Digital human intelligent inquiry method and system for teenager emotion health assessment

Abstract

The invention discloses a digital human intelligent inquiry method and system for teenager emotion health assessment, and relates to the technical field of intelligent inquiry, and the system comprises a questionnaire management module, a digital human presentation module, a voice recognition module, a large language model module, a dynamic jump logic control module and a report generation module. The method comprises the steps that a standard questionnaire for teenager emotion health screening is loaded and analyzed, natural language answers of teenagers are converted into text data through a voice recognition module, deep semantic analysis is conducted on the answer content through a large language model, and intelligent diagnosis and judgment of the teenager emotion state are achieved; meanwhile, the system is combined with screening logic to dynamically adjust the inquiry process, through the character-to-voice and mouth shape synthesis technology, multi-round interaction with teenagers is carried out through a personified digital doctor image, and the screening process is completed step by step; according to the method, efficient grading screening of teenager emotion problems can be realized, and real-time feedback and personalized intervention suggestions are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical consultation, and specifically to a digital human intelligent medical consultation method and system for adolescent emotional health assessment. Background Art

[0002] With the increasing attention of society to the mental health problems of adolescents, emotional health screening has gradually become an important part of mental health management; the existing emotional health assessment methods mainly rely on manual interviews or paper questionnaires. Although these methods have a certain degree of accuracy, they have problems such as low efficiency, high cost, and dependence on professional personnel; in addition, since adolescents often feel nervous or uncomfortable when facing manual interviews, the true results of the assessment are affected.

[0003] In recent years, with the rapid development of large language models, models such as GPT-3 have shown extremely high potential in natural language understanding and generation, being able to imitate human conversations, understand contexts, and give appropriate responses; at the same time, the progress of technologies such as speech recognition, text-to-speech, and lip-sync synthesis has made it possible to conduct emotional health screening with anthropomorphic digital human images; however, how to efficiently integrate these technologies to achieve a digital human medical consultation system for adolescent emotional health still faces technical and application challenges. Summary of the Invention

[0004] The purpose of the present invention is to provide a digital human intelligent medical consultation method and system for adolescent emotional health assessment to solve the problems raised in the prior art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A digital human intelligent medical consultation method for adolescent emotional health assessment, including the following steps:

[0006] S1. Load a standardized adolescent emotional health screening questionnaire to ensure that the questionnaire structure and logical process meet the preset specifications, providing basic data support for subsequent assessments;

[0007] S2. Use a virtual digital human to ask questions to adolescents in the form of voice and lip movement. The voice answers of adolescents are converted into text by a speech recognition module and analyzed by a large language model;

[0008] S3. Input all contents into the large language model for comprehensive evaluation, generating a report including a health score and personalized intervention suggestions.

[0009] Furthermore, in step S1: The questionnaire content includes classification questions of adolescent problems, and the questionnaire design includes jump logic, so as to dynamically adjust the order of subsequent questions according to the answers of adolescents;

[0010] In the design of questionnaire questions combined with jump logic, an additional item analysis mechanism is added. The additional item analysis mechanism marks an additional item analysis each time the content of the user's answer triggers the jump logic to jump in the direction of in-depth negative emotion analysis, and calculates the total number of marked additional item scoring times n at the end of the questionnaire survey; and at the end of the final test, the total additional score is incorporated into the analysis process of the severity of the user's emotional state for analysis and generation of a comprehensive emotional assessment report; adding the additional item analysis can record the number of times the user jumps in the direction of in-depth analysis of negative emotions. The more times the user jumps in the direction of in-depth negative emotion analysis, the more serious and complex the user's negative emotions are. An additional score is needed to evaluate and record this severity, so as to more rigorously analyze the severity of the user's emotions;

[0011] This process ensures that the consultation system has structured assessment content for subsequent emotional state analysis and dynamic jumping;

[0012] After the questionnaire is loaded, the system will initialize the settings of the questionnaire content according to the preset emotional screening goals, including the order of questions, the scoring criteria for each question, and the dynamic jump conditions triggered by them;

[0013] The content of the example questionnaire is as follows:

[0014] Question 1: In the past two weeks, when you felt down, did you have difficulty concentrating?

