Search

CN-122018699-A - Child emotion intelligent guiding method and system based on emotion instance objectification externalization

CN122018699ACN 122018699 ACN122018699 ACN 122018699ACN-122018699-A

Abstract

The invention discloses a child emotion intelligent guiding method and system based on emotion instance objectification externalization, and relates to the technical field of artificial intelligence and man-machine interaction. The method comprises the steps of obtaining interaction data and outputting emotion characteristic values, constructing a virtual object role object independent of a user body identifier based on an emotion instance objectification data structure, executing an emotion homing state machine protocol containing forced state circulation constraint, enabling a forced natural language generation model to output texts in stages according to a preset time sequence structure, enabling a stage A to generate a third person name naming externalization text, enabling a stage B to generate a pacifying text and synchronously outputting a first control instruction set for driving a terminal to execute pacifying action, and enabling a stage C to generate a homing guide text and synchronously outputting a second control instruction set for driving the terminal to output a rhythm sensory signal. According to the invention, the strong reference of the emotion label and the user identity is cut off through the unbinding of the data structure, the state machine protocol is utilized to forcedly restrict the large model output boundary, and the physiological characteristic closed loop parameter adjustment is combined, so that the cognitive load in the emotion guiding of children is effectively reduced, and the technical problems of uncontrollable large model output and lack of physiological quantitative evaluation are solved.

Inventors

  • CHEN BO
  • LIU TING

Assignees

  • 深圳市象形字科技股份有限公司

Dates

Publication Date
20260512
Application Date
20260311

Claims (10)

