An early warning method, system and device for negative emotions, thoughts and self-injurious behaviors of adolescents based on artificial intelligence

Abstract

The present application relates to the technical field of computer intelligent identification, in particular to a warning method, system and device for negative emotions, thinking and self-injurious behavior of teenagers based on artificial intelligence, comprising: obtaining interactive voice and first interactive text of a target teenager, converting the interactive voice into second interactive text through a voice recognition model, merging the first interactive text and the second interactive text to obtain communication text; performing text preprocessing on the communication text to obtain standard communication text; and inputting the standard communication text into a pre-trained second large language model. The present application maps the text labels in the labeled communication text set to abstract labels by constructing an abstract label table and a corresponding mapping relationship table, which helps to extract semantic information of the text, so that the meaning and theme of the communication text can be better understood.

Description

Technical Field

[0001] This invention relates to the field of computer intelligent recognition technology, specifically to an early warning method, system, and device for negative emotions, thoughts, and self-harm behaviors in adolescents based on artificial intelligence. Background Technology

[0002] Adolescent mood disorders and suicidal / self-harm behaviors are serious social problems that affect the safety and development of adolescents themselves, as well as the stability of families, schools, and society. However, adolescent mood disorders and self-harm behaviors are characterized by low screening rates and low rates of seeking medical treatment.

[0003] Several challenges exist in the large-scale screening of adolescent emotional disorders and suicidal behavior in the early stages of the work: there are not enough professional doctors and psychologists to engage in one-on-one psychological interviews, and the consistency of the interview results assessment is poor. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the background technology by proposing an artificial intelligence-based early warning method, system, and device for negative emotions, thoughts, and self-harm behaviors in adolescents.

[0005] The technical solution of this invention: A method for early warning of negative emotions, thoughts, and self-harm behaviors in adolescents based on artificial intelligence, comprising:

[0006] The first language model is trained to obtain the interactive voice and first interactive text of the target teenagers. The first language model is used to guide the target teenagers to talk about topics related to emotions.

[0007] The interactive speech is converted into second interactive text using a speech recognition model. The first interactive text and the second interactive text are then merged to obtain the communication text.

[0008] The exchange text is preprocessed to obtain a standard exchange text.

[0009] The standard communication text is input into a pre-trained second language model, and the trained second language model outputs a third interactive text labeled with symptom vectors, and each symptom vector includes three feature channels;

[0010] The symptom vectors are imported into a pre-trained convolutional neural network model, and the analysis results of the target adolescent's emotional disorders are output through the pre-trained convolutional neural network model.

[0011] The standard communication text and the analysis results of the emotional disorder are input into the pre-trained third language model, and the pre-trained third language model outputs graded warning results and suggestions.

[0012] The analysis results, classification, early warning, and intervention recommendations for the aforementioned mood disorders will be presented.

[0013] Preferably, the communication text is preprocessed to obtain standard communication text, including:

[0014] Define a set of communication texts, wherein the set of communication texts includes a plurality of the communication texts;

[0015] Text tags are added to each of the communication texts in the communication text set to obtain a tagged communication text set. The tagged communication text set includes multiple tagged communication texts, and each tagged communication text includes the communication text and its corresponding tag.

[0016] Construct an abstract tag table, which includes multiple abstract tags, and establish a mapping relationship table between the abstract tag table and the tags;

[0017] The tagged communication text set is mapped and matched with the abstract tag table according to the mapping relationship table, so as to obtain the abstract tag corresponding to each tagged communication text in the tagged communication text set;

[0018] The abstract tag is matched with a preset tag content library to obtain the tag content corresponding to the abstract tag. The tag content is then added to the corresponding tag communication text to obtain the standard communication text.

