Empathy guidance and psychological problem early warning method based on adolescent-specific psychological model

Abstract

The application provides a kind of empathy guidance and psychological problem early warning method based on the psychological model of adolescents, by constructing the semantic aggregation vector index containing oral expansion, the understanding ability and retrieval accuracy of the system for unstructured language of adolescents are significantly improved, the accuracy of psychological evaluation is ensured.Using double-track parallel architecture not only ensures the objectivity and quantifiability of the evaluation results by using logical enhancement model, but also improves user experience and compliance by using special adolescent psychological model, solving the technical contradiction that single model is difficult to balance accuracy and empathy.The application embeds the model and vector index through local deployment, and combines dynamic state tracking and C-SSRS-based crisis fuse mechanism, realizes real-time monitoring and automatic intervention of high-risk psychological risk on the premise of fully protecting user data privacy, greatly improves the safety and reliability of the system, and provides full-chain technical support for adolescent mental health monitoring.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and mental health support technology, and in particular to a method for empathic guidance and early warning of psychological problems based on a psychological model specific to adolescents. Background Technology

[0002] Currently, the mental health of adolescents is receiving increasing attention from society. Traditional mental health assessments mainly rely on standardized psychological scales (such as SAS and SDS) or offline professional psychological counseling. However, this model has significant limitations when dealing with adolescents. First, adolescents' cognitive development is not yet mature, and their psychological problems are often hidden and fluctuating, making it difficult for parents and teachers to detect them in a timely manner. Second, traditional scales are rigid in format and their questions are straightforward, easily triggering resistance from adolescents, leading to low compliance and difficulty in obtaining genuine feedback. With the development of artificial intelligence technology, conversational psychological assistance systems based on general large language models have emerged. While these systems can conduct natural and fluent conversations, their core flaws are as follows: First, the general models lack a deep understanding of psychometric expertise, and their assessment results are often based on semantic "illusions" rather than rigorous quantitative standards, resulting in insufficient reliability and validity. Second, the service model relying on cloud APIs poses a serious risk of user data privacy leaks, especially for sensitive mental health data, where privacy protection is a core pain point. Third, existing systems generally lack real-time identification and intervention mechanisms for extreme psychological crises (such as suicidal ideation and self-harming behavior). When users reveal high-risk signals in conversations, the system cannot promptly shut down and switch to professional intervention mode, posing significant ethical and security vulnerabilities.

[0003] Furthermore, although retrieval-enhanced generation (RAG) techniques are used to improve the accuracy of models' expertise, conventional RAG systems often suffer from poor retrieval accuracy when faced with the colloquial and unstructured expressions of adolescents, making it difficult to accurately associate statements like "I've been having trouble sleeping lately" with the "sleep disorder" item on standard scales. At the same time, how to balance the "cold" precision of logical reasoning with the "warm" empathy of emotional interaction within a single system remains a challenge that current technologies have not yet solved.

[0004] Therefore, there is an urgent need in this field for an adolescent mental health assessment system that combines the rigor of psychometrics, the interactivity of large language models, and the security of data privacy, and has the ability to intervene in real time crises, in order to solve the problems of inaccurate assessment, privacy leakage and lack of intervention mechanisms in the existing technologies mentioned above. Summary of the Invention

[0005] The purpose of this invention is to provide a method for empathic guidance and early warning of psychological problems based on a psychological model specific to adolescents, so as to solve the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides the following solution: This invention provides a method for empathy guidance and early warning of psychological problems based on a specific psychological model for adolescents, comprising: Construct a multi-source heterogeneous psychological knowledge base, which includes a semantic aggregation vector index formed by colloquially expanding the official entries of standard psychological scales; In response to user input, a dual-track parallel processing architecture is initiated, in which... The first track is an implicit analysis stream, which performs semantic retrieval and matching on the input information based on the vector index, and outputs a structured psychological assessment report based on the matching results; The second track is the explicit dialogue flow, which dynamically adjusts the dialogue strategy based on the psychological assessment data and calls upon a trained adolescent-specific psychological model to generate empathetic responses. The user's psychological state matrix is ​​maintained in real time by a state tracker, and the psychological state matrix is ​​updated based on the psychological assessment data. When the psychological state matrix meets the preset crisis triggering conditions, a circuit breaker operation is executed to interrupt the regular response of the explicit dialogue flow and initiate the crisis intervention process.

