A child positive guidance prompt word generation method and system based on personality-emotion interaction

Abstract

The application discloses a kind of child positive guidance prompt word generation method and system based on personality-emotion interaction belongs to artificial intelligence and educational technology cross technical field.Method includes: obtaining the comprehensive state vector of target child and age stage;Three-dimensional interaction tensor is constructed to obtain the interaction score of different guidance strategies;The rule prior matrix is obtained by injecting domain expert knowledge into strategy selection;Based on the comprehensive state vector, interaction score and rule prior matrix, the adaptation degree of different guidance strategies is obtained;Based on the optimal strategy combination of adaptation degree, the structured prompt word sequence is generated using conditional language model;Four-dimensional quality assessment is carried out on the structured prompt word sequence, and the optimal prompt word sequence is obtained based on the evaluation result.The application solves the problems of personality-emotion interaction modeling missing, simple strategy selection mechanism and lack of structure in the existing technology guidance process, and realizes the precise, safe and positive guidance for children and adolescents aged 6-18.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of artificial intelligence and educational technology, specifically relating to a method and system for generating positive guidance prompts for children based on personality-emotion interaction. Background Technology

[0002] With the widespread application of artificial intelligence in education, intelligent tutoring systems targeting children and adolescents are increasing. However, existing technologies have significant shortcomings in personalized guidance. Specifically: (1) General language models represented by GPT series, ChatGLM and others have the following problems in children's education scenarios: the positiveness and educational value of the generated content are difficult to guarantee; there is a lack of differentiated guidance for different personality traits; the response to emotional state is singular and the strategy selection is not accurate enough; there is a lack of age adaptation mechanism; and the security constraints are not strict enough.

[0003] (2) Existing affective computing technologies can identify user emotions, but they have limitations in terms of guidance strategies: the mapping between emotion recognition and guidance strategies is relatively simple; personality characteristics are not taken into account in strategy selection; the differentiated responses of different personality types to the same emotion are not modeled; and there is a lack of structured guidance process design.

[0004] (3) Existing educational AI patents mainly focus on knowledge question answering, learning path recommendation, etc., and have the following problems in psychological guidance: lack of personality-emotion interaction modeling, resulting in a mismatch between guidance strategies and user characteristics; guidance strategy selection relies on simple rules; the quality assessment dimension of generated content is single, which cannot guarantee positiveness and security; and a complete structured guidance framework has not been formed. Summary of the Invention

[0005] The technical problem to be solved by this invention is: how to achieve accurate matching of guidance strategies based on deep interaction modeling of children's personality traits and emotional states, and generate structured, positive and safe guidance prompts.

[0006] To achieve the above objectives, the present invention provides the following solution: a method for generating positive guidance prompts for children based on personality-emotion interaction, comprising the following steps: S1. Obtain the comprehensive state vector and age stage of the target child; S2. Construct a three-dimensional interaction tensor and obtain the interaction scores of different guidance strategies based on the three-dimensional interaction tensor; S3. Inject domain expert knowledge into strategy selection to obtain the rule prior matrix; S4. Based on the comprehensive state vector, the interaction score, and the rule prior matrix, the fit of different guidance strategies is obtained; S5. Based on the fit, obtain the optimal strategy combination, and based on the optimal strategy combination, generate a structured prompt word sequence using a conditional language model; the structured prompt word sequence includes: opening prompt words, guiding prompt words, empowering prompt words, and action prompt words; S6. Perform a four-dimensional quality assessment on the structured prompt word sequence, and obtain the optimal prompt word sequence based on the results of the four-dimensional quality assessment.

[0007] Preferably, in S2, the method for obtaining the interaction score includes: ; In the formula, Representation Strategy Interaction score; Represents a personality trait vector; superscript Indicates transpose; Representation Strategy 3D interaction tensor; Represents an emotional state vector; Represents the dimension of the personality trait vector; Represents the dimension of the emotional state vector; Indicates interaction effect; Indicates the first A dimensional personality trait vector; Indicates the first A dimensional emotional state vector.

[0008] Preferably, in S3, the rule prior matrix includes: ; In the formula, Representation Strategy The rule prior matrix; Indicates the total number of preset rules; Indicates the rule weight; Represents a rule function; This indicates an age group code.

[0009] Preferably, the rule function includes: ; ; ; ; ; ; In the formula, This indicates a rule prioritizing emotions; This indicates the outward adaptation rule; This indicates open adaptation rules; Indicates age-appropriateness rules; This indicates the rules for resilience and adaptation. This indicates the rules for calming nervousness; Indicates an indicator function; Indicates valence; Indicates intensity; Indicates extroversion; Indicates openness; Indicates resilience; It indicates neuroticism.