[0019] Preferably, text tags are added to each of the communication texts in the communication text set to obtain a tagged communication text set, including:

[0020] Extract the text topic features of each of the communication texts in the communication text set;

[0021] The similarity model is used to calculate the similarity between the text topic features and each tag in a pre-set tag table. The tag with the highest similarity is then used as the tag for the communication text. The similarity model is as follows:

[0022]

[0023] Where M represents the similarity metric, Let θ0 represent the vector residual value of the semantic information of the exchanged text fragment, θ0 represent the initial value of the vector residual of the semantic information of the exchanged text fragment, and t represent the initial value of the similarity index. This represents the general assignment conditions for substituted parameters, b1, b2, ..., b nA classification index representing the semantic information of n distinct similar text fragments;

[0024] The markers of each text in the set of text exchanges are counted, and the markers are added to the text exchanges to obtain a set of marked text exchanges.

[0025] Preferably, the first large language model is used to guide the target adolescents to talk about emotion-related topics, including:

[0026] A first knowledge base is constructed, and a first reminder is constructed based on the first knowledge base. The first reminder is used to debug the first large language model, so that the first large language model can be used to guide the target adolescents to talk about topics related to emotions.

[0027] Preferably, inputting the standard communication text into the pre-trained second language model further includes:

[0028] A second knowledge base is constructed, and a second reminder is built based on the second knowledge base. The second reminder is used to debug the second large language model, thereby obtaining a trained second large language model.

[0029] Preferably, the standard communication text and the analysis results of the emotional disorder are input into a pre-trained third language model, and the pre-trained third language model outputs graded warning results and suggestions, further including:

[0030] A third knowledge base is constructed, and a third reminder is constructed based on the third knowledge base. The third reminder is used to debug the third language model, thereby obtaining a trained third language model.

[0031] Preferably, the tiered early warning includes Level 1, Level 2, Level 3, and Level 4 early warnings, and the recommendations include Level A, Level B, Level C, Level D, and Level E handling.

[0032] The technical solution of this invention: an early warning system for negative emotions, thoughts, and self-harm behaviors in adolescents based on artificial intelligence, applicable to the aforementioned early warning method for negative emotions, thoughts, and self-harm behaviors in adolescents based on artificial intelligence, comprising an interactive data acquisition module, a data processing and conversion module, an emotion analysis module, an early warning and suggestion module, and a result presentation module, characterized in that:

[0033] The interactive data acquisition module is used to acquire the interactive voice and text of teenagers using the first language model, and guide them to talk about topics related to emotions.

[0034] The data processing and conversion module is used to convert speech into text and perform preprocessing to obtain standard communication text.

[0035] The emotion analysis module is used to analyze standard communication text using the second largest language model, output emotion symptom vector labels, and use a convolutional neural network model to analyze the emotion symptom vectors to determine the emotional state and degree of impairment of adolescents.

[0036] The warning and suggestion module is used to output warning levels and suggestions based on sentiment analysis using the third language model;

[0037] The results presentation module is used to present sentiment analysis results, warning levels, and recommendations.

[0038] The technical solution of the present invention: an early warning device for negative emotions, thoughts and self-harm behaviors of adolescents based on artificial intelligence, which is applicable to the aforementioned early warning method for negative emotions, thoughts and self-harm behaviors of adolescents based on artificial intelligence, comprising: at least one processor and a memory;

[0039] The memory is used to store computer-executed instructions;

[0040] The at least one processor is used to execute computer execution instructions stored in the memory to implement the aforementioned artificial intelligence-based early warning method for negative emotions, thoughts, and self-harm behaviors in adolescents.

[0041] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0042] 1. This invention constructs an abstract tag table and a corresponding mapping table to map text tags in a set of tagged communication texts onto abstract tags. This helps to extract semantic information from the text, thereby enabling a better understanding of the meaning and theme of the communication text. Furthermore, by matching the abstract tags with a preset tag content library, corresponding tag content can be added to the text based on the specific abstract tags, making the standard communication text more personalized and information-rich. This helps to provide more valuable information for subsequent sentiment analysis.

