Intelligent Dispatch System for Mental Health Counseling Services

Abstract

This invention relates to the field of intelligent scheduling technology for mental health, specifically an intelligent scheduling system for mental health counseling services, comprising: data integration, fit assessment, priority calculation, and scheme optimization modules. The data integration module constructs an intervention strategy set and a set of counselor candidates based on the user's current conversation data and historical mental health records. The fit assessment module combines historical records to verify the historical effectiveness of strategies and analyze counselor cooperation, achieving personalized fit assessment. The priority calculation module dynamically calculates a comprehensive scheduling priority score by integrating multi-dimensional indicators. The scheme optimization module simulates and generates multiple allocation schemes, uses a trained scheduling effect prediction model to predict the execution effect of the schemes, and selects the scheduling scheme that meets global constraints and has the best quality. This system can refine the fit assessment dimensions, optimize the task allocation logic, improve the adaptability of counseling scheduling to actual scenarios, and improve the overall scheduling execution effect.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology for mental health, and in particular to an intelligent scheduling system based on mental health counseling services. Background Technology

[0002] The existing scheduling system for mental health counseling services largely relies on matching counselors with counseling tasks based on basic information. The scheduling logic only uses basic user needs, counselor professional categories, and online status for initial screening. The selection of intervention strategies largely depends on standardized procedures, without incorporating user historical counseling data for targeted validation. This scheduling method only matches counselors based on basic information, without verifying the actual effectiveness of intervention strategies on users with similar mental states, or quantitatively analyzing the compatibility and collaboration between counselors and different users, resulting in limited dimensions for compatibility assessment. Conventional counseling task allocation relies solely on fixed priority ranking for direct assignment, failing to generate multiple allocation schemes for comparison and selection. Furthermore, it lacks a pre-implementation simulation of the scheduling scheme's effectiveness, making it difficult to balance system-wide operational constraints with the execution effect of counseling services, thus compromising the rationality and suitability of scheduling decisions.

[0003] In intelligent scheduling scenarios for mental health counseling, it is necessary to optimize the execution logic of fit assessment and task allocation to address the shortcomings of existing assessments that are limited in scope and fixed in allocation methods. To address the issues of intervention strategies and counselor fit assessments not being validated using historical data, and the inability to select the best option through multi-scheme simulation and effect extrapolation in task allocation, a complete intelligent scheduling system needs to be built. This system should improve the entire scheduling mechanism from data integration and fit assessment to scheme optimization, ensuring that the scheduling process aligns with users' personalized counseling needs and the overall operational requirements of the system. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent scheduling system for mental health counseling services.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent scheduling system for mental health counseling services, comprising: The data integration module constructs a set of potential intervention strategies and a set of counselor candidates based on the target user's current counseling session data and historical mental health records; The fit assessment module, in conjunction with the historical mental health records, conducts a personalized fit assessment of the potential intervention strategy set and the counselor candidate set. The fit assessment includes verification of the historical effectiveness of the strategies and analysis of the counselor's past cooperation. The priority calculation module dynamically calculates a comprehensive scheduling priority score for each consultant candidate based on the results of the personalized suitability assessment. The comprehensive scheduling priority score integrates multiple dimensions such as professional matching degree, resource load, user potential preferences and immediate response capability. The scheme optimization module simulates and generates multiple different consultation task allocation schemes based on the comprehensive scheduling priority score and the system's preset global scheduling constraints. For each consultation task allocation scheme, a scheduling effect prediction model trained based on historical data is used to deduce the system state changes and task completion quality indicators after its execution. From all the simulated consultation task allocation schemes, the scheme that makes the deduced task completion quality indicators optimal and satisfies the global scheduling constraints is selected as the final scheduling scheme.

[0006] As a further aspect of the present invention, the construction of a potential intervention strategy set and a set of counselor candidates based on the target user's current counseling session data and historical mental health records includes: Obtain the target user's current consultation session data and historical mental health records, which store the user's previous mental health assessment results, consultation topic records, and intervention feedback. Real-time semantic parsing and emotion feature extraction are performed on the current consultation session data to generate an instant status summary containing the current core issue, emotion intensity, and urgency tags; Based on the real-time state summary, multi-hop association retrieval is performed in the pre-constructed psychological counseling knowledge graph, which includes the association relationships between symptom descriptions, psychological theories, intervention techniques, counselor expertise areas, and counseling resource availability nodes. From the search results of the psychological counseling knowledge graph, a set of potential intervention strategies and a set of counselor candidates that match the current core issues and emotional tendencies in the instantaneous state summary are selected.

[0007] As a further aspect of the present invention, real-time semantic parsing and emotional feature extraction are performed on the current consultation session data to generate an immediate status summary containing the current core issue, emotional intensity, and urgency tags, including: Dependency parsing and named entity recognition are applied to the text stream in the current consultation session data to identify core predicate verbs, emotion-carrying objects, and key event noun phrases; Acoustic feature analysis is performed on the speech stream in the current consultation session data to extract fundamental frequency profile, speech rate variation and energy fluctuation features, and these features are quantified into speech emotion vectors. The identified text entities are fused with the speech emotion vector at the feature level, and mapped to a unified psychological state semantic space through a deep neural network model to obtain the fused psychological state vector of the user's statement. Temporal clustering analysis is performed on the continuous fused mental state vectors to identify the dominant mental state categories during the conversation, and these mental state categories are marked as the current core issue. Calculate the mean of the fused psychological state vector in terms of valence and arousal dimension under the current core issue, and use it as a quantitative representation of the intensity of the emotional tendency; Based on the suddenness of the current core issue, the extreme intensity of the emotional tendency, and whether there are preset risk keywords in the conversation, the urgency label is automatically generated by a rule engine.

[0008] As a further aspect of the present invention, based on the real-time state summary, multi-hop association retrieval is performed in the pre-constructed psychological counseling knowledge graph, including: Using the current core issue and the intensity of the emotional tendency as the initial query nodes, locate the corresponding symptom description or theoretical concept node in the psychological counseling knowledge graph; Starting from the located symptom description or theoretical concept node, perform a breadth-first traversal along the relation edges in the knowledge graph; During the traversal, all nodes with path lengths within a preset range are collected, including intervention technique nodes, common comorbid symptom nodes, nodes of counselors' areas of expertise who are good at handling such issues, and currently available counseling resource nodes. All collected nodes are sorted according to their semantic relevance to the initial query node, path hop count, and edge weight. The top-ranked subset of nodes is selected to form the preliminary search results for the potential intervention strategy set and consultant candidate set.

[0009] As a further aspect of the present invention, in conjunction with the historical mental health records, a personalized fit assessment is performed on the set of potential intervention strategies and the set of counselor candidates, including: From the aforementioned historical mental health records, extract the intervention techniques and feedback scores that users have used in the past to address similar current core issues, and construct a list of historical intervention effectiveness. From the historical mental health records, extract the identifiers of the counselors who have served the user and their corresponding counseling effectiveness evaluations and user satisfaction scores to form a historical cooperation evaluation list. Each intervention technique in the potential intervention strategy set is matched with a record in the historical intervention effectiveness list, and its historical average effectiveness score is calculated as the historical effectiveness verification score of the strategy. Each consultant in the consultant candidate set is matched with the record in the historical cooperation evaluation list. If there is a direct historical cooperation, the past consultation effect evaluation and user satisfaction score are extracted and weighted to obtain the direct cooperation suitability score. For consultants without direct prior collaboration, an indirect professional fit score is calculated based on the alignment between their area of ​​expertise and the long-term issues reflected in the user's historical profile.