[0010] Preferably, in S4, the method for calculating the adaptability of different guidance strategies includes: ; In the formula, Represents the strategy fit vector; This represents the Sigmoid activation function; Represents the strategy classification weight matrix; Represents the integrated state vector; Represents the bias vector; Represents the prior vector of the rule; Represents the interaction score vector; Indicates the number of guidance strategies.

[0011] Preferably, in S5, the method for generating prompt words includes: ; In the formula, This means generating a complete sequence of structured prompt words given a comprehensive state vector, optimal policy combination, and age stage. The probability of; Represents probability; This represents the optimal strategy combination; Indicates the total length of the prompt word sequence; Indicates the first Each word element; Indicates all locations The previous lexical sequence.

[0012] Preferably, in S6, the method for four-dimensional quality assessment includes: ; In the formula, This indicates the overall quality score; , , , These represent the fit score, positivity score, safety score, and coherence score, respectively. , , , These represent the fit rating coefficient, positiveness rating coefficient, safety rating coefficient, and consistency rating coefficient, respectively.

[0013] Preferably, the fit score includes: ; In the formula, This indicates a text editor that displays a structured sequence of prompts. Encoded as a composite state vector Semantic vectors of the same dimension; The positive score includes: ; In the formula, Indicator; This indicates a positive vocabulary database; This indicates a negative vocabulary. The security score includes: ; In the formula, Represents the hazard scoring function; Represents a text segment; The coherence score includes: ; In the formula, Indicates the first Semantic embedding vectors of class prompt words; Indicates the first Semantic embedding vectors of class prompt words.

[0014] This invention also provides a system for generating positive guidance prompts for children based on personality-emotion interaction, comprising: The data acquisition module is used to obtain the comprehensive state vector and age stage of the target child; An interaction module is used to construct a three-dimensional interaction tensor and obtain interaction scores for different guidance strategies based on the three-dimensional interaction tensor. The fusion module is used to inject domain expert knowledge into strategy selection to obtain a rule prior matrix; The fit calculation module is used to obtain the fit of different guidance strategies based on the comprehensive state vector, the interaction score and the rule prior matrix. The generation module is used to obtain the optimal strategy combination based on the adaptability, and to generate a structured prompt word sequence based on the optimal strategy combination using a conditional language model; the structured prompt word sequence includes: opening prompt words, guiding prompt words, empowering prompt words, and action prompt words; The evaluation module is used to perform a four-dimensional quality evaluation on the structured prompt word sequence, and obtain the optimal prompt word sequence based on the result of the four-dimensional quality evaluation.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves refined strategy adaptation by explicitly modeling the nonlinear interaction between personality and emotion using three-dimensional interactive tensors; it injects expert knowledge from the fields of child psychology and education into the neural network in the form of rules to improve the professionalism and interpretability of strategy selection; it uses four types of prompts—opening, guiding, empowering, and action—to form a complete guidance process; and it evaluates the generated content from four dimensions—adaptability, positivity, safety, and coherence—to ensure output quality. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are 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.

[0017] Figure 1 This is a schematic diagram of the flowchart of the method for generating positive guidance prompts for children based on personality-emotion interaction according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the rule prior matrix calculation process in an embodiment of the present invention; Figure 3 This is a schematic diagram of the strategy matching and selection process in an embodiment of the present invention; Figure 4 This is a schematic diagram of the security filtering process according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the high-risk early warning mechanism in an embodiment of the present invention; Figure 6 This is the structured prompt word output result for the scenario of failing an exam, as described in Embodiment 2 of the present invention. Figure 7 This is the structured cue word output result for social anxiety scenarios in Embodiment 3 of the present invention; Figure 8 This is the structured prompt word output result for a learning motivation lack scenario in Embodiment 4 of the present invention; Figure 9 This is the structured prompt word output result for an emergency emotional crisis scenario in Embodiment 5 of the present invention; Figure 10 This is the structured prompt output result for parent-child conflict scenarios in Embodiment Six of the present invention. Detailed Implementation

[0018] 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.

[0019] 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.