[0043] 2. This invention integrates speech recognition, large language models, and convolutional neural network models to automatically analyze and predict adolescents' emotional disorders and potential self-harm behaviors. It can also automatically collect and process data, significantly improving the timeliness of early warnings. Furthermore, by combining multi-model analysis, it improves the accuracy of emotional disorder diagnosis and can monitor in real time and provide timely early warnings and intervention suggestions, effectively preventing self-harm behaviors. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the overall method in one embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the text preprocessing process in one embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of the text tag addition process in one embodiment of the present invention;

[0047] Figure 4 This is a partial diagram of the first reminder in one embodiment of the present invention;

[0048] Figure 5 This is a partial diagram of the first knowledge base in one embodiment of the present invention;

[0049] Figure 6 This is a partial diagram of the second knowledge base in one embodiment of the present invention;

[0050] Figure 7 This is a schematic diagram illustrating the encoding of symptom severity in one embodiment of the present invention;

[0051] Figure 8 This is a schematic diagram of the encoding of emotional episodes in one embodiment of the present invention;

[0052] Figure 9 This is an excerpt of the third reminder in one embodiment of the present invention;

[0053] Figure 10 This is a schematic diagram of the warning level standard in one embodiment of the present invention;

[0054] Figure 11 This is a schematic diagram of a proposed standard in one embodiment of the present invention;

[0055] Figure 12 This is a schematic diagram of the final report template in one embodiment of the present invention;

[0056] Figure 13 This is a flowchart of the overall system in one embodiment of the present invention;

[0057] Figure 14 This is a schematic diagram of the overall equipment in one embodiment of the present invention.

[0058] Reference numerals: 1. Interactive data acquisition module; 2. Data processing and conversion module; 3. Early warning and suggestion module; 4. Early warning and suggestion module; 5. Result presentation module; 6. Processor; 7. Memory; 8. Bus. Detailed Implementation

[0059] Example 1, as Figure 1 As shown, this invention proposes an artificial intelligence-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents, comprising:

[0060] S1. Use the trained first language model to obtain the target teenagers' interactive voice and first interactive text. The first language model is used to guide the target teenagers to talk about topics related to emotions.

[0061] S2. The interactive speech is converted into the second interactive text through a speech recognition model. The first interactive text and the second interactive text are then merged to obtain the communication text.

[0062] S3. Perform text preprocessing on the communication text to obtain standard communication text;

[0063] S4. Input the standard communication text into the pre-trained second language model, and output the third interactive text with symptom vector labels through the trained second language model, and each symptom vector includes three feature channels.

[0064] S5. Input the symptom vector into the pre-trained convolutional neural network model, and output the analysis results of the target adolescent's emotional disorders through the pre-trained convolutional neural network model.

[0065] S6. Input the analysis results of standard communication text and emotional disorders into the pre-trained third language model, and output the graded warning results and suggestions through the pre-trained third language model;

[0066] S7. Present the analysis results, classification, early warning and intervention recommendations for mood disorders.

[0067] In this invention, the speech recognition model includes, but is not limited to, the Hidden Markov Model (HMM). The HMM is a statistical model that is widely used in speech recognition tasks. It aligns acoustic feature sequences with speech units such as phonemes and uses acoustic models, pronunciation dictionaries, and language models to infer the most likely text sequence.

[0068] Example 2, as Figure 2-5 As shown, the present invention proposes an artificial intelligence-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents. Compared with Embodiment 1, this embodiment further includes:

[0069] The communication text is preprocessed to obtain a standard communication text, including:

[0070] A1. Define a set of communication texts, which includes multiple communication texts;

[0071] A2. Add text tags to each text in the text exchange set to obtain a tagged text exchange set. The tagged text exchange set includes multiple tagged texts, and each tagged text exchange includes the text exchange and its corresponding tag.

[0072] A3. Construct an abstract tag table, which includes multiple abstract tags, and establish a mapping relationship table between the abstract tag table and the tags;

[0073] A4. Based on the mapping relationship table, map and match the set of marked communication texts with the abstract label table to obtain the abstract label corresponding to each marked communication text in the set of marked communication texts;

[0074] A5. Match the abstract tags with the preset tag content library to obtain the tag content corresponding to the abstract tags. Add the tag content to the corresponding tag communication text to obtain the standard communication text.

