Mental health assessment system based on large model context protocol and implementation method thereof

Abstract

The invention provides a psychological health assessment system based on a large model context protocol and an implementation method thereof, and relates to the technical field of artificial intelligence content generation, and the system comprises a large model module, an MCP tool module, a large model context protocol module, an Agent module and a tested end module; wherein the Agent module utilizes the large model module and utilizes the MCP tool module through the large model context protocol module to carry out psychological health assessment interaction with the tested end module. According to the method, the psychological health assessment table and the large language model Agent are deeply integrated, a dynamic and personalized assessment method based on the MCP protocol is adopted, and the limitation of a traditional assessment mode is broken through. The scale problem is naturally embedded into the dialogue process, so that the system not only can keep the credibility of a standardized scale, but also can dynamically adjust the evaluation content according to the real-time physiological indexes and the dialogue context of the tested person, and personalized and high-interaction psychological state analysis is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence content generation technology, and in particular to a mental health assessment system based on a large model context protocol and an implementation method thereof. Background Art

[0002] Currently, current mental health assessment systems primarily rely on standardized scales, but this traditional assessment method has certain limitations. First, paper or electronic questionnaires are susceptible to response bias and contextual dependence. While automated scoring systems based on rule engines improve assessment efficiency, their rigid structure cannot handle complex descriptions of mental states. Furthermore, the application of modern large language models in mental health assessment still faces certain technical bottlenecks, including a lack of structured indicators in the output, an inability to ensure consistent scale application, and a lack of sufficient interpretability of the assessment process, which can affect the credibility of clinical decision-making.

[0003] Therefore, there is an urgent need for a new technical means that can better combine the advantages of mental health assessment and large language models to overcome the shortcomings of existing technologies. Summary of the Invention

[0004] One of the purposes of the present invention is to provide a mental health assessment system based on a large model context protocol to solve the problems in the background technology.

[0005] An embodiment of the present invention provides a mental health assessment system based on a large model context protocol, comprising:

[0006] Large model module, MCP tool module, large model context protocol module, Agent module and tested end module;

[0007] Among them, the Agent module uses the big model module and the MCP tool module through the big model context protocol module to interact with the tested end module for mental health assessment.

[0008] Optionally, the Agent module performs the following steps:

[0009] Based on the large model module, according to the user basic information of the MCP tool module, an opening statement is generated and sent to the tested end module, and the first tested end response is obtained;

[0010] Based on the large model module, determine whether to use the psychometric scale according to the first test subject's response and the scale use judgment basis information of the MCP tool module;

[0011] If yes, based on the big model module, generate multiple easy questions according to the psychological measurement scale and the question basis information of the MCP tool module;

[0012] Send a single easy question to the tested end module in sequence and obtain the second tested end response. If the large model module determines that the user is in good condition based on the second tested end response and the user status judgment criteria of the MCP tool module, continue to send the next easy question;

[0013] The mental health assessment result is determined based on the second response obtained from the tested module after each easy-to-answer question is given to the tested module.

[0014] Optionally, when the large model module determines that the psychological measurement scale is not used, a first small talk question is generated based on the large model module and sent to the tested end module.

[0015] Optionally, when the large model module determines that the user status is not good, a second chat question is generated based on the large model module and sent to the tested end module.

[0016] Optionally, after determining the mental health assessment result, the mental health assessment result is sent to the MCP tool module for storage.

[0017] Optionally, the user basic information includes at least: a portrait of the first user's historical conversation records, the first current conversation topic, and the first user's key physiological indicators.

[0018] Optionally, the scale usage judgment basis information includes at least: a historical conversation record portrait of the second user, a second current conversation topic, key physiological indicators of the second user, a current conversation context of the first user, and a currently available MCP tool.

[0019] Optionally, the question-asking basis information includes at least: a historical conversation record portrait of the third user, a third current conversation topic, key physiological indicators of the third user, and a current conversation context of the second user.

[0020] Optionally, the user status determination basis includes at least: the reply tone of the second tested reply and the current conversation context of the third user.

[0021] An embodiment of the present invention provides a method for implementing a mental health assessment system based on a large model context protocol, comprising:

[0022] Start the large model module, MCP tool module, large model context protocol module, Agent module and the tested end module;

[0023] The Agent module uses the big model module and the MCP tool module through the big model context protocol module to interact with the tested end module for mental health assessment.