[0020] Example 1: like Figure 1 As shown, this embodiment provides a method for generating positive guidance prompts for children based on personality-emotion interaction, including the following steps: S1, obtaining the comprehensive state vector and age stage of the target child; S2, constructing a three-dimensional interaction tensor and obtaining interaction scores for different guidance strategies based on the three-dimensional interaction tensor; S3, injecting domain expert knowledge into strategy selection to obtain a rule prior matrix; S4, obtaining the fit of different guidance strategies based on the comprehensive state vector, the interaction score, and the rule prior matrix; S5, obtaining the optimal strategy combination based on the fit, and generating a structured prompt sequence using a conditional language model based on the optimal strategy combination; the structured prompt sequence includes: opening prompts, guidance prompts, empowering prompts, and action prompts; S6, performing a four-dimensional quality assessment on the structured prompt sequence, and obtaining the optimal prompt sequence based on the results of the four-dimensional quality assessment.

[0021] In S1, the comprehensive state vector This is the result of multimodal feature fusion, encompassing comprehensive information on context, personality, and emotion. Among these, the personality feature vector... Adopting a children's version of the Big Five personality traits extended model, Representing the dimensions of the personality trait vector: In the formula, It signifies openness, including curiosity, imagination, and receptiveness to new things; This indicates conscientiousness, including: self-discipline, organization, and degree of goal orientation; Extraversion is characterized by: social energy, willingness to express oneself, and interaction preferences; Agreeableness is characterized by: cooperativeness, empathy, and interpersonal harmony; Indicates neuroticism; - Learning styles are categorized into visual, auditory, kinesthetic, and reading / writing preferences, respectively. - These represent motivation types, specifically the intensity of intrinsic, extrinsic, and social drives. Represents resilience, which is the ability to recover from setbacks; values ​​range from [0,1]; superscript This represents the transpose. Emotional state vector. use +Dual-track system for discrete emotions Dimensions of the emotional state vector: In the formula, This represents valence, with a value range of [-1, 1]. Indicates wakefulness; It indicates a sense of control; - These are discrete emotions, representing happiness, sadness, anger, fear, anxiety, frustration, pride, shame, curiosity, and boredom. Indicates intensity, representing the overall intensity of the emotion, with values ​​ranging from [0,1]. The value ranges from {0, 1, 2} to represent stability, fluctuation, and rapid change, respectively. Discrete coding is used for age groups. Among them, code 1 represents 6-11 years old, the cognitive development stage is the concrete operational stage, and the appropriate strategy characteristics are: gamification, storytelling, simple language and vivid metaphors; code 2 represents 12-14 years old, the cognitive development stage is the early formal operational stage, and the appropriate strategy characteristics are: peer perspective, self-support and moderate challenge; code 3 represents 15-18 years old, the cognitive development stage is the mature formal operational stage, and the appropriate strategy characteristics are: rational discussion, future orientation and in-depth exploration.

[0022] In S2, the three-dimensional interaction tensor , This indicates the number of guidance strategies. In this embodiment, the set of guidance strategies includes at least eight strategies: emotional empathy, interactive dialogue, introspective guidance, exploratory encouragement, structured guidance, growth mindset guidance, gamification, and rational discussion. (The last sentence appears to be incomplete and possibly refers to a separate concept: "For strategies...") The calculation method for its interaction score includes: ; In the formula, Representation Strategy The interaction score indicates the current personality-emotion combination and strategy; a higher score indicates a better understanding of the interaction. The better the match; This indicates that the fixed third-dimensional index is The result obtained later matrix; Representing personality dimensions With the emotional dimension strategy The interaction weights are the three-dimensional interaction tensor. The middle is located in the first line, number Column, No. Layer elements; The first character represents the personality trait vector. One component; The first element representing the emotional state vector Each component.

[0023] For scenarios with high computational efficiency requirements, tensor decomposition can be used to reduce complexity. ; In the formula, Indicates the outer rank; Indicates the outer product operation; Representation Strategy The Individual personality basis vectors; Representation Strategy The A number of emotion basis vectors.

[0024] have There are 13 independent parameters (e.g., 13 × 15 = 195), and each component after decomposition only needs 13 independent parameters (e.g., 13 × 15 = 195). One parameter, Each component requires a total of [number] components. One parameter. When much smaller Time (e.g.) =3), the number of parameters was reduced significantly from 195 to 3×28=84, which significantly reduced the amount of computation and storage requirements, while retaining the main pattern of personality-emotion interaction.

[0025] The 3D interaction tensor explicitly models the mapping relationship of "personality × emotion → strategy," an interaction modeling method that existing linear fusion methods cannot achieve. It can capture the non-linear synergistic effect of personality and emotion. A specific example is: if a user is neurotic... (High) Anxiety (High), and If the score is relatively high after training (e.g., 0.9), then this single interaction contributes 0.82 × 0.75 × 0.9 = 0.554 points, strongly encouraging the system to choose the "emotional empathy" strategy. This bilinear interaction modeling allows the model to capture "personality". In emotions It is a "special response pattern" rather than simply treating personality and emotions separately.

