Dynamic mental health training adaptation method and system

Abstract

This application relates to the field of mental health technology and discloses a method and system for dynamic adaptation in mental health training. The method includes: acquiring initial mental state data, training needs data, training process data, and dialogue interaction data of the target user; performing text preprocessing, context fusion, and psychosemantic recognition on the dialogue interaction data to generate psychosemantic feature data; generating a dynamic mental state vector by combining the initial mental state data, training process data, and psychosemantic feature data; acquiring training target labels, adaptation state labels, and task load labels for each training unit; calculating the training fit degree based on the dynamic mental state vector, training needs data, and training unit labels, and selecting target training units based on the task load labels; generating training progress control parameters and training tasks, sending them to the target user, obtaining execution feedback data, and updating the training process data. This method improves the adaptability of mental health training and the security of process adjustments.

Description

Technical Field

[0001] This application relates to the field of mental health technology, and in particular to a dynamic adaptation method and system for mental health training. Background Technology

[0002] With the development of online mental health services, mental health training is increasingly being applied to scenarios such as emotion regulation, stress management, cognitive improvement, behavioral activation, and sleep improvement. Traditional mental health training methods typically employ pre-set courses, fixed tasks, and a uniform pace, with users completing check-ins, exercises, feedback, and periodic assessments according to a predetermined process. While this approach can meet basic mental health training needs to some extent, different users vary in their sources of stress, emotional states, cognitive habits, behavioral motivation, sleep patterns, training goals, and risk levels. A fixed training content and pace are difficult to adapt to the actual conditions of different users. During training, user states may change based on task completion, life events, and subjective feedback. If a uniform process is continued, problems such as training content not matching current needs, unreasonable training intensity, continued training despite insufficient user understanding, or low-motivation users failing to persist in completing the training can easily arise. Furthermore, traditional training effectiveness relies heavily on periodic scale results or user subjective feedback, with insufficient tracking of emotional changes, behavioral improvements, and completion quality during the training process. This leads to untimely adjustments to the training plan and difficulty in continuously verifying training effects. Therefore, traditional mental health training methods still need improvement in terms of individual adaptability, timeliness of process adjustment, and effectiveness tracking. Summary of the Invention

[0003] The implementation method of this application mainly addresses the technical problems of insufficient training content adaptation, delayed adjustment of training process, and discontinuous tracking of training effect in traditional mental health training techniques.

[0004] To address the aforementioned technical problems, the first technical solution adopted in this application is: providing a dynamic adaptation method for mental health training, comprising: acquiring initial mental state data, training demand data, training process data, and dialogue interaction data of a target user; performing text preprocessing and context fusion on the dialogue interaction data to obtain a dialogue semantic vector, and inputting the dialogue semantic vector into a mental semantic recognition model to generate mental semantic feature data; generating a dynamic mental state vector of the target user based on the initial mental state data, the training process data, and the mental semantic feature data; acquiring training unit label data for each training unit in a mental health training library, wherein the training unit label data includes training target labels, adaptation state labels, and task load labels; and generating a dynamic mental state vector based on the dynamic mental state vector. The system calculates the training fit between the target user and each training unit based on the training requirement data, the training target label, and the fit status label of each training unit; it then constrains the task load of each training unit based on the training fit and the task load label of each training unit, and filters target training units; it generates training advancement control parameters based on the dynamic psychological state vector and the task load label of the target training unit, and these parameters are used to control the advancement, pause, repetition, degradation, or replacement of the target training unit; it generates a training task based on the target training unit and the training advancement control parameters, sends the training task to the target user, obtains the execution feedback data corresponding to the training task, and updates the training process data based on the execution feedback data.

[0005] To address the aforementioned technical problems, the second technical solution adopted in this application is: providing a dynamic adaptation system for mental health training, comprising: a state data acquisition module, used to acquire initial mental state data, training requirement data, training process data, and dialogue interaction data of a target user; a psychological semantic generation module, used to perform text preprocessing and context fusion on the dialogue interaction data to obtain a dialogue semantic vector, and input the dialogue semantic vector into a psychological semantic recognition model to generate psychological semantic feature data; a state vector generation module, used to generate a dynamic mental state vector of the target user based on the initial mental state data, the training process data, and the psychological semantic feature data; a training label acquisition module, used to acquire training unit label data of each training unit in a mental health training library, the training unit label data including training target labels, adaptation state labels, and task load labels; and an adaptation score calculation module, used to... Based on the dynamic mental state vector, the training requirement data, the training target labels and adaptation status labels of each training unit, the training fit degree between the target user and each training unit is calculated; the target unit screening module is used to perform task load constraints on each training unit according to the training fit degree and the task load labels of each training unit, and screen target training units; the advancement parameter generation module is used to generate training advancement control parameters based on the dynamic mental state vector and the task load labels of the target training units, and the training advancement control parameters are used to control the advancement, pause, repetition, degradation or replacement of the target training units; the training task processing module is used to generate training tasks based on the target training units and the training advancement control parameters, send the training tasks to the target user, obtain the execution feedback data corresponding to the training tasks, and update the training process data based on the execution feedback data.

[0006] To solve the above-mentioned technical problems, the third technical solution adopted in the embodiments of this application is: to provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the dynamic adaptation method for mental health training as described above.

[0007] To solve the above-mentioned technical problems, the fourth technical solution adopted in the embodiments of this application is: to provide a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by an electronic device, the electronic device performs the mental health training dynamic adaptation method as described above.

[0008] Unlike related technologies, this application acquires the target user's initial psychological state data, training needs data, training process data, and dialogue interaction data, and performs psycho-semantic recognition on the dialogue interaction data. This allows pre-training assessment data, training process data, and dialogue semantic data to jointly participate in the generation of a dynamic psychological state vector. Therefore, the target user's psychological state no longer relies solely on the initial scale or a single feedback, but is continuously updated based on emotional changes, task completion status, user feedback, and risk signals. Based on this, this application calculates training fit according to the dynamic psychological state vector, training needs data, and training unit label data, and selects target training units by combining task load labels. Then, it controls the advancement, pause, repetition, downgrading, or replacement of training units according to training progress control parameters. This reduces problems such as content mismatch, unreasonable load, and delayed adjustments caused by fixed training content and uniform progress rhythm, improving individual adaptability and timely process adjustment in mental health training. Attached Figure Description

[0009] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0010] Figure 1 This is a schematic diagram of the operating environment of the dynamic adaptation method for mental health training provided in the embodiments of this application.

[0011] Figure 2 This is a schematic diagram of the execution flow of the dynamic adaptation method for mental health training provided in the embodiments of this application.

[0012] Figure 3 This is a schematic diagram of the execution flow for generating dynamic psychological state vectors in the dynamic adaptation method for mental health training provided in this application embodiment.

[0013] Figure 4 This is a schematic diagram of the execution flow for generating training advancement control parameters in the dynamic adaptation method for mental health training provided in this application embodiment.

[0014] Figure 5 This is a schematic diagram of the system structure of the dynamic adaptation system for mental health training provided in this application embodiment.

[0015] Figure 6 This is a schematic diagram of the hardware structure of an electronic device for implementing the dynamic adaptation method for mental health training provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Software tools, components, or servers not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0017] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the system diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the system diagram or the order in the flowchart.

[0018] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0019] To facilitate understanding of this embodiment, a detailed description of the dynamic adaptation method for mental health training disclosed in this application will be provided first. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a schematic diagram of the operating environment of the dynamic adaptation method for mental health training provided in the embodiments of this application, such as... Figure 1 As shown, the execution subject of the dynamic adaptation method for mental health training provided in this application is generally an electronic device with a certain computing power, such as a computer. In some possible implementations, this dynamic adaptation method for mental health training can be implemented by a processor calling computer-readable instructions stored in memory. Figure 1 The computer equipment mentioned can be a server. A server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. This can be understood as... Figure 1 The number of computer devices shown is merely illustrative and can be expanded in any number according to actual needs.

[0020] Please continue reading. Figure 2 , Figure 2This is a schematic diagram of the execution flow of the dynamic adaptation method for mental health training provided in the embodiments of this application, such as... Figure 2 As shown, it includes the following steps: S1. Obtain the target user's initial psychological state data, training needs data, training process data, and dialogue interaction data.

[0021] Step S1 is the data access and state initialization step in the dynamic adaptation process of mental health training, which provides basic data for subsequent psychological semantic recognition, dynamic mental state vector generation, training unit matching, and training progress control. Initial mental state data may include data such as psychological scale data, initial stress level, initial emotional state, initial sleep state, initial risk level, and initial psychological profile labels formed by the target user before the start of training; training demand data may include data such as the target user's training goals, main distress scenarios, psychological problems they hope to improve, training acceptance level, and phased training needs; training process data may include data such as the target user's emotion score, sleep duration, training task completion status, training task completion rate, AI dialogue rounds, user feedback data, task time, and training interruption status during the training process; dialogue interaction data may include natural language text data generated by the target user in the initial AI interview, training process dialogue, training feedback dialogue, and phased review dialogue. In practice, the system collects initial psychological state data and training needs data when the target user enters the mental health training process to establish a baseline for the user's initial state. During training, it continuously collects training process data and dialogue interaction data to reflect the target user's emotional changes, behavioral performance, training participation level, and subjective feedback at different training time points. Step S1 integrates pre-training assessment data, in-training behavioral data, and AI dialogue data into a unified data processing chain, forming a user state data set that can be used for subsequent calculations. This provides a multi-source data foundation for dynamic psychological state calculation, training unit adaptation, and training path adjustment, improving the accuracy of subsequent training content matching and training progress control.

[0022] S2. Perform text preprocessing and context fusion on the dialogue interaction data to obtain dialogue semantic vectors, and input the dialogue semantic vectors into the psycho-semantic recognition model to generate psycho-semantic feature data.