[0010] As a further aspect of the present invention, based on the results of the personalized fit assessment, a comprehensive scheduling priority score is dynamically calculated for each consultant candidate, including: Read the real-time maintained consultant workload status table to obtain the number of unfinished consultation tasks and the total estimated duration of each consultant candidate. A basic score for professional matching is calculated based on the degree of matching between the consultant candidate's area of ​​expertise and the current core issues in the instantaneous status summary. Based on the number of unfinished consultation tasks currently assigned to the consultant candidate and the total estimated duration of the tasks, combined with their standard workload, their current resource load coefficient is calculated, and the professional matching score is reduced accordingly. The direct collaboration fit score or the indirect professional fit score of the consultant candidate is included as a potential user preference factor in the calculation. Based on the real-time status of the consultant's online terminal and the recent average response delay, their immediate response capability is assessed and converted into a response capability coefficient. The comprehensive scheduling priority score of the consultant candidate is obtained by weighting and summing the professional matching score, the resource load coefficient, the user potential preference factor, and the response capability coefficient.

[0011] As a further aspect of the present invention, the step of simulating and generating multiple different consultation task allocation schemes based on the comprehensive scheduling priority score and the system's preset global scheduling constraints includes: Define system state variables, which include the current load of all consultants, the occupancy status of various consulting resources, and the queue of consulting tasks to be scheduled. The decision-making action is defined as assigning a consultation task to a specific consultant and occupying specific consultation resources; The initialization state inversion algorithm starts from the current actual system state and uses the comprehensive scheduling priority score as an important reference weight when allocating tasks, but not as the only basis. In each simulated allocation decision, the algorithm selects tasks from the currently scheduled tasks according to their urgency, and selects consultants from consultant candidates who meet the constraints according to a certain probability distribution based on the global scheduling constraints. The probability distribution is positively correlated with the comprehensive scheduling priority score of the consultant candidates. For each simulated allocation decision made, the simulated system state variables are updated once; Repeat the simulation allocation decision until all currently scheduled consultation tasks have been allocated, forming a complete simulation scheduling trajectory, i.e., a consultation task allocation scheme. By changing the random seed or probability selection strategy in the state inversion algorithm, multiple different simulated scheduling trajectories can be generated, thereby obtaining multiple different consultation task allocation schemes.

[0012] As a further aspect of the present invention, for each consultation task allocation scheme, a scheduling effect prediction model trained based on historical data is used to deduce the system state changes and task completion quality indicators generated after its execution, including: Each task allocation decision in the consultation task allocation scheme is transformed into a set of feature vectors, which include the real-time load changes of the assigned consultants, task urgency, professional matching degree, and user's historical cooperation rating features. The feature vector sequence is input into the scheduling effect prediction model, which is a time-series prediction model based on recurrent neural networks and attention mechanisms. The scheduling effect prediction model processes the feature vector of each task allocation decision in turn and implicitly updates the simulated system state maintained internally. After processing the feature vector sequence of the entire scheme, the scheduling effect prediction model outputs a series of predicted future states, including the predicted average completion time of consultation tasks, the overall load balance of consultants, and the predicted distribution of user satisfaction. Key indicators are extracted from the predicted future states to serve as the inferred task completion quality indicators.

[0013] As a further aspect of the present invention, the step of selecting, from all simulated consultation task allocation schemes, the scheme that optimizes the deduced task completion quality index and satisfies the global scheduling constraints as the final scheduling scheme includes: Assign a plan to each consulting task, and calculate a comprehensive quality assessment value based on the task completion quality indicators derived from it; Check whether each consultation task allocation scheme always meets the global scheduling constraints during the execution process. The global scheduling constraints include the daily consultation duration limit, the hard matching requirement of the consultant's area of ​​expertise, and the latest response time for emergency tasks. Eliminate any consultation task allocation scheme that violates any of the global scheduling constraints described above; Among all remaining consultation task allocation schemes that satisfy the global scheduling constraints, their comprehensive quality assessment values ​​are compared; The consultation task allocation scheme with the highest comprehensive quality assessment value is selected as the final scheduling scheme. If multiple consultation task allocation schemes have the same overall quality assessment value and are all the highest, then the scheme with the better overall consultant load balance will be selected.

[0014] As a further aspect of the present invention, the construction steps of the scheduling effect prediction model include: Collect historical scheduling data, which includes the feature vector sequence of each scheduling event and the subsequent actual system state changes and task completion quality indicators. The feature vector sequence of the scheduling event is composed of the feature vector corresponding to each task allocation decision in the current scheduling task allocation scheme. The feature vector includes the real-time load of the assigned consultant, task urgency, professional matching degree, and user historical cooperation rating features. The historical scheduling data is preprocessed, including normalizing the feature vectors and aligning and padding the sequences to ensure that the input sequence lengths are consistent. A model architecture is constructed, which consists of a bidirectional long short-term memory network layer and a multi-head self-attention mechanism layer. The bidirectional long short-term memory network layer is used to process the input temporal feature vector sequence and capture its long-term dependencies. The multi-head self-attention mechanism layer is used to capture the differential influence weights of different task allocation decisions in the sequence on the final scheduling effect. Define the model training objective. The prediction objective of the scheduling effect prediction model is a multi-dimensional indicator of the future system state, including the average completion time of consultation tasks, the overall load balance of consultants, and the expected value of user satisfaction distribution. The model architecture is trained under supervision using preprocessed historical scheduling data, with the actual recorded future system state indicators as the true values. The mean squared error loss function is used, and the model parameters are optimized through the backpropagation algorithm. During model training, the dataset is divided into training, validation, and test sets. The model's hyperparameters are tuned and early stopping is implemented based on the performance on the validation set. Finally, the model's predictive performance is evaluated on the test set to ensure that the model's ability to extrapolate scheduling effects meets the preset requirements.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By combining historical mental health records to validate the historical effectiveness of potential intervention strategies, and analyzing the past cooperation of counselor candidates, the effectiveness of strategy application and counselor suitability characteristics are simultaneously incorporated into the personalized suitability assessment system. This expands the reference dimensions of the suitability assessment and overcomes the limitations of conducting matching assessments solely on the counselor. The assessment process relies on users' historical counseling data to form a two-way verification logic, ensuring that the suitability assessment results simultaneously align with the actual application scenario of the intervention strategy and the collaborative suitability characteristics of the user and counselor. This reduces matching bias caused by single-dimensional assessments, making the conclusions of the suitability assessment more aligned with the actual needs of users' personalized mental health counseling. The combination of strategy and counselor is more suitable for the user's psychological state and counseling needs, further refining the accuracy of the suitability assessment and effectively improving the comprehensiveness of the assessment basis.