[0007] Preferably, the step of constructing a multi-source heterogeneous psychological knowledge base includes: We acquired data from standard psychological scales, including the Self-Rating Anxiety Scale (SAS), the Self-Rating Depression Scale (SDS), and the Columbia Suicide Severity Rating Scale (C-SSRS), and extracted the official entry texts, corresponding scoring rules, and crisis markers. For each official entry text, multiple semantically equivalent colloquial expressions for teenagers are generated using a preset prompt word engineering or manual annotation method. The official entry text is combined with its corresponding multiple colloquial expressions to form a single semantic descriptor; The semantic descriptor is vectorized and encoded using a locally deployed embedding model to generate the semantic aggregation vector index, thereby enabling offline loading of the system.

[0008] Preferably, the implicit analysis stream performs semantic retrieval and matching on the input information based on the vector index, and outputs structured psychological assessment data according to the matching results, including: The input information is converted into a query vector, and a similarity search is performed in the vector index to obtain the Top-K candidate symptom entries with the highest similarity. The input information and the Top-K candidate symptom entries are input into a logic-enhanced large language model, which determines whether the input information matches the candidate symptom entries and evaluates the symptom intensity level when it matches. The logic-enhanced large language model outputs structured psychological assessment data in JSON format, containing the hit entry ID and intensity level.

[0009] Preferably, the explicit dialogue flow dynamically adjusts the dialogue strategy based at least on the psychological assessment data, including: Read the psychological state matrix maintained in the state tracker, which includes the user's raw score, standard score and risk level on each scale; The dynamic prompt word generator selects or generates corresponding system prompt words from a preset prompt word library based on the risk level. These system prompt words are used to guide the response style and content of the adolescent-specific mental model.

[0010] Preferably, the adolescent-specific psychological model is a large-scale vertical domain model specializing in adolescent mental health, obtained through fine-tuning and reinforcement training based on a large-scale multi-turn empathic dialogue dataset.

[0011] Preferably, the crisis triggering conditions include: The structured psychological assessment data output by the implicit analysis stream was detected to contain high-risk items preset from the C-SSRS scale; Alternatively, a risk score in the psychological state matrix maintained in the state tracker exceeds a preset safety threshold.

[0012] Preferably, the step of performing the circuit breaker operation includes: Record critical alert logs in the background, including timestamps, user input, and trigger entries; The system prompts that force a switch to the explicit dialogue flow are set to the crisis intervention expert mode. In this mode, the prompts are used to guide the adolescent-specific psychological model to output intervention scripts that include information on psychological assistance hotlines and safety confirmation inquiries.

[0013] Preferably, the method further includes a report generation step: At the end of the dialogue, a deterministic algorithm for non-large language models is invoked to perform scale scoring, norm comparison, and level determination on the psychological assessment data recorded in the state tracker, so as to obtain a deterministic quantitative calculation result. The quantization results are used as immutable contextual information to construct prompt words and input them into the large language model; Based on the quantitative calculation results, the large language model generates a natural language mental health assessment report that includes a results summary, symptom analysis, and suggested guidance.

[0014] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the aforementioned empathy guidance and psychological problem early warning method based on a adolescent-specific psychological model.

[0015] The present invention also provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the steps of the described method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents.