[0023] Step S2 is the dialogue semantic structuring step, used to convert the natural language content generated by the target user during AI interviews, training dialogues, and training feedback into computable psychosemantic features. Text preprocessing may include word segmentation, noise reduction, text standardization, invalid character filtering, and expression normalization. Context fusion may involve acquiring historical dialogue texts prior to the current dialogue round and concatenating or associating the current dialogue text with historical dialogue texts to preserve the semantic changes and contextual dependencies of the target user during continuous dialogue. The dialogue semantic vector can be a vectorized representation of the contextual dialogue text after semantic encoding. The psychosemantic recognition model can be used to identify emotional states, cognitive patterns, behavioral tendencies, training comprehension levels, and risk signals, and generate psychosemantic feature data. Through step S2, unstructured dialogue text can be transformed into structured psychosemantic data, providing semantic dimension input for subsequent dynamic psychological state calculations and improving the detail of user state recognition.

[0024] As an optional implementation, the process of generating psycho-semantic feature data in step S2 above may also specifically include the following steps S21 to S25.

[0025] S21. Perform word segmentation, noise reduction, and text standardization on the dialogue interaction data to obtain the current dialogue text.

[0026] Step S21 involves cleaning and standardizing the dialogue interaction data collected in the current dialogue round. The dialogue interaction data may include natural language text input by the target user during AI interviews, training process dialogues, or training feedback dialogues; word segmentation is used to divide continuous text into words, phrases, or semantic segments; noise reduction is used to remove invalid characters, repetitive expressions, abnormal symbols, meaningless stop words, and interfering text unrelated to psycho-semantic recognition; and text standardization is used to uniformly represent synonymous expressions, colloquial expressions, abbreviations, and emotional expressions, resulting in current dialogue text that facilitates subsequent context fusion and semantic encoding.

[0027] S22. Obtain the target user's historical dialogue text before the current dialogue round, and concatenate the current dialogue text with the historical dialogue text to obtain the context dialogue text.

[0028] In step S22, the current dialogue text is associated with the historical dialogue text generated by the target user before the current dialogue round. The historical dialogue text may include the text content generated by the target user in the initial interview, the preceding training dialogue, the training feedback dialogue, and the phase review dialogue; the context splicing can be performed by splicing the historical dialogue text with the current dialogue text according to the dialogue time order, the dialogue round order, or the training phase order, or the time markers, training task markers, or user feedback markers between different dialogue rounds can be retained to obtain the contextual dialogue text containing contextual semantic relationships.

[0029] S23. Semantically encode the contextual dialogue text to obtain the dialogue semantic vector.

[0030] Step S23 converts the contextual dialogue text into a vectorized semantic representation. Semantic encoding can employ word vector encoding, sentence vector encoding, contextual semantic encoding, or neural network-based text encoding to vectorize emotional expressions, cognitive expressions, behavioral expressions, training feedback expressions, and risk-related expressions in the contextual dialogue text. The dialogue semantic vector represents the comprehensive semantic state of the target user in the current dialogue turn and its historical context, and serves as input data for the psycho-semantic recognition model.

[0031] In one implementation, the target user can be in the first... t The training time point, the first r The input text in a round-robin dialogue is denoted as It also retrieves the historical dialogue text from previous dialogue rounds to form the context dialogue text. .in, This can include the current dialogue text and one or more historical dialogue texts from previous dialogue rounds, for example: .

[0032] in, h This represents the number of previous historical dialogue rounds selected. Furthermore, the contextual dialogue text can be input into a text encoding model to obtain a dialogue semantic vector: .

[0033] in, Encoder Represents a text encoding model. Indicates the target user in the 1st month. t The training time point, the first r The dialogue semantic vector corresponds to each round of dialogue. In this way, the dialogue semantic vector not only contains the semantic information of the current input text, but also includes emotional changes, training feedback, and risk expressions in the historical dialogue, enabling subsequent psycho-semantic recognition to make judgments based on continuous context, rather than just based on a single round of text.

[0034] S24. Input the dialogue semantic vector into the psycho-semantic recognition model to generate emotional state recognition results, cognitive pattern recognition results, behavioral tendency recognition results, training understanding recognition results, and risk signal recognition results.

[0035] In step S24, the dialogue semantic vector is classified and its state is identified using a psychosemantic recognition model. The emotional state recognition result can be used to indicate the target user's current emotional state, such as anxiety, depression, stability, tension, or positivity; the cognitive pattern recognition result can be used to indicate whether the target user exhibits cognitive patterns such as negative interpretation, catastrophic thinking, self-denial, or avoidance cognition; the behavioral tendency recognition result can be used to indicate the target user's willingness to participate in training, task execution tendency, avoidance tendency, or help-seeking tendency; the training comprehension recognition result can be used to indicate the target user's level of understanding of the current training content; and the risk signal recognition result can be used to indicate whether there are signals such as high stress, strong negative emotions, risk expression, or abnormal feedback in the dialogue text.

[0036] S25. Based on the results of emotion state recognition, cognitive pattern recognition, behavioral tendency recognition, training understanding recognition, and risk signal recognition, generate psycho-semantic feature data.

[0037] In step S25, the multiple recognition results output by the psycho-semantic recognition model are structured and organized to form psycho-semantic feature data that can be used in the subsequent dynamic psychological state vector generation process. The psycho-semantic feature data may include emotional state labels, cognitive pattern labels, behavioral tendency labels, training comprehension level values, risk signal strength, stress semantic strength, and cognitive bias state, etc. The emotional state recognition results, cognitive pattern recognition results, behavioral tendency recognition results, training comprehension recognition results, and risk signal recognition results can be converted into corresponding label values, level values, confidence values, or intensity values, and written into the psycho-semantic feature data according to the preset data structure.

[0038] Through steps S21 to S25, the scattered, colloquial, and context-dependent text content of continuous dialogues with target users can be transformed into structured psychosemantic feature data. This makes psychological state information such as emotional state, cognitive patterns, behavioral tendencies, training comprehension level, and risk signals computable and accessible. Compared to methods that rely solely on single-turn dialogues for judgment, this process can reduce semantic misjudgments by incorporating historical dialogue context, improve the matching degree between psychosemantic recognition results and the user's actual state, and thus provide a more stable and complete semantic input for subsequent dynamic psychological state vector generation.

[0039] S3. Based on the initial psychological state data, training process data, and psycho-semantic feature data, generate a dynamic psychological state vector of the target user.

[0040] Step S3 is a multi-source psychological state fusion modeling step, used to integrate pre-training assessment data, training process behavioral data, and AI dialogue semantic data into a state representation of the target user at the current training time point. Initial psychological state data can represent the target user's initial basic psychological state; training process data can represent the target user's emotion score, sleep duration, training task completion rate, AI dialogue rounds, and user feedback during training; psychological semantic feature data can represent semantic recognition results such as stress semantic intensity, cognitive bias state, and risk signal intensity. In specific calculations, the initial psychological state data and training process data can be normalized and structured, and combined with the psychological semantic feature data to calculate a comprehensive stress index, emotional stability, behavioral motivation index, sleep state index, and psychological risk index, thereby generating a dynamic psychological profile and a dynamic psychological state vector. Through step S3, a user psychological state representation that updates with the training process can be formed, avoiding fixed training recommendations based solely on initial assessment results.

[0041] As an alternative implementation method, please continue reading. Figure 3 , Figure 3 This is a schematic diagram of the execution flow for generating dynamic psychological state vectors in the dynamic adaptation method for mental health training provided in this application embodiment, as shown below. Figure 3 As shown, it may also include the following steps S31 to S36.

[0042] S31. Normalize the initial psychological state data to obtain the normalized initial psychological state characteristics.

[0043] In step S31, the initial psychological state data may include psychological state data of the target user formed before training begins through psychological scales, AI initial interviews, training needs assessments, or basic state collection, such as initial stress level, initial emotional state, initial sleep state, initial psychological risk level, and initial psychological profile labels. Since initial psychological state data from different sources differ in value range, units, and expression formats, the initial psychological state data can be normalized to convert different types of data into initial psychological state features within a unified numerical range. The normalized initial psychological state features can represent the target user's basic psychological state when entering the training process and provide basic state parameters for the subsequent calculation of the comprehensive stress index, psychological risk index, and dynamic psychological state vector.

[0044] S32. Extract and process emotion scores, sleep duration, training task completion rate, AI dialogue rounds and user feedback data from the training process data to obtain normalized emotion scores, sleep duration, training task completion rate and AI dialogue rounds, and obtain structured user feedback features.

[0045] In step S32, the training process data can originate from the target user's task execution records, daily emotion records, sleep records, AI dialogue records, and training feedback records during continuous training. Emotion scores reflect the target user's subjective emotional state during the current training phase; sleep duration reflects the target user's sleep state; training task completion rate reflects the target user's performance on the training task; AI dialogue rounds reflect the target user's level of interactive activity during training; and user feedback data can include feedback on training difficulty, understanding of training content, training experience, and subjective improvement. Emotion scores, sleep duration, training task completion rate, and AI dialogue rounds can be normalized; text-based or option-based user feedback data can be labeled, ranked, or structured to obtain structured user feedback features.

[0046] S33. Based on the normalized emotion score, sleep duration, training task completion rate, AI dialogue rounds, normalized initial psychological state features, and structured user feedback features, generate a training process feature vector, and based on the most recent... n The normalized mood scores within a day are used to calculate mood fluctuation and mood stability, and a sleep state index is generated based on the normalized sleep duration.

[0047] In step S33, emotion scores, sleep duration, training task completion rate, AI dialogue rounds, initial psychological state characteristics, and user feedback characteristics can be combined in a preset order to form a training process feature vector. This training process feature vector can simultaneously represent the target user's basic psychological state, recent emotional state, sleep state, task execution status, interaction activity level, and subjective feedback. Furthermore, it can be based on recent... n The system calculates mood fluctuation values ​​based on changes in mood scores over a day, and generates mood stability based on the degree of mood fluctuation. It can also generate a sleep state index based on the correspondence between sleep duration and preset sleep state intervals. Thus, training process data can be transformed from scattered records into state features that can participate in subsequent calculations.