[0015] Question 2: When you feel down, do you often feel worthless?

[0016] Furthermore, in step S2: Analyze the content of the user's answer through a large language model and trigger the dynamic jump logic, and adjust the presentation order of the questions according to the user's answer content, so as to achieve personalized consultation; the dynamic jump mechanism can intelligently select subsequent questions or directly jump to relevant content according to the actual feedback of the user, thus improving the accuracy and pertinence of the consultation;

[0017] The control logic of the dynamic jump includes the following content:

[0018] Emotion trigger condition recognition: The system judges whether to trigger the next question or jump to a specific consultation path according to the keywords or emotional characteristics in the user's answer; for example, when the user shows significant negative emotions, the system will guide the consultation to questions related to emotion management;

[0019] Consultation logic adjustment: For different emotional expressions, the system will automatically jump to adaptive questions; for example, if the user mentions "helplessness" or "meaninglessness", the system will directly jump to in-depth emotional assessment questions instead of continuing with simple screening questions;

[0020] Personalized question generation: With the support of large language models, the system can generate adaptable questions and provide personalized guidance based on the previous round of answers; this not only enhances the naturalness of the interrogation but also better adapts to the user's emotional expression;

[0021] The example jump logic is as follows:

[0022] Question 1: In the past two weeks, when you felt down or uninterested, did you have difficulty concentrating?

[0023] User answer: "Occasionally."

[0024] Large language model analysis: The user has a slight tendency to be distracted, recorded as "yes", and the subsequent interrogation is redirected to questions related to emotions;

[0025] Question 2: When you feel down, do you have thoughts of wanting to escape or withdraw?

[0026] User answer: "Sometimes."

[0027] Large language model analysis: There is negative emotion, and the system automatically adjusts the interrogation path according to the user's answer to further explore the root cause of the emotion.

[0028] Furthermore, in step S3: The emotional health assessment process covers multi-dimensional classification of emotional problems such as depression, lack of interest, and abnormal behavior among teenagers to achieve a comprehensive assessment of the mental health status of teenagers; based on the comprehensive analysis of the emotional state of the user's answer and combined with the preset emotional health assessment criteria, personalized suggestions are generated;

[0029] The steps for generating an emotional analysis report are as follows:

[0030] S3-1. Extract the emotional state by identifying key emotional words in the user's answer and analyze the frequency and intensity of the emotional words in the conversation;

[0031] S3-2. Quantitatively score the severity of the user's emotional state according to the questionnaire scoring criteria and the user's answer content; in order to provide accurate health advice for the user;

[0032] S3-3. Generate personalized emotional management suggestions based on the user's emotional assessment results; the suggestion content may include mental health advice, relaxation exercises, etc. to help the user better manage their emotional state;

[0033] The logical jump conditions for each question are stored in the backend database, and the system uses a logic control module to judge the next question in real time; Example jump logic:

[0034] If the user answers "yes", jump to in-depth emotional assessment questions;

[0035] If the user answers "No", jump to the regular screening questions;

[0036] The user's answer is analyzed through a large language model to extract key emotional words, such as "helpless", "anxious", etc.; the analysis process is supported by a backend microservice architecture, and service calls are completed through a service gateway; the analysis results are stored in a relational database for subsequent generation of evaluation reports;

[0037] The report template is stored in the backend, and the system fills the template content according to the analysis results; it can be directly viewed through the client or downloaded as a PDF document for preservation;

[0038] In step S3-2: According to the questionnaire scoring criteria and the user's answer content, after evaluating the severity of the user's emotional state, the additional item analysis is incorporated into the user's emotional state severity analysis system. Combining the evaluation results of the normal Q&A process, the weight division is specified. The weight ratio of the normal Q&A process is a, and the additional item score ratio is b, and the comprehensive evaluation result of the user's emotional state severity is analyzed.