[0026] The mapping relationship is shown in Table 1.

[0027] Table 1 In S3, such as Figure 2 As shown, the rule prior matrix encodes the knowledge of child psychology and education experts into mathematical form: ; In the formula, Representation Strategy The rule prior matrix; This indicates the total number of preset rules, which is 6 in this embodiment; Indicates the first The weight of each rule; Represents a rule function; This indicates an age group code.

[0028] In this embodiment, the rule function and the corresponding weights include: ; =0.8; ; =0.6; ; =0.5; ; =0.7; ; =0.6; ; =0.5; In the formula, This indicates an emotion-priority rule: when strong negative emotions are detected, the "emotional empathy" strategy is selected first. This indicates the extroversion-fit rule: highly extroverted individuals prefer interactive dialogue, while less extroverted individuals prefer introspective guidance. This indicates that openness is suitable for exploration and encouragement, while low openness is suitable for structure-guided approaches. This indicates an age-appropriate rule: gamification is suitable for childhood, while rational discussion is suitable for adolescence. This indicates that individuals with low resilience need more guidance in developing a growth mindset when faced with frustrating situations. This indicates the rules for calming neuroticism; highly neurotic individuals need more emotional empathy when experiencing negative emotions. Indicates an indicator function.

[0029] In S4, the bootstrap strategy set is first defined: ,in , This indicates the number of bootstrapping strategies. The bootstrapping strategy numbers are shown in Table 2.

[0030] Table 2 like Figure 3 As shown, the strategy selection adopts rule-enhanced multi-label classification, and the calculation method for the fit of different guidance strategies includes: ; In the formula, The policy fit vector contains element each Representation Strategy The fit is within the range of [0, 1]. The closer it is to 1, the stronger the strategy. The closer it matches the current user's state. This represents the Sigmoid activation function; Represents the strategy classification weight matrix; This represents the dimension of the integrated state vector; Represents the bias vector; Represents the prior vector of the rule; This represents the interaction score vector.

[0031] In S5, a threshold-based activation strategy is used to obtain the optimal strategy combination, meaning that all activated strategies constitute the optimal strategy combination: ; ; In the formula, Representation Strategy The activation threshold is preferably set to 0.5 in this embodiment; This represents the optimal strategy combination.

[0032] The structured prompt word sequence is represented as follows: It consists of 4 prompt words; among them, These are opening words used to elevate the conversation, confirm the speaker's state, and establish a connection. They are recommended to be 15-30 characters long. These are guiding prompts used for core guiding content, question restructuring, and perspective shifts. The recommended length is 30-60 characters. Empowering prompts are used to stimulate intrinsic motivation, evoke successful experiences, and confirm abilities. The recommended length is 20-40 characters. These are action prompts used to provide specific action suggestions and next steps. They are recommended to be 15-30 characters long.

[0033] Methods for generating prompt words using conditional language models include: ; In the formula, This means generating a complete sequence of structured prompt words given a comprehensive state vector, optimal policy combination, and age stage. The probability; the higher the probability value, the more the generated structured prompt word sequence meets the requirements of the current conditions; Represents probability; Indicates the total length of the prompt word sequence; Indicates the first Each word element; Indicates all locations The previous lexical sequence.

[0034] Introducing personality fit constraints during the decoding process: ; In the formula, Indicates the adjusted number each word element The logarithmic probability; This represents the probability of a lexical term after adjustment for personality fit constraints. Indicates the original probability of a word; Indicates the first The weight coefficient of a constraint is such that the larger the value, the stronger the influence of the corresponding constraint. Indicates the first The constraint function of a certain type of constraint.

[0035] in, To constrain vocabulary complexity, the vocabulary difficulty is adjusted based on age, with a corresponding weighting coefficient of 0.3; To constrain the tone of voice, highly neurotic individuals are advised to use a gentler tone, with a weighting factor of 0.2. To constrain the interactivity matching, highly extroverted individuals are given additional interactive elements with a weighting coefficient of 0.2.