[0016] This system dynamically calculates consultants' comprehensive scheduling priority scores by integrating multiple dimensions such as professional matching, resource load, potential user preferences, and immediate response capabilities. Based on this score, it simulates and generates multiple consultation task allocation schemes. A scheduling effect prediction model trained on historical data is used to predict system state changes and task completion quality indicators after each scheme is executed. By relying on pre-calculation logic to replace the traditional model of directly allocating based on priority, and combining it with the system's preset global scheduling constraints, the system completes scheme selection. This ensures that the determination of scheduling schemes takes into account both overall system operating conditions and task execution quality, reducing the probability of misalignment between resource allocation and task assignment during scheduling. The multi-scheme comparison and effect prediction approach allows scheduling decisions to move beyond a single ranking result, resulting in a higher degree of fit between scheduling schemes and the actual system operating state, optimized task allocation rationality, and further improved adaptability and coordination during scheduling execution. Attached Figure Description

[0017] Figure 1 This is a timing diagram of the intelligent scheduling system for mental health counseling services described in this invention. Figure 2 A flowchart for generating an immediate state summary. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 During system operation, the data integration module is activated. Based on the target user's current consultation session data and their historical mental health records, this module analyzes and processes the data to construct a set of potential intervention strategies suitable for the current situation and a set of available counselor candidates. The fit assessment module is then invoked. This module, combining information recorded in the historical mental health records, performs a personalized fit assessment of the constructed set of potential intervention strategies and counselor candidates. This assessment includes validating the historical effectiveness of the strategies and analyzing the counselor's past cooperation. The priority calculation module dynamically calculates a comprehensive scheduling priority score for each counselor candidate based on the results of the personalized fit assessment. This score integrates information from multiple dimensions, including professional matching, resource availability, user potential preferences, and immediate response capabilities. The scheme optimization module begins its work. Based on the calculated comprehensive scheduling priority score and the system's preset global scheduling constraints, this module simulates and generates multiple different consultation task allocation schemes. For each generated consultation task allocation scheme, the system uses a scheduling effect prediction model pre-trained based on historical data to deduce the possible system state changes and task completion quality indicators after the scheme is executed. The scheme optimization module selects the scheme from all the simulated consultation task allocation schemes that optimizes the deduced task completion quality indicators and simultaneously satisfies all global scheduling constraints, and determines this scheme as the final scheduling scheme for output and execution.

[0021] In one embodiment of the invention, the data integration module acquires the target user's current consultation session data and historical mental health records, whereby the historical mental health records store the user's previous mental state assessment results, consultation topic records, and intervention feedback. The module performs real-time semantic parsing and emotional feature extraction on the current consultation session data to generate an immediate state summary, which includes the current core issue, emotional intensity, and urgency tag. Then, based on the generated immediate state summary, the system performs a multi-hop association search within a pre-constructed psychological counseling knowledge graph. This psychological counseling knowledge graph contains rich relationships between nodes such as symptom descriptions, psychological theories, intervention techniques, counselor expertise areas, and counseling resource availability. Through multi-hop association search, the system filters entries from the search results of the psychological counseling knowledge graph that match the current core issue and emotional intensity identified in the immediate state summary. These entries constitute a potential intervention strategy set and a counselor candidate set, respectively.

[0022] In practice, the data integration module constructs a set of potential intervention strategies and a set of counselor candidates based on the target user's current consultation session data and historical mental health records. Current consultation session data is not limited to real-time dialogue; its data sources extend to include: instant text, voice, or video communication requests initiated by the user on the platform; brief needs descriptions filled out by the user when scheduling a consultation; the user's emotional rapid test check-in results on a smart screen terminal; and questionnaire data answered by the user in a psychological assessment system. An example scenario involves an individual identified as "U123." Their historical mental health record documents the psychological state assessment results, consultation topics, and intervention feedback for "U123" in the past six consultations. For example, the topic records include "work stress management" and "coping with social anxiety," with corresponding intervention feedback scores of 4.2 and 3.8, respectively. The current consultation session data includes a 15-minute mixed text and voice interaction. The text stream contains statements such as "I feel the project deadline is approaching, I am completely unable to concentrate, and I have developed unnecessary resentment towards my colleagues," while the voice stream is the corresponding audio recording. In some embodiments, real-time semantic parsing and emotion feature extraction are performed on the current consultation session data to generate an immediate state summary containing the current core issue, emotion intensity, and urgency tags. The system applies dependency parsing and named entity recognition to the text stream in the current consultation session data, identifying the core predicate verbs as "feel" and "unable to concentrate," the emotion carriers as "project deadline" and "colleague," and the key event noun phrase as "unnecessary resistance." Acoustic feature analysis is performed on the speech stream in the current consultation session data, extracting features such as a rising fundamental frequency profile followed by a sharp drop, a speech rate that is stable in the early stages and accelerates later, and energy fluctuations showing a significant increase in energy in the middle segment. These acoustic features are quantified into a speech emotion vector [-0.2, 0.7, 0.1]. The system performs feature-level fusion of the identified text entities and the speech emotion vector, mapping them to a unified psychological state semantic space through a deep neural network model to obtain a fused psychological state vector of the user's statement [0.65, -0.3, 0.4]. Temporal clustering analysis was performed on multiple fused mental state vectors generated within a continuous time window. The dominant mental state category during the conversation was identified as "anxiety accompanied by interpersonal sensitivity," and this category was marked as the current core issue. The mean values ​​of all fused mental state vectors under the "anxiety accompanied by interpersonal sensitivity" category were calculated in terms of valence and arousal, serving as a quantitative representation of the intensity of emotional tendency. The calculation formula is as follows:

[0023] in: Represents the intensity vector of emotional tendency. The average value representing the valence dimension. This represents the average value of the arousal dimension. Based on the current core issue's suddenness assessment as "moderate," the arousal value in the emotional tendency intensity vector being relatively high, and the presence of the pre-set risk keyword "completely unable" in the conversation text, a rule engine automatically generates a urgency label of "high."

[0024] In some embodiments, multi-hop association retrieval is performed in a pre-constructed psychological counseling knowledge graph based on an immediate state summary. The system uses the current core issue "anxiety accompanied by interpersonal sensitivity" and the calculated emotional tendency intensity vector as a reference. As the initial query node, the system locates the corresponding symptom description node "Generalized Anxiety" and the theoretical concept node "Cognitive Distortion" in the psychological counseling knowledge graph. Starting from the located "Generalized Anxiety" and "Cognitive Distortion" nodes, a breadth-first traversal is performed along the relational edges defined in the psychological counseling knowledge graph, with a preset path length of 3 hops. During the traversal, the system collects all nodes with a path length within the 3-hop range, including intervention technique nodes "Cognitive Behavioral Therapy (CBT)" and "Mindfulness-Based Stress Reduction (MBSR)," common comorbid symptom nodes "Sleep Disorders," counselor specialty nodes "Anxiety Disorder Intervention" and "Stress Management," and currently available counseling resource nodes "Video Counseling Room A" and "Next Monday Afternoon Available." All collected nodes are sorted according to their semantic relevance to the initial query nodes "Generalized Anxiety" and "Cognitive Distortion," the number of path hops, and the weight of the relational edges. The system filters out a subset of top-ranked nodes to form preliminary search results for a potential intervention strategy set and a set of counselor candidates. The potential intervention strategy set initially includes {"Cognitive Behavioral Therapy (CBT)", "Mindfulness-Based Stress Reduction (MBSR)"}, and the counselor candidate set initially includes {"Counselor P_Specialty: Anxiety Disorder Intervention", "Counselor Q_Specialty: Stress Management"}. The system obtains the counselor's dynamic status in real time from the counselor management backend. This status includes at least: online status, current workload, and currently available service time slots obtained from the intelligent scheduling system. This status information, along with the counselor's area of ​​expertise and qualification tags, constitutes the real-time attributes of the counselor node in the knowledge graph, used for subsequent evaluation and calculation. Optionally, in the process of generating the instant status summary, the deep neural network model uses a pre-trained Transformer encoder structure to achieve feature-level fusion, concatenating text entity feature vectors and speech emotion vectors before inputting them into the model. The model outputs a fused psychological state vector with a dimension of 256. It is understandable that temporal clustering analysis uses the DBSCAN algorithm, which clusters based on the Euclidean distance between the fused mental state vectors in the semantic space, and determines the category represented by the centroid of the cluster with the largest number of samples as the dominant mental state category.