[0039] A digital human intelligent consultation system for adolescent emotional health assessment, which includes a questionnaire management module, a digital human presentation module, a speech recognition module, a large language model module, a dynamic jump logic control module, and a report generation module;

[0040] The questionnaire management module is used to load the standard questionnaire content and set the dynamic jump logic of the questionnaire;

[0041] The digital human presentation module is used to generate a virtual digital human image and implement voice broadcast and lip-sync synthesis functions for natural language interaction with adolescents;

[0042] The speech recognition module is used to convert the voice answers of adolescents into text data in real time;

[0043] The large language model module is used to input the text answers of adolescents into the large language model and generate emotional health assessment results;

[0044] The dynamic jump logic control module is used to control the question jump direction according to the user's state;

[0045] The report generation module is used to generate an adolescent emotional health report based on the evaluation results and attach personalized intervention suggestions.

[0046] Further, the questionnaire management module stores questionnaire data in the SSD memory of a relational database, ensuring fast retrieval with high IOPS. The physical structure of the database storage includes fields such as question number, question text, and jump logic, enabling flexible dynamic questionnaire design. The read / write latency of the memory is 1 - 2 milliseconds, and the average throughput of data query operations can reach 1000 QPS. The user input signal is sent to the backend server via the HTTP protocol, triggering the SELECT operation of the database. The database reads the corresponding questionnaire data using the instruction decoder of the memory and returns it to the front-end module through the API interface to achieve real-time questionnaire loading and dynamic jumping.

[0047] Further, the rendering data of the 3D digital human in the digital human presentation module is stored in the GDDR6X graphics card memory. The video memory capacity is 8GB, supporting high-resolution model rendering. The key frame data of the digital human animation is stored in the form of a floating-point matrix, supporting high-frame-rate animations of 60 frames per second. The display screen realizes dynamic regulation of the brightness range through the backlight adjustment signal, supporting brightness adjustment from 200 to 500 nits. The mouth movement of the digital human is synchronized with the voice output. By generating a PWM signal, it drives the display device to refresh. The Three.js rendering engine receives the rendering parameters from the backend server and generates a 3D scene through the rasterization module of the GPU. The audio signal generated by text-to-speech is played through the speaker, while driving the lip-sync movement of the digital human model. Through delay compensation technology, the lip-sync with the sound is ensured, and the overall delay is controlled within m milliseconds, thus achieving a natural and smooth user interaction experience.

[0048] Further, in the language recognition module, the voice input receives sound wave signals through a microphone sensor, and the received sound wave signals are converted into digital signals by an analog-to-digital converter. The weight file of the speech recognition model is stored in an NVMe solid-state drive and dynamically loaded into the video memory of the GPU for processing during operation. The amplitude range of the audio signal is standardized to -1 to 1 after processing, the analog-to-digital conversion error is less than x, and the voice data rate uploaded to the server is y. The user's voice is collected by the microphone to generate an electrical signal, and the electrical signal is converted into a digital signal through ADC. After the backend ASR service receives the audio signal, it passes it into a pre-trained speech recognition deep learning model, such as Wav2Vec2.0. The model extracts speech features through a convolutional neural network and decodes them, finally outputting a text result for subsequent modules to process.

[0049] Furthermore, the text data of the user in the large language model module is transmitted to the backend server via the HTTP protocol and undergoes semantic analysis and emotional state judgment through the large language model module; the weight parameters of the large language model are stored in the video memory of the GPU in floating-point form, and the loading speed and memory access rate can meet the requirements of real-time inference; the inference process of the model is executed by the CUDA cores of the GPU; after the text data is input, it is calculated through the multi-head attention mechanism of the Transformer structure to generate the emotional classification result and the jump instruction for subsequent consultation; during the inference process, physical signals are transmitted in the form of data streams among the server, video memory, and CPU to ensure the efficient completion of tasks.

[0050] Furthermore, the dynamic jump logic control module manages the questionnaire process through a state machine, with each state corresponding to a questionnaire question; the state transition is triggered by the user input signal, and the signal type is a boolean logic signal with a value of "0" or "1"; the state table is stored in the Redis in-memory database, and the query response speed is less than 1 millisecond to ensure seamless connection for real-time consultation; each state jump is recorded through an event log for subsequent analysis and auditing; the process of dynamic jump includes emotional trigger condition recognition, consultation logic adjustment, and personalized question generation; through the query and update of the state table and the real-time analysis of the user's emotional state, the module realizes a dynamic consultation process with strong pertinence and flexible path.