[0048] S34. Extract stress semantic intensity, cognitive bias state and risk signal intensity from psychosemantic feature data, calculate comprehensive stress index based on training process feature vector and stress semantic intensity, and calculate behavioral motivation index based on normalized training task completion rate, AI dialogue rounds and structured user feedback features.

[0049] In step S34, psychosemantic feature data can be generated by the psychosemantic recognition model based on dialogue interaction data. Stress semantic intensity reflects the strength of stress, anxiety, distress, or negative emotions expressed by the target user in the dialogue text; cognitive bias state reflects whether the target user exhibits tendencies such as negative interpretation, self-denial, catastrophic thinking, or avoidance cognition; risk signal intensity reflects the degree of abnormal feedback, high-stress expressions, or risk-related expressions in the dialogue text. The comprehensive stress index can be calculated by combining the initial psychological state, current emotion score, sleep status, training task completion status, and stress semantic intensity; the behavioral motivation index can be calculated by combining the training task completion rate, AI dialogue rounds, and user feedback characteristics, thereby reflecting the target user's motivation to continue participating in training and completing tasks.

[0050] S35. Calculate the psychological risk index based on the normalized initial psychological state characteristics, comprehensive stress index, emotional stability, cognitive bias state, and risk signal intensity.

[0051] In step S35, the psychological risk index can be calculated by combining the target user's basic psychological state, current stress level, emotional stability, cognitive bias, and risk signal strength. The normalized initial psychological state characteristics represent the target user's basic risk level; the comprehensive stress index represents the target user's current stress load; emotional stability indicates whether the target user's recent emotional changes are stable; cognitive bias indicates whether the target user has cognitive patterns that hinder training progress; and risk signal strength indicates the strength of risk-related signals appearing in the dialogue interaction data. By incorporating these factors into the risk calculation, the psychological risk index can more comprehensively reflect the target user's psychological risk state at the current training time point than relying solely on scale scores or single emotional feedback.

[0052] S36. Generate a dynamic psychological profile and a dynamic psychological state vector based on the comprehensive stress index, emotional stability, behavioral motivation index, cognitive bias state, sleep state index, and psychological risk index.

[0053] In step S36, a stress level can be generated based on the comprehensive stress index, an emotional stability state can be generated based on emotional stability, a behavioral motivation level can be generated based on the behavioral motivation index, a cognitive state description can be generated based on the cognitive bias state, a sleep state can be generated based on the sleep state index, and a psychological risk level can be generated based on the psychological risk index, thereby forming a dynamic psychological profile of the target user at the current training time point. The dynamic psychological profile tends to describe the target user's current psychological state in a labeled way, while the dynamic psychological state vector tends to express the target user's current psychological state numerically. The dynamic psychological profile and the dynamic psychological state vector can serve as the data basis for subsequent training fit calculation, task load constraints, and training progress control parameter generation, enabling the training content and training progress rhythm to be dynamically adjusted according to the changes in the target user's state.

[0054] Through steps S31 to S36, the initial psychological state, training execution, emotional changes, sleep status, user feedback, and AI dialogue semantic results of the target user can be uniformly integrated, transforming the originally scattered scale data, behavioral data, and semantic data into a computable and updatable dynamic psychological state representation. This processing method not only preserves the target user's basic psychological state when entering the training process but also introduces emotional fluctuations, task completion status, interaction activity, stress semantics, and risk signals during the training process. This allows the dynamic psychological state vector to reflect the target user's comprehensive psychological load, behavioral motivation, and psychological risk status at the current training time point. Therefore, subsequent training unit matching, task load constraints, and training progress control no longer rely on static evaluation results but can dynamically adapt according to the target user's state changes, improving the accuracy of training content selection and training rhythm adjustment.

[0055] As another optional implementation, the feature vector, emotion fluctuation value, dynamic psychological profile, comprehensive stress index, behavioral motivation index, psychological risk index, and dynamic psychological state vector of the above training process respectively satisfy: ; ; ; ; ; ; ; in, i Indicates the target user ID. t Indicates the training time point. Represents the feature vector during the training process. This represents the normalized sentiment score. This represents the normalized sleep duration. This represents the normalized training task completion rate. This indicates the normalized AI dialogue turn sequence. This represents the initial psychological state characteristics after normalization. This represents the characteristics of user feedback after structured processing. Indicates the value of emotional fluctuation. This represents the function for calculating standard deviation. Represents a dynamic psychological profile. Indicates stress level, Indicates emotional stability. Indicates the level of behavioral motivation. This indicates a state of cognitive bias. Indicates sleep state. Indicates the level of psychological risk. This represents the overall stress index. Indicates the semantic intensity of stress. Indicates behavioral motivation index, This indicates a psychological risk index. This indicates the normalized level of emotional stability. Indicates the intensity of perceived risk. Indicates the strength of the risk signal. Indicates the sleep state index, This indicates a state of cognitive bias. Represents a dynamic psychological state vector. , , , , , , , , , , , , All of these represent weighting coefficients. It should be noted that the stress levels mentioned above are generated based on the stress level of the comprehensive stress index, the behavioral motivation level is generated based on the behavioral motivation level of the behavioral motivation index, the sleep state is generated based on the sleep state level of the sleep state index, the psychological risk level is generated based on the risk level of the psychological risk index, and the cognitive risk intensity is obtained based on the cognitive bias state conversion.

[0056] S4. Obtain the training unit label data for each training unit in the mental health training library. The training unit label data includes training objective label, adaptation status label, and task load label.

[0057] Step S4 is the structured annotation step for training resources, which converts different training units in the mental health training library into matchable, filterable, and constrainable data objects. Training objective tags can be used to represent the training purpose corresponding to a training unit, such as stress regulation, emotional stability, cognitive restructuring, behavioral activation, or sleep improvement. Adaptation status tags can be used to represent the user's suitable state for the training unit, such as stress level, emotional stability, behavioral motivation, cognitive bias, or risk level. Task load tags can be used to represent the execution burden of the training unit, such as training difficulty, training duration, push frequency, number of task steps, and execution complexity. Through step S4, the training content can be formed into tagged data, providing standardized training unit data for subsequent training fit calculations and task load constraints.

[0058] In one implementation, the mental health training library may include multiple different types of training units. These training units may include at least one of the following: cognitive restructuring training units, behavioral activation training units, emotional stabilization training units, stress management training units, sleep improvement training units, mindfulness stabilization training units, problem-solving training units, value clarification training units, risk intervention training units, and low-load supportive training units. Each training unit may be configured with a training unit number, training type label, training goal label, adaptation status label, task load label, training difficulty, training duration, push frequency, number of task steps, applicable status conditions, and recommendation weight.

[0059] The training difficulty can be represented by preset levels, for example: in, Indicates the first j The training difficulty of each training unit is represented by 1 for low difficulty and 5 for high difficulty. Training duration can be represented by a preset time interval, for example: in, Indicates the first j The training duration of each training unit can be in minutes. Task load labels can be generated based on training difficulty, training duration, number of task steps, and emotional engagement intensity, representing the task burden the target user needs to undertake when performing the training unit. By setting these labels and parameters for training units, the training content in the mental health training library can be transformed into calculable, comparable, and filterable data objects, providing a data foundation for subsequent training fit calculations, task load constraints, and training task generation.

[0060] S5. Based on the dynamic psychological state vector, training requirement data, training target labels and adaptation state labels of each training unit, calculate the training adaptation degree between the target user and each training unit.

[0061] Step S5 is the user state and training unit matching calculation step, used to identify training units from the mental health training library that match the target user's current state and training needs. Training need data can be used to match with training target labels to determine whether the training unit meets the target user's training goals; dynamic mental state vectors can be used to match with appropriate state labels to determine whether the training unit is suitable for the target user's current stress level, emotional stability, behavioral motivation, sleep state, and psychological risk state. In specific calculations, training target matching features can be generated based on training need data and training target labels, state adaptation features can be generated based on dynamic mental state vectors and appropriate state labels, and further training unit adaptation feature vectors can be generated, thereby obtaining the training fit through similarity calculation or weighted matching calculation. Through step S5, training content matching can be expanded from single training target matching to a comprehensive matching of training needs and dynamic mental states, improving the targeting of training unit recommendations.

[0062] As an optional implementation, the process of calculating the training fitness in step S5 above may also specifically include the following steps S51 to S54.

[0063] S51. Based on the training requirement data and the training target labels of each training unit, generate the training target matching features corresponding to each training unit.

[0064] In step S51, the training requirement data may include the target user's training goals, main distress scenarios, desired psychological problems, phased training needs, and training preferences. Training goal labels can represent the goal type corresponding to each training unit, such as stress regulation, emotional stability, cognitive restructuring, behavioral activation, and sleep improvement. The training requirement data can be matched with the training goal labels of each training unit to obtain the correspondence between each training unit and the target user's training needs. For example, when the target user's training needs focus on stress regulation, training units related to stress regulation can obtain higher training goal matching features; when the training goal of a training unit is weakly correlated with the target user's current training needs, the corresponding training goal matching features can be lower.

[0065] S52. Based on the dynamic psychological state vector and the adaptation state label of each training unit, generate the state adaptation features corresponding to each training unit.

[0066] In step S52, the dynamic psychological state vector may include state parameters such as comprehensive stress index, emotional stability, behavioral motivation index, sleep state index, psychological risk index, and cognitive bias state. The adaptive state label can represent the user's suitable psychological state conditions for the training unit, such as being suitable for a high-stress state, a low-behavioral-motivation state, a state of emotional fluctuation, a state of sleep deprivation, or a state with significant cognitive bias. The dynamic psychological state vector can be compared with the adaptive state labels of each training unit to identify whether the training unit is suitable for the target user's psychological state at the current training time point, thereby generating state adaptation features.

[0067] S53. Generate the training unit adaptation feature vector corresponding to each training unit based on the training target matching features and state adaptation features corresponding to each training unit.