[0025] In one embodiment of the present invention, when generating an immediate status summary, the system applies dependency parsing and named entity recognition to the text stream in the current consultation session data to identify the core predicate verbs, the emotional carriers, and key event noun phrases. See also Figure 2 The system performs acoustic feature analysis on the speech stream in the current consultation session data, extracting fundamental frequency contours, speech rate variations, and energy fluctuations, and quantifying them into a speech emotion vector. The system then fuses the identified text entities with the speech emotion vector at the feature level, mapping them to a unified psychological state semantic space through a deep neural network model, thus obtaining a fused psychological state vector for the user's statements. The system performs temporal clustering analysis on continuous fused psychological state vectors, identifying the dominant psychological state category throughout the session and marking this category as the current core issue. The system calculates the average value of all fused psychological state vectors in terms of valence and arousal under the current core issue, using this average value as a quantitative representation of the emotional tendency intensity. Based on the suddenness of the current core issue, the extreme degree of emotional tendency intensity, and the presence of pre-set risk keywords in the session, the system automatically generates urgency tags through a rule engine. During multi-hop association retrieval, the system uses the current core issue and emotional tendency intensity as initial query nodes, locating their corresponding symptom description nodes or theoretical concept nodes in the psychological counseling knowledge graph. Starting from these located nodes, the system performs a breadth-first traversal along the relational edges defined in the knowledge graph. During the traversal, the system collects all nodes whose path lengths are within a preset range. These nodes include intervention technique nodes, common comorbid symptom nodes, nodes representing the expertise of counselors specializing in these issues, and nodes representing currently available counseling resources. The system then ranks all collected nodes based on their semantic relevance to the initial query node, path hop count, and relational edge weights, selecting a top-ranked subset of nodes. This subset forms the initial retrieval results for the potential intervention strategy set and the counselor candidate set.

[0026] In the specific implementation, the current consultation session involving user identifier "U456" is considered. The current consultation session data contains a 20-minute interaction, with the text stream including the user-inputted statement "Lately, I've lost interest in everything, I feel like a failure, even eating is a hassle," and the speech stream being the corresponding audio recording. Dependency parsing and named entity recognition are applied to the text stream in the current consultation session data, identifying the core predicate verbs as "lost interest" and "feel," the emotion carriers as "interest" and "self," and the key event noun phrase as "eating is a hassle." Acoustic feature analysis is performed on the speech stream in the current consultation session data, extracting a fundamental frequency profile that shows a flattening trend, a speech rate that is generally slow and has small fluctuations, and energy fluctuation features that show an overall low energy value. These features are quantified into a speech emotion vector [0.4, 0.6, 0.8]. The system performs feature-level fusion of the identified text entities "interests," "self," and "difficulty eating" with the voice emotion vector [0.4, 0.6, 0.8], mapping them to a unified psychological state semantic space through a deep neural network model to obtain the fused psychological state vector [0.7, -0.2, 0.5, 0.1] for each user's single statement. Temporal clustering analysis is performed on the 15 consecutively generated fused psychological state vectors in the conversation to identify the dominant psychological state category cluster. The semantic label corresponding to the cluster centroid is "depressive tendency," and the system marks "depressive tendency" as the current core issue. The mean values ​​of all fused psychological state vectors under the "depressive tendency" category are calculated in terms of valence and arousal dimensions. The mean value for valence is -0.65, and the mean value for arousal is 0.15. The quantitative representation of the emotion tendency intensity is (-0.65, 0.15). Based on the current core issue of "depressive tendency" being assessed as "high" in terms of suddenness, the valence value of the emotional tendency intensity being -0.65, which is extremely negative, and the pre-defined risk keyword "life is meaningless" being identified in the conversation text, an urgency label of "high" is automatically generated through a rule engine.

[0027] In some embodiments, the specific process of performing multi-hop association retrieval in a pre-constructed psychological counseling knowledge graph based on immediate state summaries is as follows. The system uses the current core issue "depressive tendency" and the emotional tendency intensity vector (-0.65, 0.15) as the initial query node, locating the corresponding symptom description node "depressive state" and theoretical concept node "feeling worthless" in the psychological counseling knowledge graph. Starting from the located "depressive state" and "feeling worthless" nodes, a breadth-first traversal is performed along the edges defined in the psychological counseling knowledge graph as "belongs to," "associates with," "treatment methods," and "expertise," with a preset path length range of 2 hops. During the traversal, the system collects all nodes within the 2-hop path length range, including intervention technique nodes "behavioral activation therapy" and "interpersonal psychotherapy (IPT)," common comorbid symptom nodes "excessive sleep" and "loss of appetite," counselor expertise nodes "mood disorder intervention" and "motivational interviewing" for handling such issues, and currently available counseling resource nodes "Counselor M_Wednesday afternoon free" and "Group therapy room B." All collected nodes are sorted according to their semantic relevance to the initial query nodes "depressive state" and "feeling worthless", the number of path hops, and the weight of relation edges. An example of the comprehensive scoring function used for sorting is as follows:

[0028] in: The overall score representing the node. The cosine similarity represents the semantic relevance between the node and the query node. This represents the number of hops in the path from the queried node to the current node. This represents the average weight of the relation edges along the path. , , These are preset weighting coefficients. The system selects a subset of nodes with the top 10 comprehensive scores, forming preliminary search results for a potential intervention strategy set and a set of counselor candidates. For example, in the preliminary search results, the intervention technique node "behavioral activation therapy" has a path hop count of 1, an edge weight of 0.9, and a semantic relevance of 0.88; the counselor expertise node "mood disorder intervention" has a path hop count of 2, an edge weight of 0.8, and a semantic relevance of 0.85. Optionally, the deep neural network model uses a three-layer fully connected network to achieve feature-level fusion. The network input layer receives the concatenated text feature vector and voice emotion vector, the hidden layer uses the ReLU activation function, and the output layer outputs a fused psychological state vector with a dimension of 128. It can be understood that the temporal clustering analysis uses the K-means algorithm to divide the continuous fused psychological state vector into 3 clusters, and the preset psychological state category corresponding to the centroid of the cluster containing the most samples is determined as the dominant psychological state category.

[0029] In one embodiment of the present invention, during the personalized fit assessment stage, the system extracts from the user's historical mental health records the intervention techniques and feedback scores they have previously used for similar current core issues, forming a historical intervention effectiveness list. From the same record, the system extracts the identifiers of counselors the user has previously served, along with their corresponding counseling effectiveness evaluations and user satisfaction scores, forming a historical collaboration evaluation list. The system matches each intervention technique in the potential intervention strategy set with records in the historical intervention effectiveness list, calculates the historical average effectiveness score for that technique, and uses this score as the historical effectiveness verification score for that strategy. The system matches each counselor in the counselor candidate set with records in the historical collaboration evaluation list. If a direct historical collaboration record exists, the corresponding past counseling effectiveness evaluations and user satisfaction scores are extracted and weighted to obtain the counselor's direct collaboration fit score. For counselor candidates without a direct historical collaboration record, the system calculates an indirect professional fit score based on the degree of alignment between their area of ​​expertise and the long-term issues reflected in the user's historical record. When calculating the comprehensive scheduling priority score, the system reads the real-time maintained consultant workload status table to obtain the number of unfinished consultation tasks currently assigned to each consultant candidate and the total estimated duration of these tasks. The system calculates a base score for professional matching based on the match between the consultant candidate's area of ​​expertise and the current core issues in the real-time status summary. The system calculates the consultant candidate's current resource load coefficient based on the number of unfinished consultation tasks currently assigned to them and the total estimated duration of these tasks, combined with their standard workload, and reduces the base score for professional matching accordingly. The system incorporates the consultant candidate's direct collaboration fit score or indirect professional fit score as a potential user preference factor into the calculation. The system assesses the consultant's immediate response capability based on the real-time online status of their terminal and recent average response latency data, and converts this capability into a response capability coefficient. The system then performs a weighted sum of the base score for professional matching, the resource load coefficient, the potential user preference factor, and the response capability coefficient to finally obtain the consultant candidate's comprehensive scheduling priority score.