[0075] In this embodiment, the abstract tag table is used to describe the themes or emotional aspects of the communication text. These abstract tags are usually obtained through semantic analysis and theme identification of the text content, and they reflect the important features or themes of the text. The tag content in the tag content library can be predefined or dynamically generated, depending on the needs of the preprocessing task and the actual situation. For example, for the abstract tag "happiness", the tag content library can contain a series of words or phrases describing the emotion of happiness, such as "happy", "joyful", "delight", etc.; for the abstract tag "anxiety", the tag content library can contain some descriptions related to anxiety, such as "worried", "nervous", "uneasy", etc.

[0076] In an optional embodiment, text tags are added to each communication text in the communication text set to obtain a tagged communication text set, including:

[0077] B1. Extract the text theme features of each text in the collection of exchange texts;

[0078] B2. Calculate the similarity between the text's topic features and each tag in a pre-set tag table using a similarity model. The tag with the highest similarity is then used as the tag for the exchanged text. The similarity model is as follows:

[0079]

[0080] Where M represents the similarity metric, Let θ0 represent the vector residual value of the semantic information of the exchanged text fragments, θ0 represent the initial value of the vector residual of the semantic information of the exchanged text fragments, and t represent the initial value of the similarity index. This represents the general assignment conditions for substituted parameters, b1, b2, ..., b n A classification index representing the semantic information of n distinct similar text fragments;

[0081] B3. Count the tags of each text in the text exchange set and add the tags to the text exchange to obtain the tagged text exchange set.

[0082] In an optional embodiment, the first major language model is used to guide the target adolescents to talk about emotion-related topics, including:

[0083] Construct a first knowledge base, and build a first reminder based on the first knowledge base. Use the first reminder to debug the first major language model, so that the first major language model can be used to guide the target teenagers to talk about topics related to emotions.

[0084] It should be noted that the first reminder is used to debug the first major language model. For example: Role: You are a psychiatrist specializing in mental health screening. Your role is to converse with users and screen them for common mental health disorders, including suicidal or self-harming thoughts or behaviors. After the interview, you should provide a summary. However, avoid frequently mentioning topics related to symptoms during the conversation, and do not guide users to seek professional help. Role Simulation: The robot needs to communicate with users as a psychiatrist. You should be patient, compassionate, and a good listener, with a guiding questioning style; ensure the output Chinese content conforms to the daily speaking habits of Chinese people. Avoid giving advice or solutions when the user's condition is severe. Only one question should be asked at a time, with concise and succinct answers. The task focuses on screening summary and graded diagnosis, without excessive explanation. (Each output should be less than 150 characters, with no word limit for the final summary). Sensitive Word Handling: When sensitive words such as self-harm, suicide, or self-injury are encountered, the robot will systematically ask questions according to the set steps to gain a complete and detailed understanding of the adolescent's current suicide and self-harm risks. When sensitive terms such as violence, bullying, and rape are mentioned, the robot will conduct 2-3 rounds of dialogue to understand the specific circumstances related to the sensitive term (asking only one question of less than 100 words each time); then it will continue to ask questions according to the questioning order in rob.x.1.2s2 of the knowledge base file, and finally summarize according to the report template in the knowledge base file. Objective: Self-introduction and initiation of dialogue. Start the conversation with the message: "Hello! I am a psychiatrist who focuses on adolescent mental health. Please share your feelings with me, and we can work together to find solutions to help you through this difficult time. I will keep all our conversations confidential and will not disclose them to anyone else. Now, please ensure you are in a safe and private environment. Let me know when you are ready." After receiving a response from the user such as "Ready?", begin the conversation following these steps. Conversation requirements: Do not repeatedly say that you are an AI or emphasize your limited abilities, as this will encourage the user to express their true feelings; do not offer any form of advice, especially not point-by-point suggestions. Whenever the user is about to end the conversation or leave, be sure to summarize the entire conversation as in step 6 and output the corresponding symbol as in step 5. When the user types messages such as "Thank you" or "End conversation," summarize the symptoms as in step 6 to end the conversation.

[0085] In an optional embodiment, inputting standard communicative text into a pre-trained second large language model further includes:

[0086] A second knowledge base is constructed, and a second reminder is built based on the second knowledge base. The second reminder is used to debug the second language model, thereby obtaining a trained second language model.