[0068] In step S53, the training target matching features and state adaptation features can be combined to form a training unit adaptation feature vector. This vector simultaneously represents the target matching relationship between the training unit and the target user's training needs, as well as the state adaptation relationship between the training unit and the target user's current psychological state. Through this vector, the training unit is no longer expressed solely as a training target, but rather forms a comprehensive feature representation encompassing both the degree of training target matching and the degree of user state adaptation, providing vectorized input for subsequent training fit calculations.

[0069] S54. Calculate the training fit degree based on the dynamic psychological state vector and the training unit adaptation feature vector.

[0070] In step S54, the dynamic psychological state vector of the target user and the training unit adaptation feature vector corresponding to each training unit can be used to calculate similarity or perform multi-dimensional weighted matching to obtain the training fit between the target user and each training unit. The dynamic psychological state vector represents the target user's current comprehensive psychological state, and the training unit adaptation feature vector represents the adaptation characteristics of the training unit to training requirements and psychological state. The training fit can be calculated using a similarity function or based on the weighted results of multiple matching dimensions, thus providing a ranking basis for subsequent candidate training unit selection and task load constraints.

[0071] Among them, the j The training fit of each training unit satisfies: or, in, Indicates the first i The target user in the first t The training time point and the firstj Training fit between training units Represents a dynamic psychological state vector. Indicates the first j Each training unit corresponds to a training unit adaptation feature vector. Represents the similarity function. Indicates the target user and the first j The training unit in the ... k Matching results across all dimensions Indicates the first k The weights corresponding to each dimension m Indicates the number of matching dimensions.

[0072] Through steps S51 to S54, the training needs and current dynamic psychological state of the target user can be simultaneously incorporated into the training unit matching process. This means that training fit is no longer judged solely based on whether the training objectives are consistent, but rather by comprehensively considering factors such as stress, emotional stability, behavioral motivation, sleep status, psychological risk, and cognitive biases. Consequently, different training units can form a fit ranking corresponding to the target user's current state. This ensures that the training content matches the user's subjective needs while avoiding directly recommending training units unsuitable for the current psychological state, thereby improving the accuracy of target training unit selection and individual fit.

[0073] S6. Based on the training fit and the task load label of each training unit, perform task load constraints on each training unit and select target training units.

[0074] Step S6 is the candidate training unit load constraint screening step, used to further exclude training units that do not match the target user's current state of mind from those training units that meet the training fit requirements. Training fit can be used to form a candidate training unit set; task load tags can be used to obtain the training difficulty, training duration, and execution burden of the training units; the comprehensive stress index and behavioral motivation index can be used to determine whether the target user is currently suitable for performing high-load training tasks. In specific screening, when the comprehensive stress index is high or the behavioral motivation index is low, low-difficulty, short-duration training units can be prioritized; when the comprehensive stress index is low and the behavioral motivation index is high, medium-to-high difficulty or longer-duration training units can be selected. Step S6 avoids selecting training content solely based on fit, which could lead to excessively high training load or an overly rapid pace, thus improving the match between the training task and the user's current state of mind.

[0075] As an optional implementation, the process of selecting target training units in step S6 above may also specifically include the following steps S61 to S63.

[0076] S61. Based on the task load labels of each training unit, obtain the training difficulty and training duration of each training unit.

[0077] In step S61, the task load label can be used to represent the burden level of the training unit during execution. The task load label can include information such as training difficulty, training duration, number of task steps, cognitive input intensity, and emotional exposure intensity. Training difficulty can be represented according to a preset difficulty level, such as low, medium, and high difficulty, or by using a numerical level. Training duration can represent the estimated time required for the target user to complete the corresponding training unit. By reading the task load label of each training unit, the training difficulty and training duration required for subsequent load constraints can be obtained, so that the training unit selection process not only considers whether the training content is suitable, but also whether the target user's current state can bear the corresponding training load.

[0078] S62. Include training units with a training fit greater than or equal to a preset matching threshold into the candidate training unit set.

[0079] In step S62, training fit can represent the degree of matching between the target user and each training unit, and the preset matching threshold can represent the minimum matching condition that a training unit must meet to be included in the candidate range. For each training unit in the mental health training library, the training fit can be compared with the preset matching threshold. Training units that meet the preset matching threshold are included in the candidate training unit set, while training units that do not meet the preset matching threshold are excluded from the candidate training unit set. The candidate training unit set is used to represent the range of training units that have a basic matching relationship with the target user's current training needs and dynamic psychological state.

[0080] S63. Based on the comprehensive stress index, behavioral motivation index, training difficulty and training duration of each training unit, the training units in the candidate training unit set are screened to obtain the target training units.

[0081] In step S63, the comprehensive stress index represents the target user's current stress load status, the behavioral motivation index represents the target user's motivation to continue participating in training and completing tasks, and the training difficulty and training duration represent the task load level of the candidate training units. During the screening process, the training load range that the target user can currently bear can be determined based on the comprehensive stress index and the behavioral motivation index, and training units whose training difficulty and training duration match this training load range can be selected from the candidate training unit set. For example, when the comprehensive stress index is high or the behavioral motivation index is low, low-difficulty, short-duration training units can be prioritized; when the comprehensive stress index is low and the behavioral motivation index is high, medium-to-high difficulty or longer-duration training units can be selected. Thus, task load constraints can be further implemented on the basis of meeting the training fit requirements, avoiding pushing training units that are too burdensome or too fast-paced to the target user in their current state.

[0082] The candidate training unit set satisfies: when or At that time, the target training unit satisfies: ; when and At that time, the target training unit satisfies: ; in, Indicates the first i The target user in the first t The set of candidate training units corresponding to each training time point. Indicates the first j One training unit, This indicates the preset matching threshold. This represents the training fit between the i-th target user at the t-th training time point and the j-th training unit. This represents the overall stress index. Indicates behavioral motivation index, Indicates the first j The training difficulty of each training unit Indicates the first j Training duration for each training unit.

[0083] Through steps S61 to S63, task load constraints are further introduced on top of the training fit selection, ensuring that the target training unit matches not only the target user's training needs and dynamic psychological state, but also the training difficulty and duration that the target user can currently bear. This approach avoids the problems of excessively high training burden, excessively long training time, or excessively fast pace that may occur when simply selecting training units according to fit ranking. Especially under conditions of high overall pressure and low behavioral motivation, it can prioritize the selection of easier-to-execute training content, thereby improving the feasibility of training tasks and the stability of continuous execution.

[0084] S7. Based on the dynamic mental state vector and the task load label of the target training unit, generate training advancement control parameters. The training advancement control parameters are used to control the advancement, pause, repetition, degradation or replacement of the target training unit.

[0085] Step S7 is the dynamic adjustment and control step for the training path, used to adjust subsequent training units, training difficulty, training duration, push frequency, and training execution control parameters based on the target user's state changes during continuous training. The dynamic psychological state vector can provide state parameters such as comprehensive stress index, emotional stability, behavioral motivation index, sleep state index, and psychological risk index; training process data can be used to calculate the most recent... n The training time points show trends in emotion scores and training completion; training comprehension levels can be generated by a psychosemantic recognition model based on dialogue interaction data; dynamic psychological profiles can be used to assist in determining the target user's current psychological state type. In specific control, training adjustment results can be generated based on emotion score trends, training completion trends, training comprehension levels, psychological risk index, and dynamic psychological profiles. These results can then be used to control the continuation, pause, repetition, downgrade, or switching to a stabilization training unit or a risk intervention training unit. Through step S7, the training path can be dynamically adjusted according to changes in the user's state, avoiding mechanical progression of mental health training in a fixed order or with fixed difficulty.

[0086] As an alternative implementation method, please continue reading. Figure 4 , Figure 4 This is a schematic diagram of the execution flow for generating training advancement control parameters in the dynamic adaptation method for mental health training provided in this application embodiment, such as... Figure 4 As shown, it may also include the following steps S71 to S73.

[0087] S71. Based on the training process data, calculate the target user's recent... n Trends in emotion scores and training completion within each training time point.

[0088] In step S71, the training process data may include emotion scores, training task completion results, training task completion rates, task time, and training feedback generated by the target user at multiple training time points. The most recent data can be extracted in chronological order of training time. n The system generates a sequence of emotion scores within each training time point and produces an emotion score trend based on the rise, fall, or fluctuation of emotion scores. Simultaneously, it can extract the most recent... n The training task completion rate or result within each training time point is calculated, and a training completion trend is generated based on the completion rate, continuous completion status, omissions, or interruptions of training tasks. The trend of emotion score changes can reflect whether the target user's recent emotional state has improved, and the training completion trend can reflect the target user's recent stability in executing training tasks and their continued participation.

[0089] S72. Generate training adjustment results based on the trend of emotion score changes, training completion trend, training comprehension level, psychological risk index, and dynamic psychological profile.

[0090] In step S72, the training comprehension level can be obtained from the recognition results of the dialogue interaction data by the psycho-semantic recognition model, representing the target user's understanding of the current training content; the psychological risk index can represent the target user's psychological risk level at the current training time point; the dynamic psychological profile can include state content such as stress level, emotional stability, behavioral motivation level, cognitive bias state, sleep state, and psychological risk level. The emotional score change trend, training completion trend, training comprehension level, psychological risk index, and dynamic psychological profile can be input into the adjustment rule function to generate training adjustment results. The training adjustment results can include the training unit, training difficulty, training duration, push frequency, and training execution control parameters corresponding to the next training time point. For example, if the emotional score improves and the training completion is good, the training difficulty of the next training time point can be increased; if the emotional score decreases, the training difficulty can be reduced and a stabilization training unit can be switched; if the training completion is poor, the training duration can be shortened; if the training comprehension level is low, the current training unit can be repeated; if the psychological risk index reaches the risk threshold, a risk intervention training unit can be switched to and a risk warning label can be generated.