[0024] The present invention has achieved the following beneficial effects:

[0025] By deeply integrating a mental health assessment scale with a large language model agent and employing a dynamic, personalized assessment approach based on the MCP protocol, this approach overcomes the limitations of traditional assessment models. By naturally embedding scale questions into the conversational flow, the system not only maintains the reliability and validity of standardized scales but also dynamically adjusts assessment content based on the subject's real-time physiological indicators and conversational context, enabling personalized and highly interactive mental state analysis. More importantly, the MCP protocol enables the large language model to produce interpretable, clinically accurate quantitative results during mental health assessments, improving the accuracy and credibility of assessments and providing more reliable support for clinical decision-making.

[0026] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0027] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0029] Figure 1 Schematic diagram of a mental health assessment system based on a large model context protocol in an embodiment of the present invention;

[0030] Figure 2 This is a flowchart of specific implementation steps of a mental health assessment system based on a large model context protocol in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0032] Example 1:

[0033] The embodiment of the present invention provides a mental health assessment system based on a large model context protocol, such as Figure 1 As shown, including:

[0034] Large model module, MCP tool module, large model context protocol module, Agent module and tested end module;

[0035] Among them, the Agent module uses the big model module and the MCP tool module through the big model context protocol module to interact with the tested end module for mental health assessment;

[0036] The Agent module performs the following steps:

[0037] Based on the large model module, according to the user basic information of the MCP tool module, an opening statement is generated and sent to the tested end module, and the first tested end response is obtained;

[0038] Based on the large model module, determine whether to use the psychometric scale according to the first test subject's response and the scale use judgment basis information of the MCP tool module;

[0039] If yes, based on the big model module, generate multiple easy questions according to the psychological measurement scale and the question basis information of the MCP tool module;

[0040] Send a single easy question to the tested end module in sequence and obtain the second tested end response. If the large model module determines that the user is in good condition based on the second tested end response and the user status judgment criteria of the MCP tool module, continue to send the next easy question;

[0041] Determine the mental health assessment result based on the second response obtained after each easy question is given to the module;

[0042] The user basic information includes at least: a portrait of the first user's historical conversation record, the first current conversation topic, and the first user's key physiological indicators.

[0043] The scale usage judgment basis information includes at least: a portrait of the second user's historical conversation record, a second current conversation topic, key physiological indicators of the second user, a current conversation context of the first user, and currently available MCP tools.

[0044] The information on which the question is based includes at least: a historical conversation record portrait of the third user, a third current conversation topic, key physiological indicators of the third user, and a current conversation context of the second user.

[0045] The user status determination basis includes at least: the reply tone of the second tested reply and the current conversation context of the third user.

[0046] The working principle and beneficial effects of the above technical solution are:

[0047] like Figure 2 As shown, the specific implementation steps of the system of the present invention are as follows:

[0048] Step 1. Configure candidate scale tools: Based on the test subject, staff will select 10-100 scales from a large library of psychological measurement scales for the agent to use. They will also configure the topic for the chat between the agent and the test subject. Each candidate psychological measurement scale can be encapsulated as an MCP tool for the agent to select. Configure the LLM model behind the agent (such as Claude 3.7 or later) and configure the MCP tool interface to access the user's historical profile, real-time physiological indicators (such as heart rate and galvanic skin response obtained from wearable devices), and the scale response cache. Go to Step 2.

[0049] Step 2, Agent Opening: When the subject opens the Agent conversational bot, the Agent obtains the user ID and, through the MCP protocol, calls a tool to retrieve the subject's historical conversation memory, the current subject's key real-time physiological indicators, and the staff-configured assessment topic. These are then combined into the opening prompt. The large model generates the opening content and presents it to the subject in the interaction dialog box. The brackets in the prompt below represent variables, and the corresponding information is retrieved through the MCP tool each time. Go to Step 3.

[0050] The opening prompt is as follows:

[0051] """Opening Prompt

[0052] You are a friendly and empathetic mental health assistant. Your task is to help users assess their mental health through natural conversation. Please introduce relevant topics naturally based on the user's chat history and current physiological indicators.

[0053] User historical conversation record portrait: [Historical conversation memory portrait]

[0054] Current conversation topic: [Review topic]

[0055] Key physiological indicators of users: [Physiological indicators]

[0056] Consider the user's conversation history, current topic, and key physiological indicators above to create a friendly, caring, and engaging opening statement that encourages the user to share their thoughts and feelings. Avoid asking for sensitive information directly and instead create a comfortable conversational atmosphere.