[0051] Furthermore, the report generation module stores the data of the emotional health assessment report in the MySQL database, and the backend file generator module generates a PDF format file; after the file is generated, the front-end user is notified to download it via an HTTP response signal; the report generation process includes: filling the analysis results of the large language model into a preset report template, and the template content includes emotional health summary, emotional severity assessment, and personalized intervention suggestions, etc.; after the report content is generated, the file server renders it into PDF and returns it to the user side for viewing or archiving.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] Through the deep integration of multimodal interaction and large model intelligent analysis, the present invention reduces the screening cost, improves the efficiency and accuracy, and is applicable to large-scale emotional health screening scenarios, providing a forward-looking and highly adaptable solution for the emotional health management of teenagers. The present invention also relates to an electronic device and a storage medium supporting this method. The large language model used in the present invention has stronger learning ability and generalization ability compared with traditional machine learning models or traditional deep learning models, and can use self-supervised learning technology to perform pre-training on a large amount of unlabeled data, so as to capture the complex structures and patterns of language. The pre-trained model can easily adapt to specific downstream tasks through transfer learning, significantly reducing the dependence on a large amount of labeled data and improving the efficiency of model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic flowchart of the digital human intelligent consultation method for the emotional health assessment of teenagers according to the present invention;

[0055] Figure 2 It is a schematic diagram of the operation of the digital human intelligent consultation method and system for the emotional health assessment of teenagers according to the present invention;

[0056] Figure 3 It is a program flowchart of the digital human intelligent consultation system for the emotional health assessment of teenagers according to the present invention;

[0057] Figure 4 It is a module schematic diagram of the digital human intelligent consultation system for the emotional health assessment of teenagers according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0059] As Figures 1 - 4 shown, the present invention provides a technical solution, a digital human intelligent consultation method for the emotional health assessment of teenagers, including the following steps:

[0060] S1. Load a standardized questionnaire for the emotional health screening of teenagers to ensure that the questionnaire structure and logical process meet the preset specifications, so as to provide basic data support for subsequent evaluations;

[0061] S2. The virtual digital human asks questions to teenagers in the form of voice and lip movements. The voice answers of teenagers are converted into text by the voice recognition module and analyzed by the large language model;

[0062] S3. Input all the content into a large language model for comprehensive evaluation, generate a report containing a health score and personalized intervention suggestions, and provide a basis for subsequent psychological support.

[0063] In step S1: The questionnaire content includes classification questions about adolescent problems, and the questionnaire design includes jump logic, so as to dynamically adjust the order of subsequent questions according to the answers of adolescents.

[0064] The additional item analysis mechanism is added to the questionnaire question design in combination with the jump logic. The additional item analysis mechanism marks an additional item analysis each time the user's answer content triggers the jump logic to jump in the direction of in-depth negative emotion analysis, and calculates the total marked additional item scoring times n at the end of the questionnaire survey; and at the end of the final test, the total additional score is incorporated into the analysis process of the severity of the user's emotional state for analysis and generation of a comprehensive emotion assessment report; adding the additional item analysis can record the number of times the user jumps in the direction of in-depth analysis of negative emotions. The more times the user jumps in the direction of in-depth negative emotion analysis, the more serious and complex the user's negative emotions are, and an additional score is required to evaluate and record this severity, so as to more rigorously analyze the severity of the user's emotions.

[0065] This process ensures that the interrogation system has structured evaluation content for subsequent emotional state analysis and dynamic jumping.

[0066] After the questionnaire is loaded, the system will initialize the questionnaire content according to the preset emotional screening target, including the order of questions, the scoring criteria for each question, and the dynamic jump conditions triggered by them.

[0067] The content of the example questionnaire is as follows:

[0068] Question 1: In the past two weeks, when you felt down, did you have difficulty concentrating?

[0069] Question 2: When you feel down, do you often feel worthless?

[0070] The questionnaire information is stored in a relational database (MySQL) at the back end. The questionnaire content includes question ID, question text, scoring rules, and logical jump conditions; the dynamic logical jump is stored in the form of a directed graph, each node is a question, and the weight of the edge represents the trigger condition of the answer; the questionnaire management module loads the questionnaire data through the RESTful API and transmits it to the front end. The data structure design is as follows:

[0071] {

[0072] "question_id":"Q1",

[0073] "text":"Have you felt down in the dumps in the past two weeks?