[0091] In one implementation, the training understanding level can be calculated based on the standard target semantic vector of the current training unit and the semantic vector of the target user's understanding feedback regarding the training task. Specifically, the standard target text of the current training unit can be denoted as... And generate the standard target semantic vector of the current training unit through a semantic encoding model: .

[0092] in, Indicates the first j The standard target semantic vector corresponding to each training unit Indicates the first j The standard target text for each training unit Encoder This represents a semantic encoding model. Furthermore, the third... i The target user in the first t The training time point, the first r The understanding feedback text provided during the round-robin dialogue in response to the training task is denoted as... And generate user-understanding feedback semantic vectors through a semantic encoding model: .

[0093] in, Indicates the first i The target user in the first t The training time point, the first r The semantic vector of user understanding feedback corresponding to round-robin dialogue This represents the target user's understanding feedback text in response to the training task. This feedback text can include the target user's restatement of the current training objective, their answers to the training task, task feedback content, and a description of the execution result. Then, the similarity between the standard target semantic vector and the user's understanding feedback semantic vector can be calculated to obtain the user understanding matching degree. .

[0094] in, Indicates the first i The target user in the first t The training time point, the first r The degree of user understanding matching corresponding to round-robin dialogue. Sim This represents the similarity calculation function. Sim Cosine similarity can be used for calculation. Based on user understanding of the matching degree, the training understanding level can be divided into low understanding, partial understanding, and full understanding; when 0 ≤ When <0.4, the training comprehension level is low; when 0.4 ≤ When <0.7, the training comprehension level is partial; when 0.7≤ When the value is ≤1, the training understanding level is considered sufficient. To avoid confusion between user understanding matching and training unit variables, the user understanding matching or its aggregated result can be recorded as the training understanding level value in the training tuning rules. When there is only one round of understanding feedback at the current training time point, When there are multiple rounds of understanding feedback at the current training time point, it is possible to generate a value based on the average, maximum, or weighted value of the user understanding matching degree across multiple rounds. .when Less than the understanding threshold When this occurs, it indicates that the target user does not fully understand the meaning, execution steps, or operation method of the current training unit. Instead of directly advancing to the next training unit, the current training unit is repeated, and the current training objective is reinterpreted, example information is provided, or the training task is broken down into simpler execution steps.

[0095] S73. Generate training advancement control parameters based on the training adjustment results. The training advancement control parameters include the training unit, training difficulty, training duration, push frequency, and training execution control parameters corresponding to the next training time node.

[0096] In step S73, the training adjustment results can serve as the basis for generating training progression control parameters. The training unit corresponding to the next training time node can indicate whether to continue executing the current training unit, switch to a new training unit, switch to a stabilization training unit, or switch to a risk intervention training unit; training difficulty can indicate the difficulty level of the subsequent training task; training duration can indicate the execution time of the subsequent training task; push frequency can indicate the frequency at which the training task is sent to the target user; and training execution control parameters can indicate the control method for advancing, pausing, repeating, downgrading, or replacing the target training unit in the subsequent training process. Through these parameters, the training task no longer mechanically progresses along a fixed path but can be adjusted according to the recent changes in the target user's status and risk situation.

[0097] The training adjustment results include: ; The training and adjustment results are generated by the adjustment rule function, which satisfies the following: ; when hour, ; when hour, ; when hour, ; when hour, ; when hour, ; in, Indicates the first i The target user in the first t The training adjustment results corresponding to each training time point. This indicates the training unit corresponding to the next training time node. This indicates the training difficulty corresponding to the next training time node. This indicates the training duration corresponding to the next training time node. This indicates the push frequency corresponding to the next training time node. This indicates the training execution control parameters corresponding to the next training time node. Rule This represents an adjustment rule function based on changes in user state. Indicates recent n The trend of emotion score changes within each training time point Indicates recent n Training completion trend within each training time point Indicates the first i The target user in the first t The training comprehension level value corresponds to each training time point. This value is generated by the psycho-semantic recognition model based on dialogue interaction data. This indicates a psychological risk index. Represents a dynamic psychological profile. Indicates a stable training unit. This indicates the adjustment amount for training duration. This indicates a risk intervention training unit. This indicates a risk warning sign. This indicates the threshold for understanding. This indicates the risk threshold.

[0098] Through steps S71 to S73, the subsequent training path can be dynamically adjusted based on the target user's recent emotional changes, training completion status, training comprehension level, and psychological risk state. This allows training units, training difficulty, training duration, and execution methods to no longer proceed according to fixed rules. This approach can appropriately increase the intensity of training when the target user's state improves and their completion is good, and promptly reduce the training load, repeat training content, or switch to stabilization or risk intervention training content when the target user's emotions decline, comprehension is insufficient, or risk increases. This improves the timeliness of the training process in response to changes in the user's state and the safety of training progress.

[0099] S8. Generate training tasks based on target training units and training progress control parameters, send the training tasks to the target user, obtain the execution feedback data corresponding to the training tasks, and update the training process data based on the execution feedback data.

[0100] Step S8 is an update step for training task generation, distribution, and feedback. It transforms the target training unit and training progress control parameters into specific training tasks executable by the target user and writes the task execution results back to the training process data. The target training unit can be used to determine the training content; the training progress control parameters can be used to determine the training difficulty, training duration, push frequency, and training execution control parameters corresponding to the training content; the execution feedback data can include task completion results, task time, post-training emotional feedback, post-training user feedback, and post-training dialogue interaction data. In specific execution, training tasks can be generated based on the training content, training difficulty, training duration, push frequency, and training execution control parameters, and sent to the target user; after the target user completes or provides feedback on the training task, the training task completion rate, training duration execution results, emotional score, user feedback data, AI dialogue rounds, and psycho-semantic feature data are updated based on the execution feedback data. Through step S8, a technical chain of "training task generation—task execution feedback—training process data update—next round state calculation" can be formed, enabling subsequent training adaptation and training progress control to be continuously updated based on the latest execution feedback.

[0101] In one implementation, the set of training tasks received by the i-th target user at the t-th training time node can be represented as: ; in, Represents the set of training tasks. to This represents multiple training tasks within a set of training tasks. Each training task can be represented as: ; in, Indicates the first j The task type of each training task Indicates the first j The training difficulty of each training task Indicates the first j Training duration for each training task, Indicates the first j The frequency or triggering conditions for pushing training tasks can be determined. Furthermore, task push results can be generated based on personalized training plans and dynamic mental state vectors. ; in, Indicates the first i The target user in the first t The task push results corresponding to each training time point G This represents the task push function. This indicates a personalized training plan. This represents a dynamic psychological state vector.

[0102] As an optional implementation, the process of sending the training task to the target user in step S8 above may also specifically include the following steps S81 to S85.

[0103] S81. Determine the training content based on the target training unit, and determine the training difficulty, training duration, push frequency and training execution control parameters corresponding to the training content based on the training progress control parameters.

[0104] In step S81, the target training unit can be a training unit selected after training fit calculation and task load constraints. The training content can include cognitive training content, emotion regulation content, behavioral practice content, relaxation training content, sleep improvement content, or risk intervention content, etc. The training progress control parameters can include the training difficulty, training duration, push frequency, and training execution control parameters corresponding to the next training time node. The training difficulty can represent the difficulty of understanding the training content, the difficulty of execution, or the intensity of emotional investment; the training duration can represent the estimated time required for the target user to complete the training content; the push frequency can represent the number of times the training task is sent to the target user within a certain training cycle; the training execution control parameters can indicate whether the training content should be advanced, paused, repeated, downgraded, or replaced in the current training stage.

[0105] S82. Generate training tasks based on training content, training difficulty, training duration, push frequency, and training execution control parameters.

[0106] In step S82, the training content can be combined with training difficulty, training duration, push frequency, and training execution control parameters to form a training task that the target user needs to perform at the current training time node. The training task may include a task topic, task description, training steps, estimated completion time, task feedback entry point, and task completion conditions. For example, when the training difficulty is low, the training task may contain fewer execution steps and simpler explanatory text; when the training difficulty is high, the training task may contain more complete training steps, reflection records, or behavioral practice requirements. Therefore, the target training unit can be converted into a specific task form that the target user can directly execute.

[0107] S83. When the training execution control parameter is expressed as repeating the current training unit, a training task is generated that includes the current training unit, explanation information, example information, and step breakdown information.

[0108] In step S83, repeating the current training unit can address situations where the target user has a low level of understanding of the current training content, poor training completion, or user feedback indicating difficulty in execution. In this case, the training task does not directly switch to a new training unit; instead, the current training unit is retained, and explanatory information, example information, and step breakdown information are added to the original training content. The explanatory information can supplement the training objectives and methods; the example information can provide specific scenarios or demonstrations; and the step breakdown information can break down the originally relatively complete training task into multiple smaller execution steps to reduce the difficulty for the target user to understand and execute the current training unit.

[0109] S84. When the training execution control parameter is expressed as switching between a stabilization training unit and a risk intervention training unit, a training task containing either stabilization training content or risk intervention training content is generated.

[0110] In step S84, switching the stabilization training unit can be used to address situations where the target user's emotional score decreases, overall stress increases, or the training is not suitable for further increasing the difficulty. The stabilization training content can include breathing relaxation, emotional soothing, mindfulness practice, low-load recording, etc. Switching the risk intervention training unit can be used to address situations where the psychological risk index reaches the risk threshold or obvious risk signals appear in the dialogue. The risk intervention training content can include risk warnings, emotional stabilization practice, help-seeking guidance, crisis resource prompts, or prompts for human intervention, etc.

[0111] S85. Send the training task to the target user.

[0112] In step S85, the generated training task can be sent to the target user according to the push frequency and training time schedule in the training progress control parameters. When sending the training task, the training content, training steps, training duration, task description, and feedback entry can be included, enabling the target user to complete the training and submit execution feedback according to the task requirements. After the training task is sent, the target user's task completion result, task time, post-training emotional feedback, post-training user feedback, and post-training dialogue interaction data can be used as subsequent execution feedback data for data updates and dynamic psychological state vector generation at the next training time node.