[0087] It should be noted that the second knowledge base includes 16 symptoms such as mood, activities and experiences, excessive participation in risky activities, and appetite. Each symptom contains three feature channels: symptom presentation, severity, and duration. The first two features are coded using one-hot encoding, and the last feature is coded using standardized 0-1 numbers. For example, severe low mood lasting 15 days is represented by the encoding [1,0,0,0,0,0,0,0,0,0,0,1,0,0,1].

[0088] It should be noted that the second reminder was used to debug the second language model. The main debugging method involved 5-6 psychiatric experts (doctors or graduate students with practicing physician qualifications and at least two years of experience in psychiatry) simulating patients diagnosed with various depressive disorders, bipolar disorder, schizophrenia, etc., engaging in dialogue with the first language model. A total of 100 users' dialogue texts were generated. After the above processing, all dialogue texts were converted into Markdown files and uploaded to the second language model in the form of a knowledge base. The second language model identified symptoms based on the second reminder and converted the symptoms into text with vector tags. Psychiatric experts then judged the accuracy of symptom identification and the correctness of the encoding. If the symptoms or encoding were incorrect, the second reminder was revised until all psychiatric experts believed that the second language model could correctly identify and encode symptoms in 90% of the cases.

[0089] In an optional embodiment, the analysis results of standard communication text and emotional disorders are input into a pre-trained third language model, and the pre-trained third language model outputs graded warning results and suggestions, further including:

[0090] A third knowledge base is constructed, and a third reminder is built based on the third knowledge base. The third reminder is used to debug the third language model, thereby obtaining a trained third language model.

[0091] In one optional embodiment, the tiered early warning system includes Level 1, Level 2, Level 3, and Level 4 warnings, and it is recommended to include Level A, Level B, Level C, Level D, and Level E processing.

[0092] It should be noted that the following are the treatment guidelines: Category A: Meets the diagnostic criteria for any ICD-10 mental disorder and has a high risk of suicide or harming others. Recommendations: If there are serious mental health issues, immediately inform the guardian, provide continuous care to prevent self-harm, suicide, or harming others, and seek immediate medical attention at a psychiatric clinic. Category B: Meets the diagnostic criteria for any ICD-10 mental disorder and requires scheduled psychiatric treatment, but not immediate treatment. Recommendations: If there are relatively serious psychological problems, inform the teacher and guardian to provide close attention and recommend seeking medical attention at a psychiatric clinic. Category C: Has significant psychological problems (but does not meet the diagnostic criteria for any ICD-10 mental disorder and does not require psychiatric treatment). Recommendations: If there are significant psychological problems, parents and teachers should provide close attention and regular follow-up consultations. Category D: Has some psychological problems (but does not meet the diagnostic criteria for any ICD-10 mental disorder and does not require psychiatric treatment), and is currently in a stable state and emotional condition (currently has no suicidal or self-harming thoughts, behaviors, or plans). Recommendations are as follows: There are some psychological issues, but the patient's emotions are currently stable; routine follow-up is recommended. Level E: The patient's emotions are stable, and there are no obvious psychological issues. Recommendations are as follows: The patient's emotions are currently stable; follow-up is recommended.

[0093] It should be noted that Level 1 warning: Those who meet the following main criteria and / or meet more than two of the secondary criteria will only be classified as Management Recommendation A. Main criteria: Meet the diagnostic criteria for Major Depressive Disorder (MDD) and have a definite suicide plan within the past month (including the time, place, and method of suicide); currently meet the MDD diagnosis and have strong suicidal ideation, and have recently attempted suicide and / or frequently engaged in self-harm, or feel unable or completely unable to control negative thoughts; currently have severe and varied psychotic symptoms, such as true command hallucinations, persistent persecutory delusions, etc., posing a risk of harming others, and self-reporting intention to harm others; Secondary criteria: Uncooperative during interviews, patients do not answer multiple questions or answer only a few questions, and / or may be concealing information; currently meet the MDD diagnosis, have strong negative thoughts, and have poor family and social support; recently experienced a severe traumatic or stressful event, combined with current negative thoughts; high risk of impulsivity. Level 2 Warning: A diagnosis of at least one mental disorder is established, but the severity and urgency level is not as high as Level 1 Warning. Level 3 Warning: Some psychological problems exist (including a history of mental illness, a history of strong negative thoughts and / or suicidal / self-harming behavior), but no current diagnosis of any mental disorder is established, and there is no immediate risk of suicide. Management recommendations can be categorized as C or D. Level 4 Warning: There are essentially no significant psychological problems. Management recommendations can be categorized as D or E.