[0074] "options":

[0075] {"option":"Yes","jump_to":"Q2"},

[0076] {"option":"No","jump_to":"Q3"}

[0078] }

[0079] The database server uses high-performance SSDs to ensure fast access to questionnaire data; questionnaire loading is automatically triggered by the backend service and deployed in a container environment to enhance system stability; it supports multilingual switching and version management of questionnaire content to meet the needs of different user groups.

[0080] In step S2: Analyze the content of the user's answer through a large language model and trigger the dynamic jump logic, adjust the presentation order of the questions according to the user's answer content, so as to achieve personalized consultation; the dynamic jump mechanism can intelligently select subsequent questions or directly jump to relevant content according to the user's actual feedback, thus improving the accuracy and pertinence of the consultation;

[0081] The control logic of the dynamic jump includes the following:

[0082] Emotion trigger condition recognition: The system judges whether to trigger the next question or jump to a specific consultation path according to the keywords or emotion characteristics in the user's answer; for example, when the user shows significant negative emotions, the system will guide the consultation to questions related to emotion management;

[0083] Consultation logic adjustment: For different emotion expressions, the system will automatically jump to adaptive questions; for example, if the user mentions "helplessness" or "meaninglessness", the system will directly jump to in-depth emotion assessment questions instead of continuing with simple screening questions;

[0084] Personalized question generation: With the support of a large language model, the system can generate adaptive questions and provide personalized guidance based on the previous round of answers; this not only enhances the naturalness of the consultation but also better adapts to the user's emotion expression;

[0085] The example jump logic is as follows:

[0086] Question 1: In the past two weeks, when you felt down or uninterested, did you have difficulty concentrating?

[0087] User answer: "Occasionally." ​

[0088] Analysis by the large language model: The user has a slight tendency to be distracted, recorded as "yes", and the subsequent inquiry is redirected to questions related to emotions;

[0089] Question 2: When you feel down, do you have thoughts of wanting to escape or withdraw?

[0090] User's answer: "Sometimes."

[0091] Analysis by the large language model: There is negative emotion. The system automatically adjusts the inquiry path according to the user's answer to further explore the root cause of the emotion.

[0092] In step S3: The emotional health assessment process covers multi-dimensional classification of emotional problems such as depression, lack of interest, and abnormal behavior among teenagers to achieve a comprehensive assessment of the mental health status of teenagers; Based on the comprehensive analysis of the emotional state of the user's answer, combined with the preset emotional health assessment criteria, personalized suggestions are generated;

[0093] Steps for generating an emotion analysis report are as follows:

[0094] S3-1. Extract the emotional state by identifying key emotional words in the user's answer, and analyze the frequency and intensity of the emotional words in the conversation;

[0095] S3-2. Quantitatively score the severity of the user's emotional state according to the questionnaire scoring criteria and the user's answer content; in order to provide accurate health advice for the user;

[0096] S3-3. Generate personalized emotion management suggestions according to the user's emotion assessment results; The suggested content may include mental health advice, relaxation exercises, etc. to help the user better manage their emotional state;

[0097] Example report content is as follows:

[0098] {

[0099] "emotion_summary": "The user has shown mild depression in the last two weeks and occasionally feels helpless.",

[0100] "severity_evaluation": "Mild mood swings",

[0101] "management_suggestion": "It is recommended that the user maintain a regular schedule and engage in appropriate outdoor activities. If the depression persists or worsens, it is recommended to consult a professional mental health professional."

[0102] }

[0103] In step S3-2: after evaluating the severity of the user's emotional state according to the questionnaire scoring criteria and the user's answers, the additional item analysis is incorporated into the user's emotional state severity analysis system, and combined with the evaluation results of the normal question-and-answer process, the weight division is specified, the normal question-and-answer process weight accounts for a, and the additional item score accounts for b, and the comprehensive evaluation results of the user's emotional state severity are analyzed.