[0113] Through steps S81 to S85, the selected target training units can be further transformed into specific training tasks that can be executed by the target user. The difficulty, duration, push frequency, and execution method of the training tasks are then adapted and configured according to the training progress control parameters. This process, during the task generation phase, combines different control scenarios such as repetitive training, stabilization training, and risk intervention to adjust the presentation format and execution load of the training content. This ensures that the training tasks match the target user's current level of understanding, emotional state, and risk status, thereby improving the executability, continuity, and safety of the training tasks.

[0114] As another optional implementation, the process of updating the training process data in step S8 above may also specifically include the following steps S86 to S88.

[0115] S86. Obtain the target user's task completion results, task time, post-training emotional feedback, post-training user feedback, and post-training dialogue interaction data for the training task.

[0116] In step S86, the task completion result can indicate whether the target user has completed the training task, the percentage of the task completed, whether the training task was interrupted, or whether the training task was skipped; the task time can indicate the actual time spent by the target user from starting to execute the training task to submitting feedback; the post-training emotional feedback can indicate the target user's emotional score, emotional change direction, or subjective emotional description after completing the training task; the post-training user feedback can include the target user's feedback on training difficulty, understanding of training content, training experience, task burden, and subjective improvement; the post-training dialogue interaction data can include the natural language text generated when the target user engages in feedback dialogue, status description, or question consultation with the AI ​​after completing the training task.

[0117] S87. Update the training task completion rate based on the task completion results, update the training duration execution results based on the task time, update the emotion score based on the emotion feedback after training, update the user feedback data based on the user feedback after training, and update the AI ​​dialogue rounds and psychosemantic feature data based on the dialogue interaction data after training.

[0118] In step S87, the training task completion rate of the target user in the current training cycle can be recalculated based on the task completion results, and the task time can be compared with the preset training task duration to obtain the training duration execution result. Post-training emotional feedback can be converted into new emotional scores or emotional change data, and post-training user feedback can be updated after structured processing. For post-training dialogue interaction data, the number of newly added AI dialogue rounds can be counted, and psychosemantic recognition can be performed on the newly added dialogue content to update psychosemantic feature data such as emotional state, cognitive pattern, behavioral tendency, training comprehension level, and risk signals.

[0119] In one implementation, after the training process data is updated, the training task for the next training time node can be adjusted based on the updated training task completion rate, the number of consecutive missed check-ins, and the strength of risk signals. When the training task completion rate is lower than a preset completion rate threshold, the training difficulty for the next training time node is reduced. .

[0120] in, Indicates the first i The target user in the first t The normalized training task completion rate corresponding to each training time node. Indicates the first i The target user in the first t The training difficulty corresponding to each training time point. Indicates the first i The training difficulty for each target user at the next training time point.

[0121] When a target user's consecutive missed check-ins reach a preset number, a lightweight training task will be pushed to them first. .

[0122] in, Indicates the first i The target user in the first t The number of consecutive missed check-ins corresponding to each training time point. Indicates the first i The training task for each target user at the next training time node. This indicates a light training task.

[0123] S88. Write the updated training task completion rate, training duration execution results, emotion score, user feedback data, AI dialogue rounds, and psycho-semantic feature data into the training process data for use in the generation of dynamic psychological state vectors at the next training time node.

[0124] In step S88, the updated training task completion rate, training duration execution results, emotion scores, user feedback data, AI dialogue rounds, and psycho-semantic feature data can be stored as the latest training process data for the current training time node. After the training process data is updated, it can be called upon at the next training time node to regenerate the dynamic psychological state vector and further participate in the calculation of training fit, task load constraints, and the generation of training progress control parameters.

[0125] Through steps S86 to S88, the actual execution status, time consumption, post-training emotional changes, subjective feedback, and newly added dialogue content of the target user's training task can be written back to the training process data, ensuring continuous updates to the training process data with each training task completion and feedback. This processing method allows the execution result of the current training task to directly influence the generation of the dynamic psychological state vector at the next training time node, preventing subsequent training adaptation from remaining on historical states or initial evaluation results. This forms a technical chain of "task execution—feedback collection—state update—next round of adaptation," improving the continuity and real-time nature of training content adjustment and training progress control.

[0126] In a specific application scenario, the target user could be a 28-year-old female working professional whose primary training needs are alleviating workplace evaluation anxiety, improving sleep quality, and enhancing behavioral motivation. Upon entering the mental health training process, the target user provides initial psychological state data and training needs data. Initial psychological state data may include the results of an introductory psychological scale, initial mood score, sleep duration, and initial risk level. Training needs data may include natural language descriptions such as "worried about work performance," "negative feedback from superiors," "difficulty falling asleep," and "desire to reduce anxiety responses." After performing psychosemantic recognition on these natural language descriptions, psychosemantic features such as workplace evaluation anxiety, impaired sleep, self-blaming automatic thoughts, and decreased behavioral confidence can be obtained. Based on these features, low-load emotional stabilization training, sleep management training, and cognitive adjustment training related to workplace anxiety are prioritized and matched.

[0127] At the first training time point, the normalized emotion score, sleep duration, training task completion rate, AI dialogue rounds, initial psychological state features, and user feedback features of the target user can be combined into a training process feature vector, for example: X i , 1 =[0.33,0.55,0.50,0.40,0.52,0.45].

[0128] Here, 0.33 represents the normalized emotion score, 0.55 represents the normalized sleep duration, 0.50 represents the normalized training task completion rate, 0.40 represents the normalized AI dialogue rounds, 0.52 represents the normalized initial psychological state characteristics, and 0.45 represents the structured user feedback characteristics. Based on the above training process feature vectors and psychosemantic feature data, the comprehensive stress index, behavioral motivation index, sleep state index, and psychological risk index of the target user at the first training time node can be calculated. For example, the behavioral motivation index can be calculated based on the training task completion rate, AI dialogue rounds, and user feedback characteristics: Bi , 1 =0.25+0.12+0.09=0.46.

[0129] This behavioral motivation index indicates that the target user's current behavioral motivation is moderate to low. Furthermore, a dynamic psychological state score corresponding to the first training time point can be generated: Score i , 1 ={PI:0.572, ES:low, BI:0.46, CI:high, SI:0.55, RI:medium risk}.

[0130] The corresponding dynamic psychological profile can be represented as: P i , 1 ={Stress level: moderately high, emotional stability: low, behavioral motivation: moderately low, cognitive risk labels: [self-blame, low self-esteem, evaluation anxiety], sleep status: impaired sleep, risk level: medium risk}.

[0131] Through this dynamic psychological state scoring and dynamic psychological profile, the target user's stress state, emotional stability, behavioral motivation, cognitive risk, sleep state, and risk level can be uniformly expressed at the same training time point, providing a data foundation for subsequent training unit matching and training progress control.

[0132] When generating the training plan for Day 1, a personalized training plan can be generated based on the target user's dynamic psychological state vector, training needs data, and training unit label data from the mental health training library. Plan i , 1 ={ U ER , U Sleep , U CBT , U BA}

[0133] in, U ER This can represent an emotion regulation training unit, configured as follows: U ER ={G: Working professionals, O: Reduce anxiety and arousal, M: Breathing relaxation, D: 1, T: 5, A: [High stress, impaired sleep], Y: Improve emotional stability}.

[0134] U Sleep This can represent a sleep management training unit, configured as follows: U Sleep={G: Sleep-impaired users, O: Improve sleep onset difficulties, M: Sleep management, D: 1, T: 8, A: [Sleep deprivation, high stress], Y: Improve sleep quality}.

[0135] U CBT It can represent a cognitive reconstruction training unit, configured as follows: U CBT ={G: Workplace anxiety user, O: Adjust self-blaming thoughts, M: CBT, D: 2, T: 10, A: [self-blame, evaluation anxiety, low self-esteem], Y: reduce cognitive bias}.

[0136] U BA It can represent behavioral activation training units, configured as follows: U BA ={G: Low-motivation users, O: Increase action motivation, M: Behavioral activation, D: 1, T: 5, A: [Medium to low behavioral motivation], Y: Enhance sense of action}.

[0137] In this framework, G represents the target audience, O represents the training objective, M represents the training method, D represents the training difficulty, T represents the training duration, A represents the suitability status label, and Y represents the expected training effect. Therefore, the training path for Day 1 could include: first, 5 minutes of breathing relaxation training to reduce anxiety arousal; then, 8 minutes of sleep management training to establish pre-sleep stress-reduction behaviors; followed by a 10-minute cognitive awareness task to identify self-blaming automatic thoughts such as "I can never do well"; and finally, a low-difficulty behavioral activation task, such as "Write down a minimum achievable task before work tomorrow." This approach allows the training tasks to simultaneously cover four areas: emotional stability, sleep improvement, cognitive adjustment, and behavioral activation, while maintaining a low workload in terms of both training difficulty and duration.

[0138] During the low-load stabilization phase from day 1 to day 7, low-difficulty, short-duration training tasks can be prioritized to assess the target user's ability to consistently participate in training. Training content in this phase can include 5-minute breathing relaxation exercises, daily mood ratings, sleep tracking, AI-supportive dialogue, and a low-difficulty behavioral goal. An example training task could be: "Take a 10-minute walk after get off work today and record your mood rating before and after the walk." If the target user exhibits low completion rates or strong resistance, the number of tasks can be reduced, and lighter mood-stabilization exercises can be introduced. If the target user performs well and their mood rating improves, cognitive restructuring, value clarification, or problem-solving training units can be gradually added in subsequent training phases. The focus in this phase is not on rapidly increasing training difficulty, but rather on maintaining stable training participation and achievable execution.