[0036] In S6, the methods for four-dimensional quality assessment include: ; In the formula, This represents the overall quality score, with a value range of [0,1]. A higher value indicates a better overall quality of the generated prompts. This represents the fit score, which measures the degree of matching between the generated prompts and the user's overall state. The value range is [0,1]. This represents the positiveness score, used to measure the positive guidance of the generated content. The value ranges from [-1, 1], with a larger positive value indicating stronger positiveness. This represents a safety score, used to measure the safety of generated content for children. The value ranges from [0,1], with the closer to 1 being the safer. The coherence score measures the quality of logical connection between the four types of cue words. The value ranges from [0,1], with a higher value indicating a more natural transition between the cue words. , , , These represent the fit rating coefficient, positiveness rating coefficient, safety rating coefficient, and coherence rating coefficient, respectively; this embodiment preferably uses... , , , .

[0037] The fit score uses the cosine similarity between the fused vector and the cue word embedding, including: ; In the formula, This indicates a text editor that displays a structured sequence of prompts. Encoded as a composite state vector Semantic vectors of the same dimension.

[0038] The positive positivity score is calculated by subtracting the negative word percentage from the positive word percentage, including: ; In the formula, Indicator; This indicates a positive vocabulary database; This indicates a negative vocabulary.

[0039] The safety score is calculated by subtracting the maximum hazard score from 1, including: ; In the formula, This represents a hazard rating function, with a text segment as input. and age group The output is a score in the range [0,1].

[0040] The coherence score is calculated using the average cosine similarity of adjacent cue word embeddings, including: ; In the formula, Indicates the first Semantic embedding vectors of class prompt words; Indicates the first Semantic embedding vectors of class prompt words.

[0041] At the same time, a safety assessment is conducted based on age grading thresholds: When =1, the hazard score threshold The value is 0.05, which is the strictest range, prioritizing child protection.

[0042] When =2, the hazard score threshold The value is 0.10, which is a strict interval.

[0043] When =3, the hazard score threshold The value is 0.15, which is relatively lenient but still provides some protection.

[0044] A high-level warning is triggered when a user is detected expressing self-denial or self-harm tendencies.

[0045] When the overall quality score When the value is ≥0.8, the current structured prompt word sequence is the optimal prompt word sequence. The current structured prompt word sequence is directly output and the next step, the security filtering layer, is used for rule-level detection.

[0046] When 0.6≤ When the value is less than 0.8, the structured prompt word sequence is regenerated after adjusting the strategy (up to 3 adjustments); in this embodiment, the identification... The dimension with the lowest score was regenerated after targeted adjustments: if At its lowest, the policy matching layer lowers the activation threshold. Regenerate after increasing the number of candidate strategies. At the very least, increase the weight of positive word preference in personality fit constraints. Then regenerate. If Minimum, remove The keyword type of the highest-scoring segment was replaced with a safe template and re-evaluated. If At the very least, regenerate after increasing the context window length between adjacent prompts during generation.

[0047] when If the value is less than 0.6 (meaning three consecutive adjustments fail to reach 0.8), the system reverts to the safe template library. The current generated result is discarded, and a prompt word sequence is selected from the pre-set safe template library. The safe template library is a collection of pre-written and security-approved guidance texts, organized according to a three-dimensional index of "age stage × emotion type × strategy type." Based on the current user's age stage, primary emotion type, and activated strategy type, the closest template sequence is matched and output. The template sequence... All values ​​are pre-validated to be ≥0.9 to ensure the quality and security of the output.

[0048] like Figure 4 As shown, the security filtering layer performs multi-channel content security checks. Output is allowed only if all checks pass; otherwise, content is regenerated. Specifically, the detection methods include: ; In the formula, This represents the output of the security filtering layer, with a value of 0 or 1. A value of 1 indicates that the prompt word sequence has passed all security detection channels, while a value of 0 indicates that it has not passed. This indicates the total number of detection channels; in this embodiment, =5, including: keyword matching, semantic classification, sentiment analysis, age suitability detection, and intent classification; Indicates the first The detection function for each detection channel takes a text fragment as input. and age group The output is or .

[0049] Table 3 shows examples of detection channels and prohibited content categories.

[0050] Table 3 when When the value is 0, the following processing is performed: Step 1: Locate the offending segments and channels, record the segments that failed and the channel numbers that failed, and generate an interception log.

[0051] Step 2: Targeted replacement and regeneration. Only for the prompt word type to which the violation segment belongs, the corresponding prohibition rule of the channel is added to the generation constraints and the prompt word of that type is regenerated. The other approved prompt words remain unchanged.

[0052] Step 3: Re-test. Perform security checks on all channels of the replaced sequence again.

[0053] Step 4: Repeat the test. If the test still fails after two targeted replacements, abandon the result generation and revert to the safe template library.