[0057] Example opening styles include, but are not limited to:

[0058] - Friendly greetings

[0059] -Concerned Inquiries

[0060] -Easy topic introduction

[0061] - Leading questions

[0062] - Positive affirmations

[0063] Make sure the generated opening line flows naturally and is consistent with the user's previous conversation and current input.

[0064] """

[0065] Step 3: The subject responds: After hearing the Agent's opening remarks, the subject responds to the Agent and jumps to step 4.

[0066] Step 4: Determine whether to enable a scale: Combine the user's historical conversation memory profile, current conversation topic, user's real-time key physiological indicators, user's current conversation context, and currently available MCP tools (psychological measurement scales) into a prompt to determine whether to use the scale, and send it to the LLM model to determine whether to use the scale evaluation.

[0067] Whether to use the scale prompt is as follows:

[0068] """Whether to use the gauge prompt

[0069] You are a mental health assistant and need to determine whether to select an MCP tool (psychological measurement scale) to obtain the user's mental health information based on the user's historical conversation memory portrait, the current conversation topic, the user's real-time key physiological indicators, the user's current conversation context, and the currently available MCP tools.

[0070] User historical conversation record portrait: [Historical conversation memory portrait]

[0071] Current conversation topic: [Review topic]

[0072] Key physiological indicators of users: [Physiological indicators]

[0073] User's current conversation context: [User's current conversation context]

[0074] Currently available MCP instruments: [Optional psychometric scales]

[0075] Please judge whether to select the appropriate MCP tool based on user input and history.

[0076] """

[0077] If the LLM model inference determines that the scale measurement is not enabled temporarily, jump to step 5. If the inference determines that scale A is enabled for measurement, jump to step 6.

[0078] Step 5: Normal reply to the test subject: Based on the user's reply, memory portrait, conversation topic, etc., answer directly, and jump to step 3 after finishing.

[0079] The normal reply prompt is as follows:

[0080] """

[0081] You are a friendly and empathetic chatbot. Your task is to establish a relaxed and engaging relationship with the user through natural conversation and introduce the topic of mental health assessment at the appropriate time. Based on the user's input and history, generate a friendly and engaging response.

[0082] User historical conversation record portrait: [Historical conversation memory portrait]

[0083] Current conversation topic: [Review topic]

[0084] Key physiological indicators of users: [Physiological indicators]

[0085] User's current conversation context: [User's current conversation context]

[0086] Keep the conversation light and natural, and avoid asking for sensitive information directly. Use user responses to determine when to introduce relevant MCP tools to provide personalized support and advice.

[0087] Example small talk topics include:

[0088] - Recent hobbies and interests

[0089] -Favorite movie or book

[0090] -Travel experience

[0091] Make sure your responses are empathetic and naturally lead users to share their thoughts and feelings.

[0092] """

[0093] Step 6. Rewrite the scale questions: Use the rewrite prompt to combine user information and rewrite professional questions that are not asked in the psychological scale into a relaxed way to ask the test subjects, and jump to step 7.

[0094] Rewrite Prompt as follows:

[0095] """

[0096] You are a friendly and empathetic chatbot. Your task is to combine user information, such as historical conversation profiles, current conversation topics, key physiological indicators, and the context of the current conversation, to rewrite standardized questions from professional psychological scales into a casual, conversational format to collect information about the user's mental health in natural conversation.

[0097] User historical conversation record portrait: [Historical conversation memory portrait]

[0098] Current conversation topic: [Review topic]

[0099] Key physiological indicators of users: [Physiological indicators]

[0100] User's current conversation context: [User's current conversation context]

[0101] Example question: “Feeling tired or having little energy?” from the PHQ-9

[0102] Rephrase these questions as small talk to introduce relevant topics naturally into the conversation.

[0103] The goal is to keep the conversation light and natural, without making the user feel interrogated or evaluated. Use the user's responses to gauge their well-being and offer support and advice when appropriate.

[0104] Example rewrite:

[0105] -"Have you felt particularly tired or low on energy lately?"

[0106] -"How's your energy level lately? Are you feeling particularly tired?"

[0107] Make sure your rephrase is empathetic and naturally leads users to share their feelings.

[0108] Current Issue: [Current Issue]

[0109] Rewording the question:

[0110] """

[0111] Step 7: Classify the tested responses: Classify the tested responses according to the scale question options, and save the classification results to the user scale response cache through the MCP protocol, and jump to step 8.