  1. 1. A method for intelligent guidance of a child's emotion based on emotion instance objectification externalization, performed by a computer device, comprising the steps of: acquiring interaction data of a target user, inputting an emotion recognition model, and outputting emotion characteristic values containing valence components and awakening degree components; instantiating a virtual object character object independent of a target user body identifier according to the emotion characteristic value, wherein the object carries a ternary attribute group comprising an independent object name identifier, a body action driving characteristic code and an externalized narrative logic label, so that the instance identifier of the virtual object character object and the body identifier of the target user have no reference relation in a data layer; The method comprises the steps of constructing a system-level prompt word and inputting a natural language generation model, wherein the system-level prompt word is configured with a resident subject intelligent agent identifier and a ternary attribute of a virtual object role object, and comprises a narrative visual angle constraint instruction, the natural language generation model is forced to output texts in stages strictly according to a preset time sequence structure, and free jump or skip of any stage is forbidden, and the preset time sequence structure at least comprises: Step A, naming externalization, namely generating a text describing the independent object name identifier by a third person name statement sentence based on the externalization narrative logic tag, wherein the output text does not contain a second person name emotion tag sentence pattern binding emotion attributes with a target user body; Step B, action pacifying, namely generating pacifying text based on the resident main body intelligent agent identifier, and synchronously outputting a first control instruction set, wherein the first control instruction set is used for driving a terminal to execute pacifying action or generate virtual interaction feedback; Stage C-state homing, namely generating a guide text containing a preset homing object identifier, and synchronously outputting a second control instruction set, wherein the second control instruction set is used for driving a terminal to output a sensory stimulation signal with a preset rhythm period; and outputting the text generated by stages and a corresponding control instruction set.
  2. 2. The method of claim 1, wherein the mood instance objectification data structure is stored in a mood object character database, and wherein a key value mapping rule is used to map a valence component and a arousal degree component of a mood feature value to the ternary attribute group; The mapping rule at least comprises: When the valence component of the emotion characteristic value is a negative value and the arousal degree component is larger than a first arousal degree threshold, assigning the externalized narrative logic label as an external trigger type, and assigning the body motion driving characteristic code as a high energy output type; and when the valence component of the emotion characteristic value is a negative value and the arousal degree component is smaller than a second arousal degree threshold, assigning the externalized narrative logic label as an internal consumption type, and assigning the body motion driving characteristic code as a low-energy retarded type.
  3. 3. The method of claim 1, wherein the state machine protocol further comprises content boundary constraints and subject forced replacement constraints: The content boundary constraint prohibits the natural language generation model from outputting text paragraphs containing emotion cause analysis, cognitive reconstruction suggestions or moral evaluation; and the subject forced replacement constraint forces the independent object name identifier to be inserted into the output text of the stage A as a subject of the sentence head, and prohibits the adoption of the human-called pronoun of the target user as the subject of the sentence head.
  4. 4. The method of claim 1, further comprising the step of physiological closed loop modulation: After the stage C is executed, physiological characteristic data of a target user is obtained, and a homing score is calculated based on the physiological characteristic data; if the homing score does not reach the preset safety threshold, dynamically updating the control parameters of the state machine protocol, and circularly executing the emotion homing state machine protocol until the termination condition is met; The dynamically updating control parameters includes adjusting at least one of a rhythmic period parameter of the sensory stimulation signal in the stage C, adjusting a length limitation parameter of the output text, or adjusting an audio playing rate parameter.
  5. 5. The method of claim 4, wherein the homing score is calculated based on a weighted sum of heart rate data, heart rate variability data, voice energy rate of change data, and limb motion amplitude rate of change data; Wherein, the sum of the weight coefficients of the heart rate and heart rate variability related indexes is not lower than 0.60 of the total weight and is higher than the weight duty ratio of other indexes.
  6. 6. The method of claim 4, wherein during execution of the loop, the loop is terminated and a guardian notification instruction is triggered when any one of the following conditions is met: the circulation times reach the upper limit of the maximum circulation times; Detecting preset high-risk semantic features in the interaction data, wherein the preset high-risk semantic features comprise semantic modes related to self-injury, attack of other people or extreme rejection; the physiological characteristic data center rate continuously exceeds a preset proportion of the safety baseline and the duration exceeds a preset duration.
  7. 7. The method of claim 1, further comprising the step of outputting a check and fault tolerance: Performing real-time verification on single output of the natural language generation model, and triggering a regeneration instruction if output content violates a time sequence structure, subject constraint or content boundary constraint of the state machine protocol; and when the retry times reach a preset threshold, executing a degradation response strategy, outputting a preset fixed pacifying script, and if the degradation response is triggered in the cyclic execution process, terminating the current cyclic rotation in advance.
  8. 8. An intelligent guidance system for child emotion based on emotion instance objectification and externalization, comprising: the perception acquisition module is used for acquiring interaction data of a target user; the emotion feature extraction module is used for mapping the interaction data into emotion feature vectors; The object role instantiation module is used for generating a virtual object role object instance carrying a ternary attribute group based on the emotion feature vector, wherein the ternary attribute group comprises an independent object name identifier, a body action driving feature code and an externalized narrative logic tag, and the instance identifier and a target user entity identifier are mutually independent in a data layer; A resident body agent module for generating and maintaining a resident body agent identifier; the state machine protocol engine is used for constructing a system-level prompt word and constraining a natural language generation model to output a text according to a preset time sequence structure, and the preset time sequence structure at least comprises a naming externalization stage, an action pacifying stage and a state homing stage; The multi-mode output module is used for executing the text output and the corresponding control instruction; and the physiological closed loop parameter adjusting module is used for collecting physiological characteristic data of the target user, calculating a homing score, and feeding back parameter adjusting instructions to the state machine protocol engine when the score is lower than a safety threshold value so as to dynamically update control parameters of the state machine protocol.
  9. 9. The system of claim 8, further comprising an output verification and fault tolerance module and a risk monitoring and notification module: the output verification and fault tolerance module is used for monitoring whether the output of the natural language generation model accords with the protocol constraint of the state machine and triggering regeneration or degradation strategies when violating rules; The risk monitoring and notifying module is used for triggering a guardian notification instruction when detecting that the high risk semantic feature or the physiological index is abnormal and continuously overtime.
  10. 10. An electronic device comprising a processor, a memory and a computer program stored in the memory, characterized in that the processor, when executing the computer program, implements the method of any one of claims 1 to 7 or constitutes the system of claim 8 or 9.