[0054] At the same time, such as Figure 5 As shown, this embodiment also includes a high-risk early warning mechanism. This high-risk early warning is independent of security filtering detection. For high-risk signals in user input, when it detects that the user expresses severe self-denial or persistent self-deprecation, a high-risk response is immediately triggered without waiting for the generation and evaluation process to complete. The process includes: Step 1: Generate alarm information and output a structured alarm that includes the warning level, cause, and recommended measures.

[0055] Step 2: Forced Coverage of Strategies. Regardless of the strategy matching result, force the strategy combination to cover only "emotional empathy" to ensure that the response is centered on empathy.

[0056] Step 3: Continue generating prompts. Based on the strategy after forced coverage, generate four types of prompts, but the prompts must include concern for the user's experience and inquiries about their willingness to learn more.

[0057] Step 4: Synchronize notifications and push alarm information to the guardian or teacher's end, and record it in the security event log.

[0058] Step 5: Include supplementary resources. In the output, include professional resource information such as psychological assistance hotlines.

[0059] Example 2: This embodiment uses the scenario of failing an exam as an example to illustrate the method proposed in this invention.

[0060] Input information is as follows: Subject: 14-year-old boy; Scenario: Scored 65 points on the midterm math exam, lower than expected; Personality vector: (Moderate openness, high conscientiousness, moderately high neuroticism, moderate resilience). (Negative valence, high sadness, high depression, mood swings); Age group: =2 (adolescence); Comprehensive state vector: .

[0061] Processing procedure: 1. Personality-Emotion Interaction Calculation: (Emotional empathy) = High conscientiousness × High frustration = 0.78 × 0.68 = 0.53; (Growth mindset) = Medium resilience × High frustration = (1-0.48) × 0.68 = 0.35.

[0062] 2. Prior calculation of rules: Emotion priority rule: =-0.55<-0.3, =0.72, =0.8 × 0.72 = 0.58; Age suitability: =2, does not trigger gamification or rational discussion special rules.

[0063] 3. Strategy adaptability: (Emotional Empathy Type) = 0.82 → Activated; (Growth mindset) = 0.71 → Activated; (Rational discussion type) = 0.65 → Activated.

[0064] 4. The output of structured prompt word generation is as follows: Figure 6As shown, following the structured prompts, opening prompts, guiding prompts, empowering prompts, and action prompts are provided. Combined with appropriate strategies, the following message is generated: "Your exam results didn't quite match your efforts, which must be upsetting. You've always been a person who sets high standards for yourself. Let's look together at which points were lost due to carelessness and which knowledge points need strengthening. Breaking them down will make the problems clearer. Remember that physics test last semester? You found it difficult, but you spent two weeks focusing on that section and improved by over 20 points on the final exam. If we categorize the incorrect answers into three types, which type would you like to start working on first?" Example 3: This embodiment uses a social anxiety scenario as an example to illustrate the method proposed in this invention.

[0065] Input information is as follows: Subject: 10-year-old girl; Scenario: transferred schools a month ago, always eating lunch alone; Personality vector: (Low extraversion, high agreeableness, high neuroticism, low resilience); (High anxiety, high fear, low sense of control); Age group: =1 (childhood).

[0066] Processing procedure: 1. Personality-emotion interaction calculation: Low extraversion × high anxiety → avoid direct social interaction; high agreeableness × fear → emphasize the possibility of "being accepted".

[0067] 2. Strategy adaptability: (Emotional Empathy Type) = 0.88 → Activated; (Introspective-guided) = 0.72 → Activated; (Gamified type) = 0.85 → Activated.

[0068] 3. The output of structured prompt word generation is as follows: Figure 7 As shown, the final output is: It's normal to feel a little nervous when you arrive in a new place surrounded by unfamiliar people; many children feel this way when transferring schools. Have you noticed if there are any quiet classmates? Sometimes finding one good friend is more important than knowing many people. Imagine you are a little hedgehog who has just arrived in a new forest. You can observe from your own little corner and slowly approach when you find a safe friend. Although hedgehogs are slow, they can eventually make good friends! Tomorrow at school, would you like to try smiling at a classmate who seems friendly? No need to say anything.

[0069] Example 4: This embodiment uses a scenario of lack of learning motivation as an example to illustrate the method proposed in this invention.

[0070] Input information is as follows: Subject: 16-year-old male; Scenario: Second semester of first year of high school, lost interest in studying, only wants to play games every day; Personality vector: (High openness, high outward orientation, low due diligence, strong external driving force); (Mildly negative feelings, strong sense of boredom, still possessing curiosity); Age group: =3 (adolescence).