[0112] """

[0113] You are a friendly and empathetic chatbot. Your task is to categorize users' responses onto objective multiple-choice options on a standard scale in order to gather information about their mental health during conversations.

[0114] The goal is to ensure accurate categorization and to provide support and advice when appropriate.

[0115] ###

[0116] Scale question: [Scale question]

[0117] Scale question options: [Scale question options]

[0118] Rewrite the scale question: [Rewrite the scale question]

[0119] Tested reply: [Tested reply]

[0120] Option classification:

[0121] """

[0122] Step 8: Are all the questions in the scale collected? Determine whether all the questions in the current scale have been asked and collected. If all have been completed, jump to step 12. If there are still unfinished questions, jump to step 9.

[0123] Step 9, determine whether to continue asking questions: Determine whether the current user can continue asking questions. If yes, jump to step 6. If the tested state is not good and cannot continue asking questions, jump to step 10.

[0124] Determine whether to continue asking prompts as follows:

[0125] """

[0126] You are a friendly and empathetic chatbot. Your task is to judge whether to continue asking questions or temporarily relax the atmosphere and ease the user's tension through small talk based on the user's response and current status.

[0127] Scale question: [Scale question]

[0128] Scale question options: [Scale question options]

[0129] Rewrite the scale question: [Rewrite the scale question]

[0130] Tested reply: [Tested reply]

[0131] Please consider the user's response content, tone, and history to determine whether it is appropriate to continue asking the question. Consider the following factors:

[0132] -The user's emotional state (e.g., sadness, anxiety, etc.)

[0133] - The tone of the user's response (e.g., hesitation, uncertainty, etc.)

[0134] -User's historical conversation records (such as previous answers or emotional fluctuations)

[0135] If the user's status indicates they need to relax, try to ease the tension with some small talk. Example small talk topics include:

[0136] - Recent hobbies and interests

[0137] -Favorite movie or book

[0138] -Travel experience

[0139] The goal is to ensure that the user feels comfortable and relaxed before continuing to ask questions. If the user can continue to answer, output yes; if small talk is needed, directly generate a new small talk question

[0140] """

[0141] Step 10: Chat to ease the atmosphere: Send the chat questions generated in step 9 to the user being tested to ease the atmosphere.

[0142] Step 11: Determine if the subject responds to the chat: If the subject responds to the question, proceed to step 9 to test whether the question can be asked again. Repeat the loop until step 9 and then jump back to step 6.

[0143] Step 12: Summarize and score the responses to the scale questions: Now that all the scale answers have been collected, calculate the measured scores and qualitative conclusions according to the standard method of the scale and jump to step 13.

[0144] Step 13: Update the user portrait library. Update the qualitative conclusions to the user portrait library through the MCP protocol, which will be used as a reference for the big model when the next evaluation is initiated.

[0145] This invention breaks through the limitations of traditional assessment models by deeply integrating the mental health assessment scale with the large language model agent and adopting a dynamic and personalized assessment method based on the MCP protocol. By naturally embedding scale questions into the conversation process, the system not only maintains the reliability and validity of the standardized scale, but also dynamically adjusts the assessment content based on the subject's real-time physiological indicators and conversation context, achieving personalized and highly interactive psychological state analysis. More importantly, the use of the MCP protocol can enable the large language model to produce interpretable, clinically standardized quantitative results during the mental health assessment process, improve the accuracy and credibility of the assessment, and provide more reliable support for clinical decision-making.

[0146] Example 2:

[0147] In an embodiment of the present invention, when the large model module determines that the psychological measurement scale is not used, a first small talk question is generated based on the large model module and sent to the tested end module.

[0148] Example 3:

[0149] In an embodiment of the present invention, when the large model module determines that the user status is not good, a second small talk question is generated based on the large model module and sent to the tested end module.

[0150] Example 4:

[0151] In an embodiment of the present invention, after the psychological health assessment result is determined, the psychological health assessment result is sent to the MCP tool module for storage.

[0152] Example 5:

[0153] In the embodiment of the present invention, the user being tested is a young employee named Xiao Ming, for example. He has recently felt great pressure at work and is feeling depressed.

[0154] Step 1 (Configure the Alternative Scaling Tool):

[0155] The topic of the conversation is configured as: "Assessment of recent emotions and stress conditions".

[0156] Alternative scale configurations are:

[0157] 1. PHQ-9 Depression Screening Scale (Patient Health Questionnaire-9): Quickly screen the severity of depressive symptoms.