[0139] During the cognitive adjustment and behavioral activation phase from day 8 to day 14, a review can be conducted based on the training completion rate of the previous 7 days, changes in mood scores, sleep duration, changes in AI dialogue mood, training comprehension, and user subjective feedback. If the review results show that anxiety has decreased to a mild level, the training can be upgraded to cognitive exercises and social support tasks; if sleep remains poor, sleep training can be retained while reducing the intensity of cognitive tasks. In the cognitive restructuring task, target users can be guided to identify their current thoughts, evaluate evidence supporting and not supporting those thoughts, find alternative explanations, and generate more balanced thoughts. For example, for the self-blaming automatic thought that "a poor performance in this report means the end of my career development," alternative thoughts can be generated: "This report is indeed important, but a poor performance in one report does not mean the end of my career development. I can first prepare the three most core points clearly." Subsequently, real-world behavioral tasks can be matched to integrate the cognitive adjustment results with actual behavioral exercises.

[0140] During the training process, it is possible to analyze the target user's recent... n The training plan is dynamically adjusted based on the changing trends of emotion scores and training completion rates within each training time point. If the overall emotion score increases and the training completion rate is high, the training difficulty of the next training time point can be appropriately increased; if the emotion score continuously decreases, the AI ​​support frequency can be increased, and emotion-stable training or supportive dialogue tasks can be prioritized; if the training completion rate is consistently below 50%, it can be determined that compliance is low, and the difficulty of subsequent tasks can be reduced; if the number of consecutive missed check-ins reaches a preset number, lightweight training tasks can be prioritized; if the risk signal reaches the risk threshold, regular training can be paused and the risk intervention process can be initiated.

[0141] During the consolidation and relapse prevention phase from day 15 to day 21, key consolidation areas can be automatically generated based on the target user's training data from the previous 14 days. If sleep improvement is significant but workplace evaluation anxiety remains high, scenarios involving leadership feedback, pre-report emotional stabilization exercises, and workplace communication practice can be added. If behavioral motivation remains low, low-difficulty behavioral activation tasks can continue to be configured, and these tasks can be broken down into smaller steps. If cognitive biases significantly decrease, value clarification, long-term action plans, and relapse prevention plans can be gradually added. The training objectives for this phase can include consolidating new cognitive and behavioral habits, establishing long-term stress coping strategies, developing relapse prevention plans, and improving the target user's ability to cope autonomously with future stressful scenarios. Training content can include high-pressure scenario rehearsals, relapse prevention plans, long-term sleep management, workplace communication practice, a personal support resource list, value clarification, and long-term action plans.

[0142] In the complete training process described above, training tasks can be generated without a fixed schedule. Instead, task push results can be generated based on personalized training plans and dynamic psychological state vectors. This allows training tasks to be continuously adjusted according to the target user's state changes at different training stages, rather than proceeding according to a fixed schedule. The above application scenario is merely one example of an embodiment of this application. In other implementations, the target user's training needs, training stages, training unit types, and training adjustment rules can be adaptively set according to actual mental health training scenarios.

[0143] The dynamic adaptation method for mental health training provided in this application acquires the initial psychological state data, training demand data, training process data, and dialogue interaction data of the target user. It then performs psycho-semantic recognition on the dialogue interaction data, enabling the target user's emotional state, cognitive patterns, behavioral tendencies, training comprehension level, and risk signals during the training process to be transformed into structured psycho-semantic feature data. Furthermore, it fuses the initial psychological state data, training process data, and psycho-semantic feature data to generate a dynamic psychological state vector, allowing the target user's comprehensive stress, emotional stability, behavioral motivation, sleep state, psychological risk, and cognitive bias to be dynamically updated at different training time points. Next, it calculates the training fit based on the dynamic psychological state vector, training demand data, and training unit label data, and selects target training units by combining task load labels, ensuring that the target training units simultaneously meet training objective adaptation, user state adaptation, and task load constraints. Finally, it generates training advancement control parameters based on the training adjustment results and updates the training process data based on execution feedback data, thus forming a continuous data processing chain of psycho-semantic recognition, dynamic state calculation, training unit matching, task load constraints, training advancement control, and execution feedback updates. Therefore, the training content and training pace can be dynamically adjusted based on changes in the target user's status and execution feedback, reducing problems such as content mismatch, unreasonable workload, and adjustment lag caused by fixed training paths, and improving the executability, continuity, and security of training tasks.

[0144] Please continue reading. Figure 5 , Figure 5 This is a schematic diagram of the system structure of the dynamic adaptation system for mental health training provided in this application embodiment, as shown below. Figure 5 As shown, the dynamic adaptation system 50 for mental health training includes: a state data acquisition module 51, a psychological semantic generation module 52, a state vector generation module 53, a training label acquisition module 54, an adaptation score calculation module 55, a target unit screening module 56, a propulsion parameter generation module 57, and a training task processing module 58.

[0145] The state data acquisition module 51 is used to acquire the initial psychological state data, training requirement data, training process data, and dialogue interaction data of the target user; the psychological semantic generation module 52 is used to perform text preprocessing and context fusion on the dialogue interaction data to obtain a dialogue semantic vector, and input the dialogue semantic vector into the psychological semantic recognition model to generate psychological semantic feature data; the state vector generation module 53 is used to generate the dynamic psychological state vector of the target user based on the initial psychological state data, the training process data, and the psychological semantic feature data; the training label acquisition module 54 is used to acquire the training unit label data of each training unit in the mental health training library, the training unit label data including training target label, adaptation state label, and task load label; the adaptation score calculation module 55 is used to calculate the dynamic psychological state vector, the training requirement data, and the dialogue semantic feature data based on the dialogue semantic data. The training target label and adaptation status label of each training unit are used to calculate the training fit between the target user and each training unit; the target unit screening module 56 is used to perform task load constraints on each training unit according to the training fit and the task load label of each training unit, and screen target training units; the advancement parameter generation module 57 is used to generate training advancement control parameters based on the dynamic psychological state vector and the task load label of the target training unit, and the training advancement control parameters are used to control the advancement, pause, repetition, degradation or replacement of the target training unit; the training task processing module 58 is used to generate training tasks based on the target training units and the training advancement control parameters, send the training tasks to the target user, obtain the execution feedback data corresponding to the training tasks, and update the training process data based on the execution feedback data.

[0146] It should be noted that the aforementioned dynamic adaptation system for mental health training can execute the dynamic adaptation method for mental health training provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the embodiments of the dynamic adaptation system for mental health training can be found in the dynamic adaptation method for mental health training provided in the embodiments of this application.

[0147] Figure 6 This is a schematic diagram of the hardware structure of the electronic device for implementing the dynamic adaptation method for mental health training provided in the embodiments of this application, as shown below. Figure 6 As shown, the electronic device 600 includes: One or more processors 610 and memory 620, Figure 6 Take the 610 processor as an example.

[0148] The processor 610 and the memory 620 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0149] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the dynamic adaptation method for mental health training in the embodiments of this application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, thereby implementing the dynamic adaptation method for mental health training described in the above embodiments.

[0150] The memory 620 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the mental health training dynamic adaptation system. Furthermore, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory remotely located relative to the processor 610, and these remote memories can be connected to the mental health training dynamic adaptation system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0151] The one or more modules are stored in the memory 620. When executed by the one or more processors 610, they perform the dynamic adaptation method for mental health training in any of the above method embodiments, for example, performing the above-described method. Figure 2 Method steps S1 to S8, Figure 3 Method steps S31 to S36, Figure 4 The method steps S71 to S73 are implemented. Figure 5 The functions of modules 51-58 in the document.

[0152] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0153] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 6 One of the processors 610 can enable the one or more processors to execute the mental health training dynamic adaptation method in any of the above method embodiments, for example, to execute the above-described Figure 2 Method steps S1 to S8, Figure 3 Method steps S31 to S36, Figure 4 The method steps S71 to S73 are implemented. Figure 5 The functions of modules 51-58 in the document.

[0154] This application provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions that, when executed by an electronic device, enable the electronic device to perform the dynamic adaptation method for mental health training in any of the above method embodiments, for example, to perform the above-described method. Figure 2 Method steps S1 to S8, Figure 3 Method steps S31 to S36, Figure 4 The method steps S71 to S73 are implemented. Figure 5 The functions of modules 51-58 in the document.

[0155] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0157] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A dynamic adaptation method for mental health training, characterized in that, include: Acquire initial psychological state data, training needs data, training process data, and dialogue interaction data of the target users; The dialogue interaction data is preprocessed and fused with context to obtain a dialogue semantic vector, and the dialogue semantic vector is input into a psycho-semantic recognition model to generate psycho-semantic feature data. Based on the initial psychological state data, the training process data, and the psycho-semantic feature data, a dynamic psychological state vector of the target user is generated. Obtain training unit label data for each training unit in the mental health training library. The training unit label data includes training target labels, adaptation status labels, and task load labels. Based on the dynamic psychological state vector, the training requirement data, the training target label and the adaptation state label of each training unit, the training adaptation degree between the target user and each training unit is calculated. Based on the training fit and the task load label of each training unit, task load constraints are applied to each training unit to select target training units; Training advancement control parameters are generated based on the dynamic mental state vector and the task load label of the target training unit. The training advancement control parameters are used to control the advancement, pause, repetition, degradation or replacement of the target training unit. A training task is generated based on the target training unit and the training advancement control parameters. The training task is sent to the target user. Execution feedback data corresponding to the training task is obtained, and the training process data is updated based on the execution feedback data.

2. The dynamic adaptation method for mental health training according to claim 1, characterized in that, The process involves preprocessing the dialogue interaction data with text and fusing it with context to obtain a dialogue semantic vector. This dialogue semantic vector is then input into a psycho-semantic recognition model to generate psycho-semantic feature data, including: The dialogue interaction data is processed by word segmentation, noise reduction, and text standardization to obtain the current dialogue text; Obtain the target user's historical dialogue text before the current dialogue round, and concatenate the current dialogue text with the historical dialogue text to obtain the contextual dialogue text; The context dialogue text is semantically encoded to obtain the dialogue semantic vector; The dialogue semantic vector is input into the psycho-semantic recognition model to generate emotion state recognition results, cognitive pattern recognition results, behavioral tendency recognition results, training understanding recognition results, and risk signal recognition results. Based on the emotional state recognition results, the cognitive pattern recognition results, the behavioral tendency recognition results, the training understanding recognition results, and the risk signal recognition results, the psycho-semantic feature data is generated.