[0071] Processing procedure: 1. Personality-emotion interaction calculation: High extroversion × boredom → need for novel stimulation and meaning exploration; curiosity still exists → entry point.

[0072] 2. Strategy adaptability: (Interactive Dialogue Type) = 0.78 → Activated; (Exploration-encouraging type) = 0.82 → Activated; (Rational discussion type) = 0.75 → Activated (age-appropriate).

[0073] 3. The output of structured prompt word generation is as follows: Figure 8 As shown, the final output is: It sounds like schoolwork doesn't excite you, but playing games is more fun? I'm quite curious—what's most attractive about games? Is it the thrill of competition, exploring new worlds, or the sense of accomplishment from teamwork? Actually, these feelings can be found in real life. For example, if you're interested in competition, debates or programming competitions might get you hooked; if you like exploration, there are all sorts of majors and research directions in university that you've never seen before. Would you like to spend 10 minutes exploring together which majors or fields might engage you as much as playing games? Example 5: This embodiment uses an emergency emotional crisis scenario as an example to illustrate the method proposed in this invention.

[0074] Input information is as follows: Subject: 13-year-old girl; Scenario: Publicly ridiculed by her best friend in the class group chat, feeling "I can't do anything right, I'm just a superfluous person"; Personality vector: (High agreeableness, high neuroticism, low resilience); (Extremely negative, extremely high sadness, high shame, extremely low sense of control); Age group: =2 (adolescence).

[0075] Processing procedure: 1. Personality-emotion interaction calculation: High neuroticism × extreme sadness → highest priority of emotional empathy; Low resilience × shame → requires particularly careful support.

[0076] 2. Safety Detection: If the system detects "I can't do anything right, I'm just a superfluous person" → a high-risk warning is triggered.

[0077] 3. Strategy adaptability: (Emotional Empathy Type) = 0.96 → Activation (almost the only strategy).

[0078] 4. The output of structured prompt word generation is as follows: Figure 9 As shown, the final output is: Being said that way by a good friend in a group chat must be incredibly painful. That feeling of betrayal is truly agonizing. You said, "I feel like an unwanted person," and I'm very worried about how you're feeling. Would you be willing to tell me more? What's hurting you the most right now? The fact that you're willing to tell me these things shows you're looking for a way out, which is important. The pain you're experiencing now is real, but it won't always be this intense. The most important thing right now is whether you have an adult who can help you—your parents, teacher, or other family member? If so, I hope you can tell them how you're feeling.

[0079] Example 6: This embodiment uses a parent-child conflict scenario as an example to illustrate the method proposed in this invention.

[0080] Input information is as follows: Subject: 8-year-old boy; Scenario: He had a big argument with his mother because he didn't want to practice piano, and hid in his room crying; Personality vector: (High openness, high extroversion, low conscientiousness); (High anger, high frustration, low sense of control); Age group: =1 (childhood).

[0081] The output of the structured prompt word generation is as follows: Figure 10 As shown, the final output is: You argued with your mom, and you must feel angry and sad. It's okay to cry; sometimes tears can help us release our unhappiness. What do you least want to do when practicing the piano? Is it because the piece is too difficult, the practice time is too long, or you want to do something else? You know, many children don't like practicing the piano. The fact that you're willing to tell me shows that you're a brave child. Mom might just be too anxious; she does love you. When you're in a better mood, we can think together about how to make practicing the piano less unpleasant, okay? For example, how about starting with a piece you like? Example 7: This embodiment provides a positive guidance prompt word generation system for children based on personality-emotion interaction, used to implement the generation method provided in Embodiment 1, including: a data acquisition module for acquiring the comprehensive state vector and age stage of the target child; an interaction module for constructing a three-dimensional interaction tensor and obtaining interaction scores for different guidance strategies based on the three-dimensional interaction tensor; a fusion module for injecting domain expert knowledge into strategy selection to obtain a rule prior matrix; a fit calculation module for obtaining the fit of different guidance strategies based on the comprehensive state vector, interaction scores, and rule prior matrix; a generation module for obtaining the optimal strategy combination based on the fit, and generating a structured prompt word sequence using a conditional language model based on the optimal strategy combination; the structured prompt word sequence includes: opening prompt words, guiding prompt words, empowering prompt words, and action prompt words; and an evaluation module for performing a four-dimensional quality evaluation on the structured prompt word sequence, and obtaining the optimal prompt word sequence based on the results of the four-dimensional quality evaluation.