[0158] 2.GAD-7 Generalized Anxiety Disorder Scale (Generalized Anxiety Disorder 7-item): Quickly screen the severity of generalized anxiety symptoms.

[0159] 3. Beck Depression Inventory (BDI-II): Assess the severity of depressive symptoms and is more detailed than PHQ-9.

[0160] 4. State-Trait Anxiety Inventory (STAI): Distinguishes and assesses current anxiety state and long-term anxiety tendency.

[0161] 5. Perceived Stress Scale (PSS): Assess the degree of life stress that an individual subjectively feels.

[0162] 6. Pittsburgh Sleep Quality Index (PSQI): Assess sleep quality over the past month.

[0163] 7. UCLA Loneliness Scale: Assess the subjective degree of social isolation and loneliness.

[0164] 8. Rosenberg Self-Esteem Scale (RSES): Assess an individual's overall self-esteem level.

[0165] 9. Social Phobia Inventory (SPIN): Screens and assesses symptoms of social anxiety disorder.

[0166] 10. Obsessive-Compulsive Inventory-Revised (OCI-R): Assess obsessive thoughts and compulsive behaviors.

[0167] 11. Positive and Negative Affect Schedule (PANAS): Assess recent positive and negative emotional experiences.

[0168] 12. Connor-Davidson Resilience Scale (CD-RISC): Assess an individual's psychological resilience in the face of stress.

[0169] Step 2 (Agent opening remarks): Xiao Ming opens the Agent dialogue robot. The Agent calls the tool through the MCP protocol:

[0170] 1. Obtain Xiao Ming’s historical conversation memory portrait (assuming that Xiao Ming mentioned “heavy workload” in the last conversation).

[0171] 2. Obtain Xiao Ming’s current real-time physiological indicators (heart rate slightly higher than normal resting level).

[0172] 3. Obtain the assessment topic "Recent Emotional and Stress Status Assessment" assigned by the staff.

[0173] 4. Integrate this information into the "Opening Prompt" and send it to the large language model.

[0174] 5. LLM generates an opening line: "Hello, Xiaoming, welcome back. I noticed you mentioned last time that you've been busy at work lately. How are you feeling? Your heart rate seems a little fast. Is there something bothering you, or were you busy with something else?"

[0175] 6.Agent presents this opening statement to Xiao Ming.

[0176] Step 3 (Tested Response): After seeing the opening line, Xiao Ming responded: "Oh, forget it. I've been feeling really tired lately. I don't have the energy to do anything, and I can't sleep well at night."

[0177] Step 4 (Determining Whether to Use a Scale): The agent integrates Xiao Ming's response, "I'm extremely tired. I have no energy for anything, and I don't sleep well at night," along with his historical profile, physiological indicators, conversation topics, and available MCP tools (PHQ-9, GAD-7), into a "Prompt to Use Scale" and sends it to the LLM. After analysis, the LLM determines that keywords like "exhausted" and "tired" are related to depressive symptoms and consistent with the assessment topic, and decides to use the "PHQ-9 Depression Screening Scale," an MCP tool.

[0178] Step 6 (Rewriting the Question): The system identifies the first question on the PHQ-9 scale as "In the past two weeks, how often have you felt troubled by a lack of interest or enjoyment in doing things?" The agent integrates this question and Xiao Ming's context (historical profile, physiological indicators, current conversation, etc.) into a "Rewriting the Question Prompt" and sends it to the LLM. The LLM rewrites the question into a more natural conversational format: "I see. That does sound like a lot of work. So, are you feeling less interested in doing things you might normally enjoy, like listening to music, exercising, or chatting with friends?" The agent then sends the rewritten question to Xiao Ming.

[0179] Step 7 (Categorizing Participant Responses): Xiao Ming responded, "Yeah, I used to enjoy playing games after get off work to relax, but now I just want to lie down and do nothing." The agent combined Xiao Ming's response, the original scale question, the options (e.g., "Not at all," "Several days," "More than half the time," "Almost every day"), and the rewritten question into the "Participant Response Categorization Prompt" and sent it to the LLM. After analysis, the LLM categorized Xiao Ming's response, "Yes...I don't want to do anything," as the PHQ-9 option for that question, "Almost every day." Using the MCP protocol, the tool was called to store the result, "PHQ-9 Question 1: Almost every day," in Xiao Ming's scale response cache.

[0180] Step 8 (Are all the questions collected?): The system checks whether all the questions in the PHQ-9 scale have been asked and the answers collected. It is found that there are still some questions that have not been completed.