[0030] In the specific implementation, the target user is identified as "U789". This user's historical mental health record includes three previous consultations on the issue of "social anxiety". The feedback score for the intervention technique "exposure therapy" was 4.0, and the feedback score for the intervention technique "cognitive restructuring" was 4.5. In the immediate status summary generated from the current consultation session data, the current core issue is "social anxiety", with an emotional tendency intensity of (-0.6, 0.7). The potential intervention strategy set constructed by the data integration module includes {"Cognitive Behavioral Therapy (CBT)", "Systematic Desensitization"}, and the counselor candidate set includes {Counselor A_Specialty: Anxiety Disorders, Counselor B_Specialty: Interpersonal Relationships, Counselor C_Specialty: Stress Management}. The fit assessment module, combined with the historical mental health record, conducts a personalized fit assessment. This module extracts from the historical mental health record the intervention techniques and feedback scores that user "U789" has previously used for similar issues like the current core issue "social anxiety", including "exposure therapy: 4.0" and "cognitive restructuring: 4.5", forming a list of historical intervention effectiveness. The adaptation assessment module extracts the identifiers of the counselors previously served by user "U789" from the historical mental health records, along with their corresponding counseling effectiveness evaluations and user satisfaction scores, and records them as "Counselor A: Effectiveness evaluation 4.2 points, User satisfaction 4.3 points", thus forming a historical cooperation evaluation list.

[0031] The adaptation assessment module matches each intervention technique in the potential intervention strategy set with records in the historical intervention effectiveness list, comparing the semantic similarity of the intervention techniques. For the potential intervention strategy "Cognitive Behavioral Therapy (CBT)," "Cognitive Restructuring" was found to be one of its core components in the historical intervention effectiveness list, showing high semantic similarity. The historical score of "Cognitive Restructuring" was calculated as 4.5 points, serving as the historical effectiveness verification score for "CBT" at 4.5 points. For the potential intervention strategy "Systematic Desensitization," no direct matching record was found in the historical intervention effectiveness list, but "Exposure Therapy" was found to belong to the same behavioral therapy category, showing moderate semantic similarity. The historical score of "Exposure Therapy" was calculated as 4.0 points, serving as the historical effectiveness verification score for "Systematic Desensitization" at 4.0 points. The matching assessment module matches each consultant in the pool of candidates with records in the historical collaboration evaluation list. For consultant candidate A, there is a direct historical collaboration record in the historical collaboration evaluation list. Their past consultation effectiveness evaluation score of 4.2 and user satisfaction score of 4.3 are extracted and weighted for calculation, with effectiveness evaluation weighted at 0.6 and user satisfaction weighted at 0.4, resulting in a direct collaboration matching score of 4.2 + 0.6 + 4.3 + 0.4 = 4.24. For consultant candidates B and C, there is no direct historical collaboration record in the historical collaboration evaluation list. The matching assessment module calculates the matching score based on the degree of consistency between consultant candidate B's expertise in "interpersonal relationships" and the long-term issue "social anxiety" reflected in their user history profile. The consistency score is 0.76, and mapping this score to a scale of 1-5 yields an indirect professional matching score of 3.8. Consultant candidate C's expertise in "stress management" has a consistency score of 0.5 with "social anxiety," resulting in an indirect professional matching score of 2.5 after mapping.

[0032] In some embodiments, when the priority calculation module calculates the base score for professional matching, the matching score is calculated based on the semantic distance between the consultant's area of ​​expertise and the current core issue within a predefined knowledge graph. This semantic distance is converted into a score of 1-5 using the shortest path hop count between nodes. This can be understood as a resource load factor. The calculation formula is:

[0033] in: This refers to the number of uncompleted consultation tasks. It is a weighting factor for the average estimated duration of a single task. This is the total estimated duration of the task. This refers to the standard daily workload. In practical implementation, the weighting coefficients in the comprehensive scheduling priority score calculation formula... , , , The system can be configured through the management interface to suit the different priorities of various scheduling strategies. Before each scheduling calculation, the priority calculation module retrieves the latest real-time data from the consultant workload status table. The consultant workload status table is continuously updated by the backend service every minute.

[0034] In one embodiment of the present invention, when simulating the generation of a consultation task allocation scheme, the system defines a set of system state variables, including the current load of all consultants, the occupancy status of various consultation resources, and the queue of consultation tasks to be scheduled. The system defines the decision action as assigning a consultation task to a specific consultant and occupies consultation resources. The system initializes a state inversion algorithm, which simulates from the current actual system state. The comprehensive scheduling priority score is an important reference weight when allocating tasks, but not the only criterion. In each simulated allocation decision, the algorithm selects from the currently scheduled tasks based on the urgency of the tasks, and selects a consultant from all consultant candidates who meet the constraints according to a probability distribution based on a global scheduling constraint. This probability distribution is positively correlated with the comprehensive scheduling priority score of the consultant candidate. After each simulated allocation decision, the algorithm updates the simulated system state variables. The simulated allocation decision is repeated until all currently scheduled consultation tasks are allocated, thus forming a complete simulated scheduling trajectory, i.e., a consultation task allocation scheme. By changing the random seed or probability selection strategy used in the state inversion algorithm, the system can generate multiple different simulated scheduling trajectories, thus obtaining multiple different consultation task allocation schemes. When extrapolating the effects of these schemes, the system transforms each task allocation decision in the scheme into a set of feature vectors. These feature vectors contain features such as the real-time load changes of the assigned consultants, task urgency, professional matching degree, and user historical cooperation ratings. The system inputs this feature vector sequence into the scheduling effect prediction model, a time-series prediction model built on recurrent neural networks and attention mechanisms. The scheduling effect prediction model processes the feature vector of each task allocation decision in the input sequence sequentially, implicitly updating its internally maintained simulated system state in the process. After processing the feature vector sequence of the entire scheme, the scheduling effect prediction model outputs a series of predicted future state indicators, including the predicted average completion time of consultation tasks, the overall load balance of consultants, and the predicted user satisfaction distribution. The system extracts key components from these predicted future state indicators as the extrapolated task completion quality indicators. To achieve an efficient consultation service experience, the system can set "effective connection between user and consultant within 6 seconds" as a key business indicator and performance goal, transforming this goal into a core global scheduling constraint. Under this configuration, when the solution optimization module uses a scheduling effect prediction model to extrapolate the execution effect of each consultation task allocation scheme, it compares the estimated connection establishment time with the 6-second threshold. Only schemes whose extrapolation results meet this time constraint will proceed to the subsequent comparison and final selection stages regarding task completion quality indicators.To support this goal from the outset, the priority calculation module assigns higher weight to the coefficient reflecting immediate responsiveness when dynamically calculating the comprehensive scheduling priority score of consultant candidates. This allows candidates who can respond quickly to occupy a more advantageous position in the ranking. By embedding clear business time objectives into global scheduling constraints and using a scheduling effect prediction model for preliminary feasibility analysis and screening, the system can ensure the overall quality of the scheduling plan while prioritizing its responsiveness requirements for rapid connection.