[0016] The present invention achieves the following beneficial technical effects compared to the prior art: This invention provides a method for empathic guidance and early warning of psychological problems based on a specific adolescent psychological model. It constructs a semantic aggregation vector index with colloquial extensions and employs a dual-track parallel architecture that separates rigorous quantitative analysis (implicit flow) from warm and natural empathic interaction (explicit flow). This approach utilizes a logic-enhanced model to ensure the objectivity and quantifiability of assessment results while enhancing user experience and compliance through a specific adolescent psychological model. Furthermore, it employs localized deployment of the embedded model and vector index, combined with dynamic state tracking and a crisis circuit breaker mechanism based on C-SSRS. These methods significantly improve the system's ability to understand unstructured language from adolescents and enhance retrieval accuracy, ensuring the accuracy of psychological assessments. This invention resolves the technical contradiction of balancing accuracy and empathy in a single model. While fully protecting user data privacy, it achieves real-time monitoring and automatic intervention for high-risk psychological issues, greatly improving the system's security and reliability, and providing end-to-end technical support for adolescent mental health monitoring. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of the empathy guidance and psychological problem early warning method based on a specific psychological model for adolescents provided by this invention; Figure 2 This is a schematic diagram illustrating the semantic aggregation vectorization principle in the empathy guidance and psychological problem early warning method based on a adolescent-specific psychological model provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the psychological problem early warning method based on semantic aggregation vector index provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the empathic guided dialogue system based on a specific psychological model for adolescents, as described in this invention.

[0021] Figure 5 This is a simplified diagram showing the sequential workflow of key nodes in each system of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The purpose of this invention is to address the technical problems in existing technologies for adolescent mental health assessment, such as low assessment accuracy, high risk of data privacy leakage, and lack of real-time crisis intervention mechanisms. It provides a method for empathic guidance and early warning of mental health problems based on a specific adolescent psychological model. This method constructs a semantic aggregation vector index that integrates standard psychological scales and colloquial expressions, and combines a dual-track parallel processing architecture with a dynamic crisis circuit breaker mechanism. This achieves accurate quantitative assessment and humanized empathic interaction while ensuring data privacy, significantly improving the intelligence and security of adolescent mental health monitoring.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Example 1: like Figure 1The diagram illustrates the overall process of an empathy guidance and psychological problem early warning method based on a specific adolescent psychological model, as provided in an embodiment of the present invention. The method first executes step S1, constructing a multi-source heterogeneous psychological knowledge base. Specifically, this knowledge base is not simply data storage, but rather deeply integrates the official entries of standard psychological scales with multi-dimensional colloquial expressions used by adolescents through an innovative "semantic aggregation" strategy. For example, for the official entry "I don't sleep well at night" in the SDS scale, the system will pre-set or generate multiple colloquial expressions commonly used by adolescents in natural conversation, such as "I've been having insomnia lately," "I can't sleep because I stay up late," and "I have heavy dark circles under my eyes." Subsequently, the official text and these colloquial texts are concatenated into a complete semantic descriptor, and a locally deployed embedding model (such as bge-m3) is used to vectorize and encode it, ultimately generating a high-dimensional semantic vector index. This approach allows subsequent retrieval to no longer rely on literal keyword matching, but rather to deeply understand the semantic connotation of user input, significantly improving the ability to recognize unstructured and implicit expressions of adolescents. Meanwhile, all vectorization processing is completed locally, and the generated binary index file (such as .npy format) can achieve offline loading and second-level cold start of the system, fundamentally eliminating the risk of privacy leakage caused by uploading data to the cloud.

[0026] After completing the construction of the knowledge base, the method proceeds to step S2, establishing a dual-track parallel processing architecture based on the vector index. For example... Figure 2 As shown, this architecture is one of the core innovations of this invention. It uses a dual-track dispatch scheduler to simultaneously distribute user input information to two independent yet collaborative subsystems: an implicit analysis stream (early warning system) and an explicit dialogue stream (interaction system). This design cleverly separates the logically rigorous quantitative analysis task from the warm and natural empathetic interaction task, allowing different types of models to handle them separately.

[0027] For implicit analysis, it is completely invisible to the user and runs silently in the background. When user input is received (such as the text "I've been feeling very stressed lately and can't calm down"), the system first converts the input into a query vector and performs a similarity search in the semantic aggregation vector index built in step S1 to obtain the Top-K candidate symptom entries (e.g., K=4) that are semantically most similar. This process utilizes Retrieval Augmentation (RAG) technology to ensure the relevance of subsequent analysis. Next, the system feeds these candidate entries, along with the original user input, into a logic-enhanced large language model that emphasizes logical reasoning and instruction following. The model's task is to accurately determine whether the user input matches a candidate entry and, if so, assess its symptom intensity level (e.g., level 1-4). To ensure that downstream modules can stably and accurately parse the processing results, this invention requires that the logic-enhanced model output strictly formatted JSON data, such as {"hit_entries": [{"id": "SAS_01", "intensity": 3}]}, clearly indicating the ID of the hit entry and the corresponding intensity level, thereby realizing the transformation of ambiguous natural language into accurate structured psychological assessment data.