[0181] Step 9 (Determine whether to continue asking questions): The agent integrates Xiao Ming's response, current status, and other information into the "Determine whether to continue asking questions prompt" and sends it to the LLM. The LLM analyzes Xiao Ming's response, which is calm in tone and relevant to the assessment, and determines that it is appropriate to continue asking questions, outputting "yes."

[0182] Step 6 (Continue Questioning): The system retrieves the next unasked question from the PHQ-9, such as "In the past two weeks, how often have you been troubled by 'feeling down, depressed, or hopeless'?" and rephrase the question again using the LLM: "I see. Besides a lack of interest, have you also often felt a little down or hopeless recently?" The agent sends the rephrased question.

[0183] Step 7 (reclassification): Xiao Ming replied: "Sometimes, especially at night when I'm alone, I'll have random thoughts and feel so stressed." LLM classified it as "more than half the time" and stored it in the cache through the MCP protocol.

[0184] Example of loops and interruptions: Suppose that when asked a question (for example, about thoughts of self-harm), Xiao Ming's response shows obvious avoidance, anxiety, or resistance ("I don't want to talk about this, can we change the subject?"), and real-time physiological indicators (such as heart rate and galvanic skin response) show significant fluctuations.

[0185] Step 9 (Determine whether to continue asking questions): LLM analyzes Xiao Ming's response and physiological indicators, determines that it is not appropriate to continue asking in-depth questions at this time, decides to ease the atmosphere, and generates a chat question.

[0186] Step 10 (small talk to ease the atmosphere): The agent sends the small talk question generated by the LLM: "Okay, okay, no problem. Let's talk about something relaxing first. The weather has been nice recently. Did you go out for a walk and relax this weekend?"

[0187] Step 11 (Assessing Participant's Response to Small Talk): Xiao Ming responds to the small talk questions. The agent continues a few rounds of casual conversation. Then, the LLM returns to Step 9, where it reassesses Xiao Ming's status. If Xiao Ming's status has calmed down, it attempts to return to Step 6 and continue with the previous scale questions. If Xiao Ming's status remains unsuitable, it continues the small talk or ends the assessment.

[0188] Step 12 (Score Summary of Responses): Assume that after several rounds of conversation (possibly including interruptions and resumptions), the agent successfully collects answers to all questions on the PHQ-9 scale. The system retrieves all categorized results from the cache using the MCP protocol and calculates a total score (e.g., 12) according to the standard PHQ-9 scoring rules. Based on this score, the system draws a qualitative conclusion: "Moderate Depressive Tendency."

[0189] Step 13 (Update User Profile Library): The Agent invokes the User Profile Library Update Tool via the MCP protocol and updates Xiao Ming's historical conversation memory profile library with the assessment conclusion (PHQ-9 score 12, moderate depressive tendency) and other information, including the assessment date, for future assessments or interventions. A structured assessment report summary is also generated for staff reference.

[0190] This embodiment of the present invention demonstrates how the system uses the MCP protocol to flexibly embed the standardized PHQ-9 scale into a natural conversation with user Xiao Ming. Through the understanding, judgment, and rewriting capabilities of the LLM, the following are achieved:

[0191] Contextual opening: Initiate conversations by combining user history and real-time data.

[0192] Dynamic tool selection: Introduce appropriate assessment scales (MCP tools) at the right time based on the content and topic of the conversation.

[0193] Naturalized questioning: Rewrite the objective selection questions of the scale into a colloquial, non-interrogative dialogue.

[0194] Intelligent categorization: Understands user natural language responses and maps them to scale options.

[0195] Adaptive interaction: Adjust the conversation rhythm based on the user's status and insert small talk at the right time to relieve tension.

[0196] Structured output: After completing the evaluation, calculate the score and give qualitative conclusions, and update the user profile.

[0197] Overall, it overcomes the rigidity of traditional questionnaires and the limitations of direct application of LLM, achieving a more humane, personalized and professional standard mental health assessment.

[0198] Example 6:

[0199] Directly disseminating mental health assessment results after they are determined can cause emotional fluctuations in users, especially if the results don't meet their expectations or reveal psychological challenges they face. This emotional fluctuation can affect their mental state and, in some cases, even lead to adverse consequences such as anxiety, depression, or emotional distress. Therefore, the communication of assessment results must be handled with caution to avoid causing excessive emotional impact on users.