[0035] The construction of the scheduling effect prediction model is completed through the following steps: First, historical scheduling data is collected, including the feature vector sequence of each scheduling event and its subsequent actual system state changes and task completion quality indicators. The feature vector sequence of a scheduling event consists of the feature vector corresponding to each task allocation decision in the current scheduling task allocation scheme. Second, the historical scheduling data is preprocessed, including normalizing the feature vectors and aligning and padding the sequences to ensure that all input sequences have consistent lengths. Third, the model architecture is constructed, consisting of a bidirectional long short-term memory network layer and a multi-head self-attention mechanism layer. The bidirectional long short-term memory network layer processes the input temporal feature vector sequence and captures its long-term dependencies, while the multi-head self-attention mechanism layer captures the differentiated impact weights of different task allocation decisions on the final scheduling effect. Fourth, the model training objective is defined. The prediction objective of the scheduling effect prediction model is a multi-dimensional indicator of the future system state, including the average completion time of consultation tasks, the overall load balance of consultants, and the expected value of user satisfaction distribution. Supervised training of the constructed model architecture is performed using preprocessed historical scheduling data. Real-world recorded future system state indicators are used as ground truth, and the mean squared error loss function is employed. Model parameters are optimized using the backpropagation algorithm. During model training, the dataset is divided into training, validation, and test sets. Model hyperparameter tuning and early stopping operations are implemented based on performance on the validation set. Finally, the model's predictive performance is evaluated on the test set.

[0036] In practical implementation, the scheme optimization module initializes the state inversion algorithm. The state inversion algorithm simulates the current actual system state. In each simulated allocation decision, the algorithm selects tasks from the current queue of tasks to be scheduled, in descending order of urgency. The first round selects task T1. Based on global scheduling constraints, the algorithm selects a consultant from the constrained consultant candidates according to a probability distribution. For task T1, the candidates satisfying the constraint "high-urgency tasks must be assigned to online consultants within 30 minutes" are consultant Alpha and consultant Beta. Consultant Gamma does not satisfy the immediate response constraint. The state inversion algorithm selects a consultant according to a probability distribution positively correlated with the comprehensive scheduling priority score. The ratio of the probability of selecting consultant Alpha to the probability of selecting consultant Beta is approximately... The state inversion algorithm, based on the result of a random number generator, selected Consultant Alpha in this round of simulation. The algorithm made a simulated allocation decision: "Assign task T1 to Consultant Alpha and occupy resource Video Room 2," updating the simulated system state variables: Consultant Alpha's load increased by 1 hour to 4 hours, the consultation resource "Video Room 2" became occupied, and task T1 was removed from the task queue. The algorithm repeated the simulated allocation decision, processing tasks T2 and T3 sequentially. Each decision was based on the updated simulated system state, global scheduling constraints, and probabilistic selection, ultimately forming a complete simulated scheduling trajectory, i.e., a consultation task allocation scheme. By changing the random seed in the state inversion algorithm, the scheme optimization module generated three different simulated scheduling trajectories, resulting in three different consultation task allocation schemes, as shown in Table 1.

[0037] Table 1: Simulated Consulting Task Allocation Scheme

[0038] In some embodiments, for each consultation task allocation scheme, a scheduling effect prediction model trained based on historical data is used to deduce the system state changes and task completion quality indicators after its execution. The scheme optimization module transforms each task allocation decision in the consultation task allocation scheme into a set of feature vectors. For the decision "assign task T1 to consultant Alpha" in scheme A, its feature vector includes features such as the real-time load change of the assigned consultant Alpha, the urgency of task T1, the professional matching degree between task T1 and consultant Alpha, and the user's historical cooperation rating. The scheme optimization module arranges the feature vectors corresponding to the three decisions of scheme A in the execution order to form a feature vector sequence. The scheme optimization module inputs this feature vector sequence into the scheduling effect prediction model, which is a time-series prediction model based on recurrent neural networks and attention mechanisms, containing a bidirectional long short-term memory network layer and a multi-head self-attention mechanism layer. The scheduling effect prediction model processes each vector in the feature vector sequence sequentially. It captures long-term dependencies in the sequence through a bidirectional long short-term memory network layer and captures the differentiated impact weights of different allocation decisions on the final effect through a multi-head self-attention mechanism layer. Simultaneously, it implicitly updates the simulated system state it maintains internally. After processing the feature vector sequence of the entire scheme, the scheduling effect prediction model outputs a series of predicted future state indicators. These indicators include the predicted average completion time of consultation tasks, the overall load balancing of consultants, and the predicted user satisfaction distribution. The scheme optimization module extracts key indicators from the predicted future states as inferred task completion quality indicators. For example, the inferred task completion quality indicators for scheme A are: predicted average completion time 1.15 hours, predicted load balancing 0.82, and predicted user satisfaction 4.3.

[0039] The construction steps of the scheduling effect prediction model include: collecting historical scheduling data, which includes the feature vector sequence of each scheduling event in the past six months and records of subsequent actual system state changes and task completion quality indicators; preprocessing the historical scheduling data, including normalizing features of different dimensions in the feature vectors and padding the feature vector sequences of historical scheduling events of varying lengths to ensure that the sequence length input to the scheduling effect prediction model is consistent; constructing the model architecture, which consists of a bidirectional long short-term memory network layer and a multi-head self-attention mechanism layer; the bidirectional long short-term memory network layer has 128 hidden units to process the input temporal feature vector sequence and capture the long-term dependencies of the sequence; and the multi-head self-attention mechanism layer has four attention heads to capture the differentiated impact weights of different task allocation decisions on the final scheduling effect; and defining the model training objective, the prediction objective of the scheduling effect prediction model is a multi-dimensional indicator of the future system state, including the average completion time of consultation tasks, the overall load balance of consultants, and the expected value of user satisfaction distribution. Supervised training of the model architecture is performed using preprocessed historical scheduling data. Real-world recorded future system state indicators are used as ground truth, and the mean squared error loss function is employed. Model parameters are optimized using the backpropagation algorithm. During model training, the dataset is divided into training, validation, and test sets in a 7:2:1 ratio. Model hyperparameter tuning and early stopping are implemented based on performance on the validation set. Finally, the model's predictive performance is evaluated on the test set. Optionally, in the state inversion algorithm, when selecting consultants from constrained candidates using probability, the probability distribution used is based on the Softmax function, which converts the comprehensive scheduling priority score into probability. The formula used in the probability selection strategy can be understood as follows:

[0040] in: Representative selects consultant The probability, It is a temperature parameter used to control exploratory activities. Consultant The comprehensive scheduling priority score represents the set of consultant candidates that currently meet the constraints. In practice, the bidirectional long short-term memory network layer of the scheduling effect prediction model is followed by a fully connected output layer, with the number of neurons in the output layer corresponding to the number of multi-dimensional indicators to be predicted. Task completion quality indicators in historical scheduling data are actually recorded and stored by the system after scheduling execution, including the actual time from task allocation to closure, the statistical variance of consultant workload, and the satisfaction ratings given by users in follow-up surveys.

[0041] In one embodiment of the invention, the system calculates a comprehensive quality assessment value for each simulated consultation task allocation scheme based on the derived task completion quality indicators. The system checks whether each consultation task allocation scheme consistently meets the system's preset global scheduling constraints throughout the entire simulation execution process. These global scheduling constraints include, but are not limited to, the daily consultation duration limit, hard matching requirements for consultants' expertise areas, and the latest response time for emergency tasks. The system eliminates any consultation task allocation scheme that violates any global scheduling constraint. Among all remaining consultation task allocation schemes that meet all global scheduling constraints, the system compares their comprehensive quality assessment values. The system selects the consultation task allocation scheme with the highest comprehensive quality assessment value and determines it as the final scheduling scheme. If multiple consultation task allocation schemes have the same and highest comprehensive quality assessment value, the system selects the scheme with the best derived overall consultant load balancing.