[0104] A digital human intelligent consultation system for adolescent emotional health assessment, which includes a questionnaire management module, a digital human presentation module, a speech recognition module, a large language model module, a dynamic jump logic control module and a report generation module;

[0105] The questionnaire management module is used to load the standard questionnaire content and set the dynamic jump logic of the questionnaire;

[0106] The digital human presentation module is used to generate a virtual digital human image and realize voice broadcast and lip synthesis functions so as to interact with teenagers in natural language;

[0107] The speech recognition module is used to convert the teenager's voice answer into text data in real time;

[0108] The large language model module is used to input the text answer of the teenager into the large language model and generate an emotional health assessment result;

[0109] The dynamic jump logic control module is used to control the direction of the question bar according to the user status;

[0110] The report generation module is used to generate a youth emotional health report based on the assessment results, along with personalized intervention suggestions.

[0111] Furthermore, the questionnaire management module stores the questionnaire data in the SSD memory of the relational database, and uses high IOPS to ensure fast retrieval; the physical structure of the database storage includes the fields of question number, question text and jump logic, so as to support flexible dynamic questionnaire design; the read and write delay of the memory is 1-2 milliseconds, and the average throughput of data query operations can reach 1000QPS; the user input signal is sent to the back-end server via the HTTP protocol, triggering the SELECT operation of the database; the database uses the instruction decoder of the memory to read the corresponding questionnaire data, and returns it to the front-end module through the API interface, realizing real-time questionnaire loading and dynamic jump.

[0112] The rendering data of the 3D digital human in the digital human presentation module is stored in the GDDR6X graphics card memory; the video memory capacity is 8GB, which is used to support high-resolution model rendering; the key frame data of the digital human animation is stored in the form of a floating-point matrix, supporting 60 high-frame-rate animations per second; the display screen realizes dynamic regulation of the brightness range through the backlight adjustment signal, supporting brightness adjustment from 200 to 500 nits; the mouth movement of the digital human is synchronized with the voice output. By generating a PWM signal, it drives the display device to refresh, and the PWM signal has a frequency of 10 kHz; the Three.js rendering engine receives rendering parameters from the backend server and generates a 3D scene through the rasterization module of the GPU; the audio signal generated by text-to-speech is played through the speaker, while driving the lip-sync movement of the digital human model; through the delay compensation technology, the lip-sync with the sound is ensured, and the overall delay is controlled within 20 milliseconds, thus achieving a natural and smooth user interaction experience.

[0113] In the language recognition module, the voice input receives the sound wave signal through the microphone sensor. The sensitivity of the microphone is -42dB, which can accurately capture the user's voice content; the received sound wave signal is converted into a digital signal by the analog-to-digital converter, and the sampling rate is set to 16kHz to maintain high voice quality; the weight file of the voice recognition model is stored in the NVMe solid-state drive and dynamically loaded into the video memory of the GPU for processing during operation; the amplitude range of the audio signal is standardized to -1 to 1 after processing, and the analog-to-digital conversion error is less than 0.1dB. The voice data rate uploaded to the server is 128kbps; the user's voice is collected by the microphone to generate an electrical signal, and the electrical signal is converted into a digital signal through the ADC. After the backend ASR service receives the audio signal, it passes it into a pre-trained voice recognition deep learning model, such as Wav2Vec2.0; the model extracts voice features through a convolutional neural network and decodes them, and finally outputs the text result for subsequent modules to process.

[0114] In the large language model module, the user's text data is transmitted to the backend server through the HTTP protocol for semantic analysis and emotional state judgment by the large language model module; the weight parameters of the large language model are stored in the form of floating-point numbers in the video memory of the GPU. The model size is about 20GB, and the loading speed and memory access rate can meet the real-time inference requirements; the inference process of the model is executed through the CUDA cores of the GPU, and the matrix operation processing time is controlled within 200 milliseconds; after the text data is input, through the multi-head attention mechanism of the Transformer structure, the emotion classification result and the jump instruction for subsequent consultation are generated; during the inference process, physical signals are transmitted in the form of data streams between the server, video memory, and CPU to ensure the efficient completion of tasks.