[0200] To this end, in an embodiment of the present invention, the mental health assessment system based on the large model context protocol further includes:

[0201] Acceptance Assessment and Acceptance Support Modules, including:

[0202] After the Agent module determines the mental health assessment results, the user of the tested module is evaluated to see how well they can directly accept each result item in the mental health assessment results; the acceptance is calculated as the weighted sum of the severity of the result item and the average acceptance of the result item by other users similar to the user;

[0203] Outputting result items whose acceptance degree exceeds the first threshold to the user, and supporting the user in accepting the result items whose acceptance degree does not exceed the first threshold.

[0204] The mental health assessment results are divided into multiple items. The system evaluates the user's acceptance of each item. If the acceptance exceeds a first threshold (a preset threshold representing a high level of acceptance), the corresponding item is directly output to the user. If the output result does not exceed the first threshold, the system supports the user in accepting it and avoids directly outputting the result. This effectively avoids emotional fluctuations in the user when outputting the mental health assessment results, helps users face their mental health status with a positive attitude, and promotes future improvement.

[0205] When evaluating acceptance, a weighted calculation is performed on the severity of the result item (the inverse of the severity of the mental health status involved in the result item) and the average acceptance of the result item by other users similar to the user (similarity refers to other users who are similar to the user in age, gender, occupation, and life experience) (the average of the acceptance of the corresponding result item by these other users) (the weights of the severity of the result item and the average acceptance of the result item by other users similar to the user are set in advance according to their respective representations of the user's ability to directly accept the result item, and the sum of the two weights is 1. The weights are multiplied by their corresponding values, and the sum of the two multiplication results is the weighted calculation sum). This improves the accuracy, comprehensiveness and applicability of the assessment of the acceptance of each result item in the mental health assessment results that users can directly accept.

[0206] Supporting the user in accepting the result item whose acceptance level does not exceed the first threshold includes:

[0207] Traverse the result items whose acceptance does not exceed the first threshold in order from large to small acceptance;

[0208] During each traversal, multiple supporting contents and their respective support levels are determined from the user's personal experience to support the user's acceptance of the traversed result item; the support level is calculated as the weighted sum of the total number of times the user has experienced the supporting contents and the freshness of their memory;

[0209] sorting the supporting contents whose support exceeds the second threshold in ascending order of support to obtain a supporting content sequence;

[0210] Identify the relay capability relationship between any supporting content in the supporting content sequence and any subsequent supporting content; the relay capability relationship means that the negative emotions that may be generated by the user when the former supporting content is output can be resolved by the latter supporting content output;

[0211] Eliminate support content that does not have a relay capability relationship from the support content sequence;

[0212] Based on the relay capability relationships of the supporting content in the eliminated supporting content sequence, relay triggering timing and relay control rules are set for the corresponding supporting content; the relay triggering timing refers to the negative emotions that may arise when the user outputs the supporting content; the relay control rules refer to the output of other supporting content that has a relay capability relationship with the supporting content;

[0213] Outputting the supporting contents in the support content sequence after elimination to the user in sequence order;

[0214] Each time an output is made, it is identified whether the user has triggered the corresponding relay triggering opportunity. If so, the corresponding relay control rule is executed.

[0215] The user's personal experience includes at least: life experience, living habits, mental health history, historical mental health assessment results, professional experience and skill background. From the user's personal experience, multiple supporting contents and their respective support levels that can support the user to accept the traversed result items are determined. The support level represents the degree of support that the supporting content can support the user to accept the traversed result items. Specifically, for example: the traversed result item is mild depression, and the supporting content is a positive lifestyle in living habits, etc. When determining the support level, the total number of times the user has experienced the supporting content (the total number of times the supporting content has been experienced in history) and the memory freshness (the inverse of the time difference from the last time the supporting content was experienced) are weighted and summed (different weights are pre-set based on the ability of each of the two to represent the degree of support for the user to accept the traversed result items). The more total experiences and the higher the memory freshness, the clearer the subjective memory related to the user when viewing the supporting content, and the more it can help the user understand the corresponding result items.

[0216] The second threshold represents a higher support level. Support contents with support levels exceeding the second threshold are sorted from small to large according to support levels to obtain a support content sequence. When outputting support contents in sequence, support contents with lower support levels but sufficient support levels are output first, so that support contents with higher support levels can serve as a backup for their support effects.

