scheduling strategy matching degree calculation, representing the comprehensive state features of the current question interaction; The state classification layer uses a classification algorithm to construct a classifier. The fused feature vector is input into the classifier to classify and identify the response state, environment state, and running state. The scheduling rule matching module constructs a set of strategies containing parameterized policies based on the state assessment results, and continuously monitors the state changes during the question-and-answer interaction process; it constructs a state feature vector containing multiple state feature dimensions from the state assessment results, and constructs a corresponding policy parameter vector for each scheduling policy; based on a preset matching degree calculation model, it calculates the matching degree of the state feature vector and the policy parameter vector, and outputs the matching degree result for selecting the scheduling policy; if the state is normal, the question-and-answer process continues; if an early warning is triggered, an adjustment-type scheduling policy is executed; if an anomaly is determined, an emergency-type scheduling policy is executed. Scheduling Iteration Optimization Module: Executes scheduling actions according to the scheduling strategy, obtains interactive feedback data after scheduling in real time and evaluates the scheduling effect; associates and stores the multi-source state data, scheduling strategy and scheduling effect data, and calculates the matching degree of the scheduling strategy under different state scenarios; when the matching degree is lower than the preset threshold after continuous scheduling, updates the scheduling strategy parameters under the corresponding state type.

Images