[0042] In practice, the solution optimization module checks whether each consultation task allocation scheme consistently meets the system's preset global scheduling constraints during execution. These constraints include a daily consultation duration limit of 8 hours, a hard match requirement between consultants' expertise areas, and a maximum response time of 30 minutes for emergency tasks. The module simulates the allocation sequence of scheme Alpha and finds that consultant Epsilon's total assigned task time in scheme Alpha reaches 8.5 hours, exceeding the 8-hour daily consultation limit. Therefore, scheme Alpha is deemed to violate global scheduling constraints. The module then simulates the allocation sequence of scheme Beta and finds that the total task time for all consultants does not exceed 8 hours. Task Tau is assigned to consultant Zeta, who is "online," 25 minutes after its generation, satisfying the maximum response time constraint for emergency tasks. All allocations meet the hard match requirement between consultants' expertise areas, therefore scheme Beta is deemed to satisfy all global scheduling constraints. The solution optimization module simulates the allocation sequence of solution Gamma and finds that task Upsilon is assigned to counselor Eta, whose specialty is "family therapy," while the core issue of task Upsilon is "job burnout." The matching degree between the two in the system's knowledge graph is below the threshold for a hard match requirement; therefore, solution Gamma is judged to violate the hard match requirement for counselor specialty areas. The solution optimization module then simulates the allocation sequence of solution Delta and confirms that it complies with all global scheduling constraints. The solution optimization module eliminates any counseling task allocation solution that violates any global scheduling constraint; therefore, solutions Alpha and Gamma are eliminated, and solutions Beta and Delta proceed to the next comparison stage.

[0043] In some embodiments, the solution optimization module compares the comprehensive quality assessment values ​​of solution Beta and solution Delta among all remaining consultation task allocation solutions that satisfy the global scheduling constraints. The comprehensive quality assessment value is calculated by standardizing and weighting the various prediction indicators, and the specific formula is as follows:

[0044] in: Represents the overall quality assessment value. This represents the score corresponding to the predicted average completion time, which is obtained by mapping the predicted average completion time to the interval between 0 and 1 through a normalization function. Represents the predicted load balancing degree. Represents predicted user satisfaction. , , This represents the preset weights of each indicator. Assume the normalization function is set as follows:

[0045] in: To predict the average completion time, and The system is set with a minimum and maximum reference duration, for example, 0.5 hours and 2.0 hours respectively. Based on this function, the predicted average completion time of scheme Beta, 1.1 hours, corresponds to a score... The score corresponding to the predicted average completion time of the Delta scheme is 1.05 hours. Assuming the weights are set... , , The predicted load balancing score for solution Beta is 0.78, the predicted user satisfaction score is 4.5, and the calculated overall quality assessment score is... The proposed Delta model has a predicted load balancing score of 0.75 and a predicted user satisfaction score of 4.6, resulting in a calculated overall quality assessment value. The solution optimization module compares the comprehensive quality assessment values ​​of solution Beta and solution Delta. Solution Delta's comprehensive quality assessment value of 1.85832 is higher than that of solution Beta (1.824). The module selects the consulting task allocation scheme with the highest comprehensive quality assessment value as the final scheduling scheme; therefore, solution Delta is selected as the final scheduling scheme. Optionally, the hard matching requirement of consultant expertise in the global scheduling constraints can be implemented using a binary function. A decision is considered to satisfy the constraint if and only if the matching score between the consultant's expertise and the core issue of the task is higher than a preset threshold. This can be understood as a normalization function. The design ensures that its output value is a dimensionless score, consistent with the predicted load balancing. and predicting user satisfaction The dimensions are consistent, thus ensuring that the dimensions of both sides of the formula are consistent. In specific implementation, after the scheme optimization module eliminates schemes that violate the constraints, if the number of remaining schemes is zero, the system relaxes some non-core constraints and re-simulates and re-generates and selects schemes, or triggers manual scheduling intervention.

[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications 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 protection scope of the present invention.

Claims

1. A smart scheduling system for mental health counseling services, characterized in that: include: The data integration module constructs a set of potential intervention strategies and a set of counselor candidates based on the target user's current counseling session data and historical mental health records; The fit assessment module, in conjunction with the historical mental health records, performs a personalized fit assessment of the potential intervention strategy set and the counselor candidate set. The fit assessment includes verification of the historical effectiveness of the strategies and analysis of the counselor's past cooperation. The priority calculation module dynamically calculates a comprehensive scheduling priority score for each consultant candidate based on the results of the personalized suitability assessment. The comprehensive scheduling priority score integrates multiple dimensions such as professional matching degree, resource load, user potential preferences and immediate response capability. The scheme optimization module simulates and generates multiple different consultation task allocation schemes based on the comprehensive scheduling priority score and the system's preset global scheduling constraints. For each consultation task allocation scheme, a scheduling effect prediction model trained based on historical data is used to deduce the system state changes and task completion quality indicators after its execution. From all the simulated consultation task allocation schemes, the scheme that makes the deduced task completion quality indicators optimal and satisfies the global scheduling constraints is selected as the final scheduling scheme.

2. The intelligent scheduling system for mental health counseling services as described in claim 1, characterized in that, The process involves constructing a set of potential intervention strategies and a set of counselor candidates based on the target user's current counseling session data and historical mental health records, including: Obtain the target user's current consultation session data and historical mental health records, which store the user's previous mental health assessment results, consultation topic records, and intervention feedback. Real-time semantic parsing and emotion feature extraction are performed on the current consultation session data to generate an instant status summary containing the current core issue, emotion intensity, and urgency tags; Based on the real-time state summary, multi-hop association retrieval is performed in the pre-constructed psychological counseling knowledge graph, which includes the association relationships between symptom descriptions, psychological theories, intervention techniques, counselor expertise areas, and counseling resource availability nodes. From the search results of the psychological counseling knowledge graph, a set of potential intervention strategies and a set of counselor candidates that match the current core issues and emotional tendencies in the instantaneous state summary are selected.

3. The intelligent scheduling system for mental health counseling services as described in claim 2, characterized in that, Real-time semantic parsing and sentiment feature extraction are performed on the current consultation session data to generate an immediate status summary containing the current core issue, sentiment intensity, and urgency tags, including: Dependency parsing and named entity recognition are applied to the text stream in the current consultation session data to identify core predicate verbs, emotion-carrying objects, and key event noun phrases; Acoustic feature analysis is performed on the speech stream in the current consultation session data to extract fundamental frequency profile, speech rate variation and energy fluctuation features, and these features are quantified into speech emotion vectors. The identified text entities are fused with the speech emotion vector at the feature level, and mapped to a unified psychological state semantic space through a deep neural network model to obtain the fused psychological state vector of the user's statement. Temporal clustering analysis is performed on the continuous fused mental state vectors to identify the dominant mental state categories during the conversation, and these mental state categories are marked as the current core issue. Calculate the mean of the fused psychological state vector in terms of valence and arousal dimension under the current core issue, and use it as a quantitative representation of the intensity of the emotional tendency; Based on the suddenness of the current core issue, the extreme intensity of the emotional tendency, and whether there are preset risk keywords in the conversation, the urgency label is automatically generated by a rule engine.