3. The dynamic adaptation method for mental health training according to claim 1, characterized in that, The step of generating the dynamic psychological state vector of the target user based on the initial psychological state data, the training process data, and the psycho-semantic feature data includes: The initial psychological state data is normalized to obtain the normalized initial psychological state features. Emotional scores, sleep duration, training task completion rate, AI dialogue rounds, and user feedback data are extracted and processed from the training process data to obtain normalized emotional scores, sleep duration, training task completion rate, and AI dialogue rounds, and to obtain structured user feedback features. Based on the normalized emotion score, sleep duration, training task completion rate, AI dialogue rounds, normalized initial psychological state features, and structured user feedback features, a training process feature vector is generated. Emotional fluctuation value and emotional stability are calculated based on the normalized emotion score over the most recent n days. A sleep state index is generated based on the normalized sleep duration. The stress semantic intensity, cognitive bias state, and risk signal intensity are extracted from the psycho-semantic feature data. A comprehensive stress index is calculated based on the training process feature vector and the stress semantic intensity. A behavioral motivation index is calculated based on the normalized training task completion rate, AI dialogue rounds, and the structured user feedback features. The psychological risk index is calculated based on the normalized initial psychological state characteristics, the comprehensive stress index, the emotional stability, the cognitive bias state, and the risk signal intensity. Based on the comprehensive stress index, the emotional stability, the behavioral motivation index, the cognitive bias state, the sleep state index, and the psychological risk index, a dynamic psychological profile and a dynamic psychological state vector are generated.

4. The dynamic adaptation method for mental health training according to claim 3, characterized in that, The training process feature vector, the emotion fluctuation value, the dynamic psychological profile, the comprehensive stress index, the behavioral motivation index, the psychological risk index, and the dynamic psychological state vector respectively satisfy the following: ； ； ； ； ； ； ； in, i Indicates the target user ID. t Indicates the training time point. This represents the feature vector of the training process. This represents the normalized sentiment score. This represents the normalized sleep duration. This represents the normalized training task completion rate. This represents the normalized AI dialogue turn. This represents the normalized initial psychological state characteristics. This represents the user feedback features after the structured processing. This represents the emotional fluctuation value. This represents the function for calculating standard deviation. This refers to the dynamic psychological profile. Indicates stress level, Indicates emotional stability. Indicates the level of behavioral motivation. This indicates a state of cognitive bias. Indicates sleep state. Indicates the level of psychological risk. This indicates the comprehensive pressure index. Indicates the semantic intensity of the pressure. This represents the behavioral dynamics index. This represents the psychological risk index. This represents the normalized emotional stability. Indicates the intensity of perceived risk. Indicates the strength of the risk signal. This represents the sleep state index. This indicates the state of cognitive bias. This represents the dynamic psychological state vector. , , , , , , , , , , , , All represent weighting coefficients; The stress level is generated based on the stress level of the comprehensive stress index, the behavioral motivation level is generated based on the behavioral motivation level of the behavioral motivation index, the sleep state is generated based on the sleep state level of the sleep state index, and the psychological risk level is generated based on the risk level of the psychological risk index.

5. The dynamic adaptation method for mental health training according to claim 1, characterized in that, The step of calculating the training fit between the target user and each training unit based on the dynamic psychological state vector, the training requirement data, the training target label and the adaptation state label of each training unit includes: Based on the training requirement data and the training target labels of each training unit, generate training target matching features corresponding to each training unit; Based on the dynamic psychological state vector and the adaptation state label of each training unit, generate the state adaptation feature corresponding to each training unit. Based on the training target matching features and state adaptation features corresponding to each training unit, a training unit adaptation feature vector is generated for each training unit. The training fit is calculated based on the dynamic psychological state vector and the training unit adaptation feature vector. Among them, the j The training fit for each training unit satisfies: or, in, Indicates the first i The target user in the first t The training time point and the first j Training fit between training units This represents the dynamic psychological state vector. Indicates the first j Each training unit corresponds to a training unit adaptation feature vector. Represents the similarity function. Indicates that the target user and the first j The training unit in the ... k Matching results across all dimensions Indicates the first k The weights corresponding to each dimension m Indicates the number of matching dimensions.

6. The dynamic adaptation method for mental health training according to claim 3, characterized in that, The step of constraining the task load of each training unit based on the training fit and the task load label of each training unit, and selecting target training units, includes: Based on the task load labels of each training unit, obtain the training difficulty and training duration of each training unit; Training units with a training fit greater than or equal to a preset matching threshold are included in the candidate training unit set. Based on the comprehensive stress index, the behavioral motivation index, the training difficulty and training duration of each training unit, the training units in the candidate training unit set are screened to obtain the target training unit. Wherein, the set of candidate training units satisfies: when or At that time, the target training unit satisfies: ； when and At that time, the target training unit satisfies: ； in, Indicates the first i The target user in the first t The set of candidate training units corresponding to each training time point. Indicates the first j One training unit, This indicates the preset matching threshold. Indicates the first i The target user in the first t The training time point and the first j Training fit between training units This indicates the comprehensive pressure index. This represents the behavioral dynamics index. Indicates the first j The training difficulty of each training unit Indicates the first j Training duration for each training unit.

7. The dynamic adaptation method for mental health training according to claim 1, characterized in that, The generation of training advancement control parameters based on the dynamic mental state vector and the task load label of the target training unit includes: Based on the training process data, calculate the target user's recent... n Trends in emotion scores and training completion within each training time point; Based on the trend of emotion score change, the trend of training completion, the degree of training comprehension, the psychological risk index, and the dynamic psychological profile, training adjustment results are generated. The training advancement control parameters are generated based on the training adjustment results. The training advancement control parameters include the training unit, training difficulty, training duration, push frequency, and training execution control parameters corresponding to the next training time node. The training adjustment results include: ； The training adjustment result is generated by an adjustment rule function, which satisfies the following: ； when hour, ; when hour, ; when hour, ; when hour, ; when hour, ; in, Indicates the first i The target user in the first t The training adjustment results corresponding to each training time point. This indicates the training unit corresponding to the next training time node. This indicates the training difficulty corresponding to the next training time node. This indicates the training duration corresponding to the next training time node. This indicates the push frequency corresponding to the next training time node. This indicates the training execution control parameters corresponding to the next training time node. Rule This represents an adjustment rule function based on changes in user state. Indicates recent n The trend of emotion score changes within each training time point Indicates recent n Training completion trend within each training time point Indicates the first i The target user in the first t The training comprehension level value corresponds to each training time point, and the training comprehension level value is generated by the psycho-semantic recognition model based on dialogue interaction data. This represents the psychological risk index. This refers to the dynamic psychological profile. Indicates a stable training unit. This indicates the amount of training duration adjustment. This indicates a risk intervention training unit. This indicates a risk warning sign. This indicates the threshold for understanding. This indicates the risk threshold.

8. The dynamic adaptation method for mental health training according to claim 1, characterized in that, The step of generating a training task based on the target training unit and the training advancement control parameters, and sending the training task to the target user, includes: The training content is determined based on the target training unit, and the training difficulty, training duration, push frequency and training execution control parameters corresponding to the training content are determined based on the training advancement control parameters. The training task is generated based on the training content, the training difficulty, the training duration, the push frequency, and the training execution control parameters; When the training execution control parameter indicates that the current training unit is repeated, a training task is generated that includes the current training unit, explanation information, example information, and step breakdown information. When the training execution control parameter indicates switching between a stabilization training unit and a risk intervention training unit, a training task containing either stabilization training content or risk intervention training content is generated. The training task is sent to the target user.

9. The dynamic adaptation method for mental health training according to claim 1, characterized in that, The step of obtaining the execution feedback data corresponding to the training task and updating the training process data based on the execution feedback data includes: Acquire the target user’s task completion results, task time, post-training emotional feedback, post-training user feedback, and post-training dialogue interaction data for the training task. Update the training task completion rate based on the task completion result, update the training duration execution result based on the task time, update the emotion score based on the post-training emotion feedback, update the user feedback data based on the post-training user feedback, and update the AI ​​dialogue rounds and psycho-semantic feature data based on the post-training dialogue interaction data. The updated training task completion rate, training duration execution result, emotion score, user feedback data, AI dialogue rounds, and psycho-semantic feature data are written into the training process data for use in the generation of dynamic psychological state vectors at the next training time node.

10. A dynamic adaptation system for mental health training, characterized in that, include: The state data acquisition module is used to acquire the target user's initial psychological state data, training requirement data, training process data, and dialogue interaction data; The psycho-semantic generation module is used to perform text preprocessing and context fusion on the dialogue interaction data to obtain a dialogue semantic vector, and input the dialogue semantic vector into the psycho-semantic recognition model to generate psycho-semantic feature data. The state vector generation module is used to generate the dynamic psychological state vector of the target user based on the initial psychological state data, the training process data, and the psycho-semantic feature data. The training tag acquisition module is used to acquire training unit tag data for each training unit in the mental health training library. The training unit tag data includes training target tags, adaptation status tags, and task load tags. The adaptation score calculation module is used to calculate the training adaptation degree between the target user and each training unit based on the dynamic psychological state vector, the training requirement data, the training target label and adaptation state label of each training unit. The target unit filtering module is used to perform task load constraints on each training unit based on the training fit and the task load label of each training unit, and to filter target training units. The advancement parameter generation module is used to generate training advancement control parameters based on the dynamic mental state vector and the task load label of the target training unit. The training advancement control parameters are used to control the advancement, pause, repetition, degradation or replacement of the target training unit. The training task processing module is used to generate training tasks based on the target training unit and the training advancement control parameters, send the training tasks to the target user, obtain the execution feedback data corresponding to the training tasks, and update the training process data based on the execution feedback data.

Images