[0094] Example 3, as Figure 6As shown, this invention proposes an artificial intelligence-based early warning system for negative emotions, thoughts, and self-harming behaviors in adolescents. It is applicable to the aforementioned artificial intelligence-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents. The system includes an interactive data acquisition module 1, a data processing and conversion module 2, an emotion analysis module 3, an early warning and suggestion module 4, and a result presentation module 5. Its key features are:

[0095] Interactive data acquisition module 1 is used to acquire interactive voice and text of teenagers using the first major language model, and guide them to talk about topics related to emotions;

[0096] Data processing and conversion module 2 is used to convert speech into text and perform preprocessing to obtain standard communication text;

[0097] The emotion analysis module 3 is used to analyze standard communication text using the second largest language model, output emotion symptom vector labels, and use a convolutional neural network model to analyze the emotion symptom vectors to determine the emotional state and degree of impairment of adolescents.

[0098] The warning and suggestion module 4 is used to output warning levels and suggestions based on sentiment analysis using the third language model;

[0099] Module 5, the results presentation module, is used to present the sentiment analysis results, warning levels, and recommendations.

[0100] Example 4, as Figure 7 As shown, the present invention proposes an early warning device for negative emotions, thoughts and self-harm behaviors of adolescents based on artificial intelligence, which is applicable to the aforementioned early warning method for negative emotions, thoughts and self-harm behaviors of adolescents based on artificial intelligence, including: at least one processor 6 and a memory 7.

[0101] Memory 7 is used to store instructions executed by the computer;

[0102] At least one processor 6 is used to execute computer execution instructions stored in memory 7 to implement the aforementioned artificial intelligence-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents.

[0103] In an alternative embodiment, the memory 7 can be either standalone or integrated with the processor 6. When the memory 7 is set up independently, the device also includes a bus 8 for connecting the memory 7 and the processor 6.

[0104] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. An artificial intelligence-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents, characterized in that, include: The first language model, which has been trained, is used to obtain the interactive voice and text of the target teenagers. The first language model is used to guide the target teenagers to talk about topics related to emotions. The interactive speech is converted into second interactive text using a speech recognition model. The first interactive text and the second interactive text are then merged to obtain the communication text. The exchange text is preprocessed to obtain standard exchange text; The standard communication text is input into a pre-trained second language model, and the trained second language model outputs a third interactive text labeled with symptom vectors, and each symptom vector includes three feature channels; The symptom vectors are imported into a pre-trained convolutional neural network model, and the analysis results of the target adolescent's emotional disorders are output through the pre-trained convolutional neural network model. The standard communication text and the analysis results of the emotional disorder are input into the pre-trained third language model, and the pre-trained third language model outputs graded warning results and suggestions. The analysis results, classification, early warning, and intervention recommendations for the aforementioned mood disorders will be presented.

2. The method for early warning of negative emotions, thoughts, and self-harming behaviors in adolescents based on artificial intelligence, as described in claim 1, is characterized in that: The exchange text is preprocessed to obtain standard exchange text, including: Define a set of communication texts, wherein the set of communication texts includes a plurality of the communication texts; Text tags are added to each of the communication texts in the communication text set to obtain a tagged communication text set. The tagged communication text set includes multiple tagged communication texts, and each tagged communication text includes the communication text and its corresponding tag. Construct an abstract tag table, which includes multiple abstract tags, and establish a mapping relationship table between the abstract tag table and the tags; The tagged communication text set is mapped and matched with the abstract tag table according to the mapping relationship table, so as to obtain the abstract tag corresponding to each tagged communication text in the tagged communication text set; The abstract tag is matched with a preset tag content library to obtain the tag content corresponding to the abstract tag. The tag content is then added to the corresponding tag communication text to obtain the standard communication text.