Description

Child emotion intelligent guiding method and system based on emotion instance objectification externalization Technical Field The invention relates to the technical field of artificial intelligence and man-machine interaction, in particular to a child emotion intelligent guiding method and system based on emotion instance objectification externalization. Background With the rapid development of Large Language Models (LLM) and intelligent terminal technologies, AI interactive systems with emotion computing capability are becoming a research and industrialization hotspot. However, the prior art has three classes of system-level drawbacks of technical nature, not just application-level optimization problems, when dealing with the negative emotional burst scenario of a young child. And (one) cognitive overload problems caused by emotion label binding. The existing AI emotion interaction system commonly adopts a 'direct semantic co-emotion' mode, and the underlying data processing logic is that emotion recognition results (such as 'anger') are directly added into the conversation context of the target user as attribute tags and output in a second person sentence pattern (such as 'you are now very anger'). From a computer data structure perspective, this approach creates a strong reference relationship between emotional features and user ontology identifiers at the semantic layer. Existing researches in the field of cognitive neuroscience show that when emotion tags are directly bound with self-identification, the regulating function of the prefrontal cortex of the brain is inhibited, and amygdala is continuously activated, so that cognitive load overload is caused. For children under 5 years old, the prefrontal cortex development is not complete, and the negative effect is more remarkable. At the engineering level, such strong citation binding also results in large language models tending to generate rumination text (emotion analysis, reason discussion, instruction advice) around the emotion tag in multiple conversations, further exacerbating cognitive load, forming a technical emotion strengthening loop rather than a guiding loop. And (II) lack effective closed loop control over large language model outputs. Large language models are probabilistic text generation systems in nature, with uncertainty in their output. In emotional intervention scenarios, existing systems lack a mechanism to technically force suspension and redirection once the model output deviates from a preset guide path (e.g., generating misleading content, teaching text that triggers cognitive rumination). Studies have shown that relying solely on Reinforcement Learning Human Feedback (RLHF) fine tuning and generic security alignment is not sufficient to guarantee output boundaries for models in certain scenarios. In this highly sensitive vertical scenario of child emotion guiding, a control protocol with more engineering certainty than the Prompt word (promt) constraint is needed. And (III) lack of quantitative homing assessment problems based on physiological characteristics. Feedback assessment by existing emotion computing systems typically relies on speech emotion recognition or facial expression analysis, and both of these types of signals are noisier and less reliable during a child emotion outbreak. Heart Rate Variability (HRV) is an objective indicator of autonomic nervous system activity and can more stably reflect physiological changes in emotional arousal. While HRV has found application in adult stress assessment, in the area of childhood emotion guidance, the prior art has not systematically incorporated it into a real-time closed-loop control function. The existing children emotion computing system depends on external signals, and lacks a quantitative homing evaluation mechanism based on objective indexes of an autonomic nervous system such as HRV and the like, so that an intervention strategy (such as regulating the rhythm period of sensory stimulation) cannot be dynamically adjusted according to the real-time physiological level of children. The "open loop" or "weak feedback" mode is difficult to cope with individual variability and dynamic changes in the case of a child emotion outbreak, and is a technical problem to be solved in the field. In summary, in the prior art, technical blanks exist in three dimensions of an emotion tag unbinding mechanism, an LLM output forced constraint protocol and a physiological characteristic closed loop parameter adjustment, and the invention provides a solution to the three specific technical problems. Disclosure of Invention Problems to be solved by the invention The invention aims to solve the following three specific technical problems: The technical problem is how to realize unbinding and isolating of emotion examples and user body identifiers in a data structure level, so that strong reference binding of emotion labels and user identities in large language model output is cut off from a