[0217] Identify the relay capability relationship between any supporting content in the supporting content sequence and any supporting content after it, and eliminate supporting content that does not have a relay capability relationship from the supporting content sequence. The relay capability relationship means that the negative emotions that may be generated by the user when the former supporting content is output can be resolved through the output of the latter supporting content. For example: the former supporting content is a positive lifestyle in life habits. When it is output, the user may have negative emotions that they never want to adopt this lifestyle again. Then the latter supporting content is the relatives and friends who have often accompanied the user to adopt this lifestyle in history and their companionship history, which can help the user resolve the negative emotions.

[0218] Support content that does not have a relay capability relationship is removed from the support content sequence, so that each support content in the support content sequence after the removal is followed by other support content to further back it up for the support effect.

[0219] The support contents in the support content sequence after elimination are output to the user in sequence order. Each time the support contents are output, it is identified whether the user triggers the corresponding relay triggering opportunity. If so, the corresponding relay control rule is executed.

[0220] This embodiment of the present invention first traverses the result items whose acceptance level does not exceed a first threshold, and then, based on the user's historical experience, determines which support content can help the user better accept these result items. The support level is sorted from smallest to largest, ensuring that the user can gradually receive sufficient help when accepting support.

[0221] In particular, this process also considers the relay capacity relationship between supporting content. This relationship refers to whether, when a certain supporting content triggers negative emotions in the user, subsequent supporting content can help alleviate these emotions, ensuring that the user does not fall into a cycle of negative emotions throughout the support process. By eliminating supporting content that does not have a relay capacity relationship, an effective supporting content sequence is ultimately formed. This sequence is output sequentially. Whenever a user triggers negative emotions, the system will automatically use relevant supporting content to alleviate and support the user's emotions, enhancing the overall effect.

[0222] Overall, users can receive effective help through multi-level and systematic support, and gradually accept and cope with the results of mental health assessments.

[0223] Example 7:

[0224] An embodiment of the present invention provides a method for implementing a mental health assessment system based on a large model context protocol, characterized by comprising:

[0225] Start the large model module, MCP tool module, large model context protocol module, Agent module and the tested end module;

[0226] The Agent module uses the big model module and the MCP tool module through the big model context protocol module to interact with the tested end module for mental health assessment.

[0227] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A mental health assessment system based on a large model context protocol, characterized in that: include: Large model module, MCP tool module, large model context protocol module, Agent module and tested end module; Among them, the Agent module uses the big model module and the MCP tool module through the big model context protocol module to interact with the tested end module for mental health assessment.

2. The mental health assessment system based on the large model context protocol according to claim 1, characterized in that: The Agent module performs the following steps: Based on the large model module, according to the user basic information of the MCP tool module, an opening statement is generated and sent to the tested end module, and the first tested end response is obtained; Based on the large model module, determine whether to use the psychometric scale according to the first test subject's response and the scale use judgment basis information of the MCP tool module; If yes, based on the big model module, generate multiple easy questions according to the psychological measurement scale and the question basis information of the MCP tool module; Send a single easy question to the tested end module in sequence and obtain the second tested end response. If the large model module determines that the user is in good condition based on the second tested end response and the user status judgment criteria of the MCP tool module, continue to send the next easy question; The mental health assessment result is determined based on the second response obtained from the tested module after each easy-to-answer question is given to the tested module.

3. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: When the large model module determines that the psychological measurement scale is not used, a first small talk question is generated based on the large model module and sent to the tested end module.

4. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: When the large model module determines that the user status is not good, a second chat question is generated based on the large model module and sent to the tested end module.

5. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: After determining the mental health assessment results, the mental health assessment results are sent to the MCP tool module for storage.

6. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: The user basic information includes at least: a portrait of the first user's historical conversation record, the first current conversation topic, and the first user's key physiological indicators.

7. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: The scale usage judgment basis information includes at least: a historical conversation record portrait of the second user, a second current conversation topic, key physiological indicators of the second user, a current conversation context of the first user, and a currently available MCP tool.

8. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: The question basis information includes at least: a historical conversation record portrait of the third user, a third current conversation topic, key physiological indicators of the third user, and a current conversation context of the second user.

9. The mental health assessment system based on the large model context protocol according to claim 2, characterized in that: The user status determination basis includes at least: the reply tone of the second tested reply and the current conversation context of the third user.

10. A method for implementing a mental health assessment system based on a large model context protocol, characterized in that: include: Start the large model module, MCP tool module, large model context protocol module, Agent module and the tested end module; The Agent module uses the big model module and the MCP tool module through the big model context protocol module to interact with the tested end module for mental health assessment.

Images