3. The method for early warning of negative emotions, thoughts, and self-harming behaviors in adolescents based on artificial intelligence, as described in claim 2, is characterized in that: Text tags are added to each of the communication texts in the communication text set to obtain a tagged communication text set, including: Extract the text topic features of each of the communication texts in the communication text set; The similarity model is used to calculate the similarity between the text topic features and each tag in a pre-set tag table. The tag with the highest similarity is then used as the tag for the communication text. The similarity model is as follows: Where M represents the similarity metric, Let θ0 represent the vector residual value of the semantic information of the exchanged text fragment, θ0 represent the initial value of the vector residual of the semantic information of the exchanged text fragment, and t represent the initial value of the similarity index. This represents the general assignment conditions for substituted parameters, b1, b2, ..., b n A classification index representing the semantic information of n distinct similar text fragments; The markers of each text in the set of text exchanges are counted, and the markers are added to the text exchanges to obtain a set of marked text exchanges.

4. The method for early warning of negative emotions, thoughts, and self-harming behaviors in adolescents based on artificial intelligence, as described in claim 1, is characterized in that: The first major language model is used to guide the target adolescents to talk about emotion-related topics, including: A first knowledge base is constructed, and a first reminder is constructed based on the first knowledge base. The first reminder is used to debug the first large language model, so that the first large language model can be used to guide the target adolescents to talk about topics related to emotions.

5. The method for early warning of negative emotions, thoughts, and self-harming behaviors in adolescents based on artificial intelligence, as described in claim 1, is characterized in that: The step of inputting the standard communication text into the pre-trained second language model also includes: A second knowledge base is constructed, and a second reminder is built based on the second knowledge base. The second reminder is used to debug the second large language model, thereby obtaining a trained second large language model.

6. The method for early warning of negative emotions, thoughts, and self-harming behaviors in adolescents based on artificial intelligence, as described in claim 1, is characterized in that: The standard communication text and the analysis results of the emotional disorder are input into a pre-trained third language model. The pre-trained third language model outputs graded warning results and suggestions, and also includes: A third knowledge base is constructed, and a third reminder is constructed based on the third knowledge base. The third reminder is used to debug the third language model, thereby obtaining a trained third language model.

7. The method for early warning of negative emotions, thoughts, and self-harming behaviors in adolescents based on artificial intelligence, as described in claim 1, is characterized in that: The tiered early warning system includes Level 1, Level 2, Level 3, and Level 4 warnings, and the recommendations include Level A, Level B, Level C, Level D, and Level E handling.

8. An AI-based early warning system for negative emotions, thoughts, and self-harming behaviors in adolescents, applicable to the AI-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents as described in any one of claims 1-7, comprising an interactive data acquisition module (1), a data processing and conversion module (2), an emotion analysis module (3), an early warning and suggestion module (4), and a result presentation module (5), characterized in that: The interactive data acquisition module (1) is used to acquire the interactive voice and text of teenagers using the first language model, and guide them to talk about topics related to emotions; The data processing and conversion module (2) is used to convert speech into text and perform preprocessing to obtain standard communication text; The emotion analysis module (3) is used to analyze standard communication text using the second largest language model, output emotion symptom vector labels, and use a convolutional neural network model to analyze the emotion symptom vectors to determine the emotional state and degree of impairment of adolescents. The warning and suggestion module (4) is used to output warning levels and suggestions based on sentiment analysis using the third language model; The results presentation module (5) is used to present the sentiment analysis results, warning levels, and recommendations.

9. An artificial intelligence-based early warning device for negative emotions, thoughts, and self-harming behaviors in adolescents, applicable to the artificial intelligence-based early warning method for negative emotions, thoughts, and self-harming behaviors in adolescents as described in any one of claims 1-7, characterized in that, include: At least one processor (6) and memory (7); The memory (7) is used to store computer execution instructions; The at least one processor (6) is used to execute computer execution instructions stored in the memory (7) to implement an artificial intelligence-based early warning method for negative emotions, thoughts and self-harm behaviors in adolescents as described in any one of claims 1-7.

Images