[0028] Meanwhile, the second track's explicit dialogue flow is responsible for front-end interaction with the user. At the core of this flow is a specially trained adolescent-specific psychological model (which can be named the "Muxin Model" in this embodiment). This model is not a general dialogue model, but a large-scale, vertically-domain model fine-tuned and reinforced based on a large-scale, high-quality multi-turn empathic dialogue dataset. Its training corpus deeply covers typical confusions, emotional fluctuations, and real-life psychological counseling cases faced by adolescents during their growth, enabling the model to embed rich psychological knowledge and coping strategies, possessing expert-level emotion perception and empathic interaction capabilities. Before generating a response, the explicit dialogue flow reads the user's psychological state matrix maintained in the state tracker in real time and uses a dynamic prompt word generator to dynamically adjust the system prompt words input to the proprietary model based on the current risk level. For example, if the state is normal, the prompt words guide the model to engage in regular, supportive dialogue; if anxiety risk is detected, the prompt words will inject caring instructions such as "Please use anxiety management strategies to respond," thereby indirectly and effectively guiding the model's output style and content.

[0029] Next, the method proceeds to step S3, implementing dynamic state tracking and crisis circuit breaking. The StateTracker is the key hub connecting the implicit analysis flow and the explicit dialogue flow. It internally caches detailed scoring rules for all psychological scales (SAS, SDS, C-SSRS) (such as positive and negative scoring, raw score and standard score conversion formulas, norm judgment thresholds, etc.). Whenever the implicit analysis flow outputs new structured assessment data, the StateTracker receives this data and calculates and updates the user's psychological state matrix in real time according to the preset scoring logic. This matrix dynamically records the user's raw score, standard score, and the resulting risk level (e.g., no risk, mild, moderate, severe) on each scale during the current session. The system has a built-in circuit breaking logic based on the C-SSRS scale, which monitors in real time whether items with specific "crisis markers" are triggered. Once the system detects that the user's input matches the semantics of a high-risk C-SSRS item, such as "I want to end my life," or that the accumulated risk score (e.g., depression standard score) exceeds the preset safety threshold, the system immediately triggers the circuit breaking operation. This operation comprises three simultaneous actions: First, a critical alert log at the CRITICAL level is recorded in the background, including a timestamp, the user's original input, and the triggering item. Second, the regular response generation process of the explicit dialogue flow is forcibly interrupted. Finally, through a dynamic prompt word generator, the system prompt words of the proprietary mental model are overridden with the highest privileges, forcibly switching to "Crisis Intervention Expert Mode." In this mode, the instructions received by the model are no longer ordinary empathetic companionship, but rather it is guided to output standard intervention scripts including authoritative psychological assistance hotlines and safety confirmation inquiries (such as "Are you safe now?"), ensuring that the model can provide the user with the correct help guidance at critical moments.

[0030] Furthermore, a preferred embodiment of the present invention also includes a report generation step under data constraints. At the natural end of each conversation, the system does not rely on the large language model for numerical calculations. Instead, it first invokes a deterministic algorithm that is not based on the large model, strictly adhering to the accumulated data recorded in the state tracker and the scale scoring rules, to perform precise numerical calculations, total score statistics, and normative level determination, resulting in an accurate quantitative calculation result. Subsequently, the system embeds these deterministic calculation results as an immutable and unshakeable "factual context" into a specially designed prompt word before feeding it into the large language model. The large language model here acts as a "text polisher," generating a professional and readable mental health assessment report based on this accurate data. The report may include a results summary, symptom analysis, and personalized suggestions. This hybrid model of "logical calculation + text polishing" fundamentally eliminates the "illusion" risk of numerical calculations in the large language model, ensuring the authority and reliability of the assessment report.