4. The intelligent scheduling system for mental health counseling services as described in claim 3, characterized in that, Based on the aforementioned real-time state summary, multi-hop association retrieval is performed in the pre-constructed psychological counseling knowledge graph, including: Using the current core issue and the intensity of the emotional tendency as the initial query nodes, locate the corresponding symptom description or theoretical concept node in the psychological counseling knowledge graph; Starting from the located symptom description or theoretical concept node, perform a breadth-first traversal along the relation edges in the knowledge graph; During the traversal, all nodes with path lengths within a preset range are collected, including intervention technique nodes, common comorbid symptom nodes, nodes of counselors' areas of expertise who are good at handling such issues, and currently available counseling resource nodes. All collected nodes are sorted according to their semantic relevance to the initial query node, path hop count, and edge weight. The top-ranked subset of nodes is then selected to form the preliminary search results for the potential intervention strategy set and consultant candidate set.

5. The intelligent scheduling system for mental health counseling services as described in claim 4, characterized in that, Based on the aforementioned historical mental health records, a personalized fit assessment is conducted on the set of potential intervention strategies and the set of counselor candidates, including: From the aforementioned historical mental health records, extract the intervention techniques and feedback scores that users have used in the past to address similar current core issues, and construct a list of historical intervention effectiveness. From the historical mental health records, extract the identifiers of the counselors who have served the user and their corresponding counseling effectiveness evaluations and user satisfaction scores to form a historical cooperation evaluation list. Each intervention technique in the potential intervention strategy set is matched with a record in the historical intervention effectiveness list, and its historical average effectiveness score is calculated as the historical effectiveness verification score of the strategy. Each consultant in the consultant candidate set is matched with the record in the historical cooperation evaluation list. If there is a direct historical cooperation, the past consultation effect evaluation and user satisfaction score are extracted and weighted to obtain the direct cooperation suitability score. For consultants without direct prior collaboration, an indirect professional fit score is calculated based on the alignment between their area of ​​expertise and the long-term issues reflected in the user's historical profile.

6. The intelligent scheduling system for mental health counseling services as described in claim 5, characterized in that, Based on the results of the personalized fit assessment, a comprehensive scheduling priority score is dynamically calculated for each consultant candidate, including: Read the real-time maintained consultant workload status table to obtain the number of unfinished consultation tasks and the total estimated duration of each consultant candidate. A basic score for professional matching is calculated based on the degree of matching between the consultant candidate's area of ​​expertise and the current core issues in the instantaneous status summary. Based on the number of unfinished consultation tasks currently assigned to the consultant candidate and the total estimated duration of the tasks, combined with their standard workload, their current resource load coefficient is calculated, and the professional matching score is reduced accordingly. The direct collaboration fit score or the indirect professional fit score of the consultant candidate is included as a potential user preference factor in the calculation. Based on the real-time status of the consultant's online terminal and the recent average response delay, their immediate response capability is assessed and converted into a response capability coefficient. The comprehensive scheduling priority score of the consultant candidate is obtained by weighting and summing the professional matching score, the resource load coefficient, the user potential preference factor, and the response capability coefficient.

7. The intelligent scheduling system for mental health counseling services as described in claim 6, characterized in that, The process involves simulating and generating multiple different consultation task allocation schemes based on the comprehensive scheduling priority score and the system's preset global scheduling constraints, including: Define system state variables, which include the current load of all consultants, the occupancy status of various consulting resources, and the queue of consulting tasks to be scheduled. The decision-making action is defined as assigning a consultation task to a specific consultant and occupying specific consultation resources; The initialization state inversion algorithm starts from the current actual system state and uses the comprehensive scheduling priority score as an important reference weight when allocating tasks, but not as the only basis. In each simulated allocation decision, the algorithm selects tasks from the currently scheduled tasks according to their urgency, and selects consultants from consultant candidates who meet the constraints according to a certain probability distribution based on the global scheduling constraints. The probability distribution is positively correlated with the comprehensive scheduling priority score of the consultant candidates. For each simulated allocation decision made, the simulated system state variables are updated once; Repeat the simulation allocation decision until all currently scheduled consultation tasks have been allocated, forming a complete simulation scheduling trajectory, i.e., a consultation task allocation scheme. By changing the random seed or probability selection strategy in the state inversion algorithm, multiple different simulated scheduling trajectories can be generated, thereby obtaining multiple different consultation task allocation schemes.

8. The intelligent scheduling system for mental health counseling services as described in claim 7, characterized in that, For each consultation task allocation scheme, a scheduling effect prediction model trained based on historical data is used to deduce the system state changes and task completion quality indicators after its execution, including: Each task allocation decision in the consultation task allocation scheme is transformed into a set of feature vectors, which include the real-time load changes of the assigned consultants, task urgency, professional matching degree, and user's historical cooperation rating features. The feature vector sequence is input into the scheduling effect prediction model, which is a time-series prediction model based on recurrent neural networks and attention mechanisms. The scheduling effect prediction model processes the feature vector of each task allocation decision in turn and implicitly updates the simulated system state maintained internally. After processing the feature vector sequence of the entire scheme, the scheduling effect prediction model outputs a series of predicted future states, including the predicted average completion time of consultation tasks, the overall load balance of consultants, and the predicted distribution of user satisfaction. Key indicators are extracted from the predicted future states to serve as the inferred task completion quality indicators.

9. The intelligent scheduling system for mental health counseling services as described in claim 8, characterized in that, The step of selecting the final scheduling scheme from all simulated consultation task allocation schemes that optimizes the deduced task completion quality index and satisfies the global scheduling constraints includes: Assign a plan to each consulting task, and calculate a comprehensive quality assessment value based on the task completion quality indicators derived from it; Check whether each consultation task allocation scheme always meets the global scheduling constraints during the execution process. The global scheduling constraints include the daily consultation duration limit, the hard matching requirement of the consultant's area of ​​expertise, and the latest response time for emergency tasks. Eliminate any consultation task allocation scheme that violates any of the global scheduling constraints described above; Among all remaining consultation task allocation schemes that satisfy the global scheduling constraints, their comprehensive quality assessment values ​​are compared; The consultation task allocation scheme with the highest comprehensive quality assessment value is selected as the final scheduling scheme. If multiple consultation task allocation schemes have the same overall quality assessment value and are all the highest, then the scheme with the better overall consultant load balance will be selected.

10. The intelligent scheduling system for mental health counseling services as described in claim 9, characterized in that, The steps for constructing the scheduling effect prediction model include: Collect historical scheduling data, which includes the feature vector sequence of each scheduling event and the subsequent actual system state changes and task completion quality indicators. The feature vector sequence of the scheduling event is composed of the feature vector corresponding to each task allocation decision in the current scheduling task allocation scheme. The feature vector includes the real-time load of the assigned consultant, task urgency, professional matching degree, and user historical cooperation rating features. The historical scheduling data is preprocessed, including normalizing the feature vectors and aligning and padding the sequences to ensure that the input sequence lengths are consistent. A model architecture is constructed, which consists of a bidirectional long short-term memory network layer and a multi-head self-attention mechanism layer. The bidirectional long short-term memory network layer is used to process the input temporal feature vector sequence and capture its long-term dependencies. The multi-head self-attention mechanism layer is used to capture the differential influence weights of different task allocation decisions in the sequence on the final scheduling effect. Define the model training objective. The prediction objective of the scheduling effect prediction model is a multi-dimensional indicator of the future system state, including the average completion time of consultation tasks, the overall load balance of consultants, and the expected value of user satisfaction distribution. The model architecture is trained under supervision using preprocessed historical scheduling data, with the actual recorded future system state indicators as the true values. The mean squared error loss function is used, and the model parameters are optimized through the backpropagation algorithm. During model training, the dataset is divided into training, validation, and test sets. The model's hyperparameters are tuned and early stopping is implemented based on the performance on the validation set. Finally, the model's predictive performance is evaluated on the test set to ensure that the model's ability to extrapolate scheduling effects meets the preset requirements.

Images