[0082] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for generating positive guidance prompts for children based on personality-emotion interaction, characterized in that, Includes the following steps: S1. Obtain the comprehensive state vector and age stage of the target child; S2. Construct a three-dimensional interaction tensor and obtain the interaction scores of different guidance strategies based on the three-dimensional interaction tensor; S3. Inject domain expert knowledge into strategy selection to obtain the rule prior matrix; S4. Based on the comprehensive state vector, the interaction score, and the rule prior matrix, the fit of different guidance strategies is obtained; S5. Based on the fit, obtain the optimal strategy combination, and based on the optimal strategy combination, use a conditional language model to generate a structured prompt word sequence; The structured prompt sequence includes: opening prompts, guiding prompts, empowering prompts, and action prompts; S6. Perform a four-dimensional quality assessment on the structured prompt word sequence, and obtain the optimal prompt word sequence based on the results of the four-dimensional quality assessment.

2. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 1, characterized in that, In S2, the method for obtaining the interaction score includes: ； In the formula, Representation Strategy Interaction score; Represents a personality trait vector; superscript Indicates transpose; Representation strategy 3D interaction tensor; Represents an emotional state vector; Represents the dimension of the personality trait vector; Represents the dimension of the emotional state vector; Indicates interaction effect; Indicates the first A dimensional personality trait vector; Indicates the first A dimensional emotional state vector.

3. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 2, characterized in that, In S3, the rule prior matrix includes: ； In the formula, Representation strategy The rule prior matrix; Indicates the total number of preset rules; Indicates the rule weight; Represents a rule function; This indicates an age group code.

4. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 3, characterized in that, The rule functions include: ； ； ； ； ； ； In the formula, This indicates a rule prioritizing emotions; This indicates the outward adaptation rule; This indicates open adaptation rules; Indicates age-appropriateness rules; This indicates the rules for resilience and adaptation. This indicates the rules for calming nervousness; Indicates an indicator function; Indicates valence; Indicates intensity; Indicates extroversion; Indicates openness; Indicates resilience; It indicates neuroticism.

5. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 4, characterized in that, In S4, the methods for calculating the fit of different guidance strategies include: ； In the formula, Represents the strategy fit vector; This represents the Sigmoid activation function; Represents the strategy classification weight matrix; Represents the integrated state vector; Represents the bias vector; Represents the prior vector of the rule; Represents the interaction score vector; Indicates the number of guidance strategies.

6. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 5, characterized in that, In S5, the methods for generating prompt words include: ； In the formula, This means generating a complete sequence of structured prompt words given a comprehensive state vector, optimal policy combination, and age stage. The probability of; Represents probability; This represents the optimal strategy combination; Indicates the total length of the prompt word sequence; Indicates the first Each word element; Indicates all locations The previous lexical sequence.

7. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 6, characterized in that, In S6, the method for four-dimensional quality assessment includes: ； In the formula, This indicates the overall quality score; , , , These represent the fit score, positivity score, safety score, and coherence score, respectively. , , , These represent the fit rating coefficient, positiveness rating coefficient, safety rating coefficient, and consistency rating coefficient, respectively.

8. The method for generating positive guidance prompts for children based on personality-emotion interaction according to claim 7, characterized in that, The compatibility score includes: ； In the formula, This indicates a text editor that displays a structured sequence of prompts. Encoded as a composite state vector Semantic vectors of the same dimension; The positive score includes: ； In the formula, Indicator; This indicates a positive vocabulary database; This indicates a negative vocabulary. The security score includes: ； In the formula, Represents the hazard scoring function; Represents a text segment; The coherence score includes: ； In the formula, Indicates the first Semantic embedding vectors of class prompt words; Indicates the first Semantic embedding vectors of class prompt words.

9. A system for generating positive guidance prompts for children based on personality-emotion interaction, the system being used to implement the generation method according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to obtain the comprehensive state vector and age group of the target child. An interaction module is used to construct a three-dimensional interaction tensor and obtain interaction scores for different guidance strategies based on the three-dimensional interaction tensor. The fusion module is used to inject domain expert knowledge into strategy selection to obtain a rule prior matrix; The fit calculation module is used to obtain the fit of different guidance strategies based on the comprehensive state vector, the interaction score and the rule prior matrix. The generation module is used to obtain the optimal strategy combination based on the fit, and to generate a structured prompt word sequence based on the optimal strategy combination using a conditional language model. The structured prompt sequence includes: opening prompts, guiding prompts, empowering prompts, and action prompts; The evaluation module is used to perform a four-dimensional quality evaluation on the structured prompt word sequence, and obtain the optimal prompt word sequence based on the result of the four-dimensional quality evaluation.

Images