[0031] To more clearly illustrate the technical details of this invention, the following description is provided in conjunction with... Figure 3 The "semantic aggregation vectorization" process in step S1 is explained in detail. For example... Figure 3 As shown, this invention first obtains raw data from professional psychological assessment scales, such as SAS, SDS, and C-SSRS. These scales contain official entry text, item numbers, scoring directions (positive / negative), and special crisis markers. Subsequently, the system uses a cue word engineering approach, calling a large language model or combining it with manual annotation, to verbally expand each official entry. For example, for SAS scale entry 1, "I feel more nervous and anxious than usual," the system generates multiple expressions that teenagers might use in real-life situations, such as "I've been particularly nervous lately," "I'm always anxious," "I feel restless," "I can't calm down," and "I feel inexplicably flustered." These verbal expressions greatly enrich the semantic coverage of the original entries. Next, the system concatenates the official text with all the generated verbal texts to form a complete semantic descriptor. Finally, a locally deployed general embedding model (such as bge-m3) is used to vectorize and encode this semantic descriptor, and the encoded results of all entries, along with their metadata (ID, scoring rules, crisis markers, etc.), are stored as a binary vector index file. In this way, when a user enters "I've been feeling particularly stressed lately", their query vector can be highly matched in semantic space with the semantic descriptor vector containing the entry "SAS_01", thus achieving accurate retrieval.

[0032] In a specific application example, a user inputs "I've been feeling very stressed lately and can't calm down" through the front-end interface. The system backend then... Figure 2The logical structure shown is processed. The dispatcher sends the input to both the early warning system and the interaction system simultaneously. In the early warning system, the semantic retrieval device quickly matches the "SAS_01" item on the SAS scale, the logic enhancement model determines the intensity level as "high," and outputs JSON data. The state tracker updates the user's psychological state matrix accordingly, and the SAS anxiety score increases. Since the score has not yet reached the circuit breaker threshold, the dynamic prompt word generator reads the risk of "mild anxiety" and sends prompt words containing anxiety relief strategies to the "Muxin Model" in the interaction system. The Muxin Model generates an empathetic response based on this, such as "It sounds like you've been under a lot of pressure lately. It must be tiring to be so tense all the time. Would you like to tell me more about what's making you feel nervous?" This response is checked for harmlessness by the safety and compliance filter and then displayed to the user. At the same time, the state tracker starts an observation period based on a sliding time window to continuously monitor subsequent input. If the user continues to express high levels of anxiety in subsequent conversations and the accumulated score exceeds the safety threshold, the system will immediately trigger a circuit breaker, forcibly switch the prompt words, and guide the Muxin model to output intervention scripts containing content such as "You can try calling the psychological assistance hotline 12320 for professional help. Would you like to confirm with me whether you are safe now?" and send a warning notification to the preset guardian in the background.

[0033] In summary, this invention, through a series of innovative designs such as semantic aggregation vector indexing, dual-track parallel processing architecture, dynamic state tracking, and crisis circuit breaker mechanism, effectively solves the problems of low assessment accuracy, insufficient data privacy protection, and lack of real-time intervention capabilities in existing technologies, and realizes all-weather, precise, and humanized intelligent monitoring of the mental health status of adolescents.

[0034] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0035] It should be noted that the components mentioned in the above embodiments are all general standard parts or components known to those skilled in the art. Their structures and principles can be learned by those skilled in the art through technical manuals or conventional experimental methods.

[0036] This invention has illustrated its principles and implementation methods using specific examples. The descriptions of these embodiments are merely illustrative of the method and its core ideas; furthermore, those skilled in the art will recognize that modifications may be made to the specific implementation methods and application scope based on the principles of this invention. Therefore, the content of this specification should not be construed as limiting the invention.