[0115] The dynamic jump logic control module manages the questionnaire process through a state machine, with each state corresponding to a questionnaire question; the state transition is triggered by a user input signal, the signal type is a boolean logic signal, and the value is "0" or "1"; the state table is stored in the Redis in-memory database, and the query response speed is less than 1 millisecond to ensure seamless connection for real-time medical consultations; each state jump is recorded through an event log for subsequent analysis and auditing; the process of dynamic jump includes emotional trigger condition identification, medical consultation logic adjustment, and personalized question generation; through querying and updating the state table and combining with real-time analysis of the user's emotional state, the module realizes a highly targeted and flexible dynamic medical consultation process.

[0116] The report generation module stores the data of the emotional health assessment report in the MySQL database, and the backend file generator module generates a PDF format file; after the file is generated, the front-end user is notified to download it through an HTTP response signal; the report generation time is controlled within 500 milliseconds, and the data format size of a single report is about 50KB; the report generation process includes: filling the analysis results of the large language model into a preset report template, and the template content includes emotional health summary, emotional severity assessment, and personalized intervention suggestions, etc.; after the report content is generated, the file server renders it into PDF and returns it to the user side for viewing or archiving.

[0117] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

Claims

1. A digital human intelligent consultation method for adolescent emotional health assessment, characterized by: The following steps are involved: S1. Collect and load the standardized adolescent emotional health screening questionnaire to ensure that the questionnaire structure and logical process meet the preset specifications to provide basic data support for subsequent evaluation; S2, ask questions to teenagers in the form of voice and lip shape through virtual digital people, and the teenagers' voice answers are converted into text by the speech recognition module and analyzed using the large language model; S3. All content is input into a large language model for comprehensive evaluation, generating a report containing health scores and personalized intervention recommendations.

2. The digital human intelligent diagnosis method for adolescent emotional health assessment according to claim 1 is characterized by: In step S1: the questionnaire content includes categorized questions about adolescent issues, and the questionnaire design includes jump logic, so as to dynamically adjust the order of subsequent questions according to the adolescents' answers; An additional item analysis mechanism is added to the jump logic in the questionnaire question design. Each time the user's answer triggers the jump logic to jump to the in-depth negative emotion analysis direction, the additional item analysis is marked once, and the total number of marked additional item scores n is calculated at the end of the questionnaire survey; and at the end of the final test, the total additional score is included in the severity analysis process of the user's emotional state, which is used to analyze and generate a comprehensive emotional assessment report.

3. The digital human intelligent diagnosis method for adolescent emotional health assessment according to claim 1 is characterized by: In step S2: the user's answer is analyzed through a large language model, and the dynamic jump logic is triggered to adjust the presentation order of questions according to the user's answer, thereby realizing personalized consultation; the dynamic jump mechanism can intelligently select follow-up questions or directly jump to related content based on the user's actual feedback; The control logic of dynamic jump includes the following: Emotional trigger condition recognition: The system determines whether to trigger the next question or jump to a specific consultation path based on the keywords or emotional characteristics in the user's answer; Adjustment of the diagnosis logic: the system will automatically jump to adaptive questions for different emotional expressions; Personalized question generation: With the support of a large language model, the system can generate adaptive questions and provide personalized guidance based on the previous round of answers.

4. The digital human intelligent diagnosis method for adolescent emotional health assessment according to claim 1 is characterized by: In step S3: the emotional health assessment process covers the multi-dimensional classification of emotional problems of adolescents such as depression, lack of interest and abnormal behavior; based on a comprehensive analysis of the emotional state of the user's answers and combined with the preset emotional health assessment standards, personalized suggestions are generated; The steps to generate a sentiment analysis report are: S3-1, extracting emotional states by identifying key emotional words in the user's answer, and analyzing the frequency and intensity of the emotional words in the conversation; S3-2, assessing the severity of the user's emotional state based on the questionnaire scoring criteria and the user's responses; S3-3, generating personalized emotion management suggestions based on the user's emotion assessment results; In step S3-2: after evaluating the severity of the user's emotional state according to the questionnaire scoring criteria and the user's answers, the additional item analysis is incorporated into the user's emotional state severity analysis system, and combined with the evaluation results of the normal question-and-answer process, the weight division is specified, the normal question-and-answer process weight accounts for a, and the additional item score accounts for b, and the comprehensive evaluation results of the user's emotional state severity are analyzed.

5. The digital human intelligent consultation system for adolescent emotional health assessment is characterized by: The system includes a questionnaire management module, a digital human presentation module, a speech recognition module, a large language model module, a dynamic jump logic control module and a report generation module; The questionnaire management module is used to load the standard questionnaire content and set the dynamic jump logic of the questionnaire; The digital human presentation module is used to generate a virtual digital human image and realize voice broadcast and lip synthesis functions so as to interact with teenagers in natural language; The speech recognition module is used to convert the teenager's voice answer into text data in real time; The large language model module is used to input the text answer of the teenager into the large language model and generate an emotional health assessment result; The dynamic jump logic control module is used to control the question jump direction according to the user status; The report generation module is used to generate a youth emotional health report based on the assessment results, along with personalized intervention suggestions.

6. The digital human intelligent consultation system for adolescent emotional health assessment according to claim 5 is characterized by: The questionnaire management module stores the questionnaire data in the SSD memory of the relational database, and uses high IOPS to ensure fast retrieval; the user input signal is sent to the back-end server via the HTTP protocol, triggering the SELECT operation of the database; the database uses the instruction decoder of the memory to read the corresponding questionnaire data, and returns it to the front-end module through the API interface, realizing real-time questionnaire loading and dynamic jumping.

7. The digital human intelligent consultation system for adolescent emotional health assessment according to claim 5 is characterized by: The rendering data of the three-dimensional digital human of the digital human presentation module is stored in the GDDR6X graphics card memory; the key frame data of the digital human animation is stored in the form of a floating-point matrix, the display screen realizes dynamic regulation of the brightness range through the backlight adjustment signal, the digital human's mouth movement is synchronized with the voice output, and the display device is driven to refresh by generating a PWM signal; The Three.js rendering engine receives rendering parameters from the backend server and generates a three-dimensional scene through the GPU's rasterization module. The audio signal generated by text-to-speech is played through the speaker, while driving the lip movements of the digital human model to synchronize. The synchronization of lip movement and sound is ensured through delay compensation technology, and the overall delay is controlled within m milliseconds.

8. The digital human intelligent consultation system for adolescent emotional health assessment according to claim 5 is characterized by: In the language recognition module, the voice input receives the sound wave signal through the microphone sensor, and the received sound wave signal is converted into a digital signal through the analog-to-digital converter. The weight file of the voice recognition model is stored in the NVMe solid-state hard disk and dynamically loaded into the GPU memory for processing during operation; the amplitude range of the audio signal is -1 to 1 after standardization, the analog-to-digital conversion error is lower than x, and the voice data rate uploaded to the server is y; the user's voice is collected by the microphone to generate an electrical signal, which is converted into a digital signal through the ADC. After the back-end voice recognition service receives the audio signal, it passes it into the pre-trained voice recognition deep learning model; the model extracts and decodes voice features through a convolutional neural network, and finally outputs a text result.

9. The digital human intelligent consultation system for adolescent emotional health assessment according to claim 5 is characterized by: The user's text data in the large language model module is transmitted to the back-end server via the HTTP protocol, and semantic analysis and emotional state judgment are performed through the large language model module; the weight parameters of the large language model are stored in the GPU's video memory in the form of floating-point numbers, and the loading speed and memory access rate can meet the real-time reasoning requirements; the reasoning process of the model is executed by the GPU's CUDA core, and after the text data is input, it is calculated through the Transformer structure's multi-head attention mechanism to generate emotion classification results and jump instructions for subsequent consultations; during the reasoning process, physical signals are transmitted between the server, video memory and CPU in the form of data streams.

10. The digital human intelligent consultation system for adolescent emotional health assessment according to claim 5 is characterized by: The dynamic jump logic control module manages the questionnaire process through a state machine, and each state corresponds to a questionnaire question; the state switch is triggered by a user input signal, and the signal type is a Boolean logic signal with a value of "0" or "1"; The status table is stored in the Redis memory database, and the query response speed is less than 1 millisecond, ensuring seamless connection of real-time consultation; Each state jump is recorded through the event log; the dynamic jump process includes emotion trigger condition identification, consultation logic adjustment and personalized question generation.

Images