Claims

1. A method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents, characterized in that: include: Construct a multi-source heterogeneous psychological knowledge base, which includes a semantic aggregation vector index formed by colloquially expanding the official entries of standard psychological scales; In response to user input, a dual-track parallel processing architecture is initiated, in which... The first track is an implicit analysis stream, which performs semantic retrieval and matching on the input information based on the vector index, and outputs a structured psychological assessment report based on the matching results; The second track is the explicit dialogue flow, which dynamically adjusts the dialogue strategy based on the psychological assessment data and calls upon a trained adolescent-specific psychological model to generate empathetic responses. The user's psychological state matrix is ​​maintained in real time by a state tracker, and the psychological state matrix is ​​updated based on the psychological assessment data. When the psychological state matrix meets the preset crisis triggering conditions, a circuit breaker operation is executed to interrupt the regular response of the explicit dialogue flow and initiate the crisis intervention process.

2. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 1, characterized in that, The steps for constructing a multi-source heterogeneous psychological knowledge base include: We acquired data from standard psychological scales, including the Self-Rating Anxiety Scale (SAS), the Self-Rating Depression Scale (SDS), and the Columbia Suicide Severity Rating Scale (C-SSRS), and extracted the official entry texts, corresponding scoring rules, and crisis markers. For each official entry text, multiple semantically equivalent colloquial expressions for teenagers are generated using a preset prompt word engineering or manual annotation method. The official entry text is combined with its corresponding multiple colloquial expressions to form a single semantic descriptor; The semantic descriptor is vectorized and encoded using a locally deployed embedding model to generate the semantic aggregation vector index, thereby enabling offline loading of the system.

3. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 1, characterized in that, The implicit analysis stream performs semantic retrieval and matching on the input information based on the vector index, and outputs structured psychological assessment data based on the matching results, including: The input information is converted into a query vector, and a similarity search is performed in the vector index to obtain the Top-K candidate symptom entries with the highest similarity. The input information and the Top-K candidate symptom entries are input into a logic-enhanced large language model, which determines whether the input information matches the candidate symptom entries and evaluates the symptom intensity level when it matches. The logic-enhanced large language model outputs structured psychological assessment data in JSON format, containing the hit entry ID and intensity level.

4. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 1, characterized in that, The explicit dialogue flow dynamically adjusts the dialogue strategy based at least on the psychological assessment data, including: Read the psychological state matrix maintained in the state tracker, which includes the user's raw score, standard score and risk level on each scale; The dynamic prompt word generator selects or generates corresponding system prompt words from a preset prompt word library based on the risk level. These system prompt words are used to guide the response style and content of the adolescent-specific mental model.

5. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 4, characterized in that, The adolescent-specific psychological model is a large-scale vertical model specializing in adolescent mental health, obtained through fine-tuning and reinforcement training based on a large-scale multi-turn empathic dialogue dataset.

6. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 1, characterized in that, The crisis triggering conditions include: The structured psychological assessment data output by the implicit analysis stream was detected to contain high-risk items preset from the C-SSRS scale; Alternatively, a risk score in the psychological state matrix maintained in the state tracker exceeds a preset safety threshold.

7. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 6, characterized in that, The circuit breaker operation includes: Record critical alert logs in the background, including timestamps, user input, and trigger entries; The system prompts that force a switch to the explicit dialogue flow are set to the crisis intervention expert mode. In this mode, the prompts are used to guide the adolescent-specific psychological model to output intervention scripts that include information on psychological assistance hotlines and safety confirmation inquiries.

8. The method for empathic guidance and early warning of psychological problems based on a specific psychological model for adolescents as described in claim 1, characterized in that, The method also includes a report generation step: At the end of the dialogue, a deterministic algorithm for non-large language models is invoked to perform scale scoring, norm comparison, and level determination on the psychological assessment data recorded in the state tracker, so as to obtain a deterministic quantitative calculation result. The quantization results are used as immutable contextual information to construct prompt words and input them into the large language model; Based on the quantitative calculation results, the large language model generates a natural language mental health assessment report that includes a results summary, symptom analysis, and suggested guidance.

9. An electronic device, including a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the empathy guidance and psychological problem early warning method based on a specific adolescent psychological model as described in any one of claims 1 to 8.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that, When executed by the processor, this instruction implements the steps of the empathy guidance and psychological problem early warning method based on a adolescent-specific psychological model as described in any one of claims 1 to 8.

Images