Search

CN-122025025-A - Method, system and computing device for generating future scene for emotional intervention

CN122025025ACN 122025025 ACN122025025 ACN 122025025ACN-122025025-A

Abstract

The invention discloses a method for generating future scenes for emotion intervention, which comprises the steps of obtaining user related data, utilizing at least one large language model to extract a plurality of memory elements from the user related data, determining generation preference parameters according to a preset mapping relation or algorithm model, utilizing the at least one large language model to generate one or more future scenes according to the generation preference parameters and the memory elements, wherein the future scenes comprise one or more of a text scene, a picture scene, an audio scene, a video scene, a virtual reality scene and an augmented reality scene, performing quality evaluation on the one or more future scenes, determining at least one target future scene according to a quality evaluation result, sending the at least one target future scene to a client for displaying, interfering the emotion of the user, and optimizing the preset mapping relation or algorithm model so as to optimize the generation preference parameters. The invention can generate the individualized future scene with target sense, meaning sense and positive and reasonable sense related to the past experience of the user, thereby helping the user to improve the control sense and the safety sense of the future and immediately and effectively improving the positive emotion of the user.

Inventors

  • ZHANG LERAN
  • DONG YAN

Assignees

  • 北京乐象筑梦互动科技有限公司

Dates

Publication Date
20260512
Application Date
20260123

Claims (17)

  1. 1. A method of generating a future scene for emotional intervention, comprising: Acquiring user-related data, wherein the user-related data comprises one or more of user instant input content, user historical behavior data and user multidimensional image data; Extracting a plurality of memory elements from the user-related data using at least one large language model; Determining and generating preference parameters according to a preset mapping relation or algorithm model; generating one or more future scenes according to the generation preference parameters and the memory elements by using at least one large language model, wherein the future scenes comprise one or more of text scenes, picture scenes, audio scenes, video scenes, virtual reality scenes and augmented reality scenes; Performing quality evaluation on the one or more future scenes, determining at least one target future scene according to a quality evaluation result, and sending the at least one target future scene to the client for display so as to intervene in emotion of a user; And optimizing the preset mapping relation or algorithm model so as to optimize the generation preference parameters.
  2. 2. The method of claim 1, wherein optimizing the preset mapping relationship or algorithm model comprises: and receiving user feedback information from a client, and adjusting parameters of the preset mapping relation or algorithm model based on the user feedback information, wherein the user feedback information comprises at least one of positive emotion test scores, implicit behavior characteristic data and physiological sign data.
  3. 3. The method of claim 1 or 2, wherein the generation preference parameters include semantic feature constraints or model calculation intervention factors for intervention in a large language model generation direction.
  4. 4. The method of any of claims 1-3, wherein the generated preference parameters include at least one of a proportioning parameter, a cue word weight parameter, a model attention distribution parameter, a feature space distance parameter; The proportioning parameters are used for indicating the ratio of the first quantity corresponding to the memory elements to the second quantity corresponding to the creative elements in the future scene generated by the large language model; The prompt word weight parameter is used for indicating the weight distributed to the memory element when the large language model generates a future scene; the model attention distribution parameters are used for increasing the attention degree of the large language model to the memory elements when generating future scenes; The feature space distance parameter is used for indicating that the feature space distance between a future scene generated by the large language model and the memory elements does not exceed a distance threshold.
  5. 5. The method of any one of claims 1-4, wherein determining generating preference parameters according to a predetermined mapping relationship or algorithm model comprises determining proportioning parameters according to a predetermined mapping relationship or algorithm model; The preset mapping relation or algorithm model comprises a core regression equation, wherein the core regression equation comprises a first independent variable, a second independent variable and a dependent variable, the first independent variable and the second independent variable respectively represent a first quantity corresponding to memory elements and a second quantity corresponding to creative elements, and the dependent variable represents a positive emotion value; wherein, determining the proportioning parameter according to a preset mapping relation or algorithm model comprises the following steps: Respectively solving a first partial derivative and a second partial derivative of the first independent variable and the second independent variable of the core regression equation, wherein the first partial derivative and the second partial derivative are respectively 0, so as to obtain a partial derivative equation; and solving the core regression equation and the partial derivative equation simultaneously to obtain the proportioning parameter.
  6. 6. The method of claim 4 or 5, wherein generating one or more future scenes from the generation preference parameters and the plurality of memory elements using at least one large language model comprises: And acquiring a first number of memory elements from the plurality of memory elements and generating a second number of creative elements according to the proportioning parameters by using at least one large language model, and generating one or more future scenes based on the first number of memory elements and the second number of creative elements.
  7. 7. The method of any of claims 1-6, wherein generating one or more future scenes from the generation preference parameters and the plurality of memory elements using at least one large language model comprises: Generating a prompt word according to the generated preference parameters and the memory elements; at least one creative element is generated from the hint words using at least one large language model, and one or more future scenes are generated from the plurality of memory elements and the at least one creative element.
  8. 8. The method of any of claims 1-7, wherein the quality assessment results include a composite score for each of the future scenes, performing a quality assessment on the one or more future scenes, determining at least one target future scene based on the quality assessment results, comprising: and determining the comprehensive score of each future scene according to the preset mapping relation or algorithm model and the rationality function by utilizing at least one large language model, and screening at least one target future scene with the comprehensive score reaching a preset score threshold value or highest comprehensive score from the one or more future scenes.
  9. 9. The method of any of claims 1-8, wherein quality assessment of the one or more future scenes, determining at least one target future scene from the quality assessment results, comprises: Determining a comprehensive score of each future scene according to the preset mapping relation or algorithm model and a rationality function by using at least one large language model; Acquiring detected physiological indexes of the user for each future scene; at least one target future scene is determined based on the composite score for each of the future scenes and the user's physiological index for each of the future scenes.
  10. 10. The method of claim 8 or 9, wherein determining a composite score for each of the future scenes according to the preset mapping relationship or algorithm model and a rationality function comprises: For each future scene, determining a positive emotion value of the future scene according to the preset mapping relation or algorithm model; Determining a rationality value of the future scene according to a rationality function; And carrying out weighted summation according to the positive emotion value and the rationality value of the future scene to obtain the comprehensive score of the future scene.
  11. 11. The method of any of claims 8-10, further comprising: Receiving a rationality score for the target future scene from a client; And correcting parameters of the rationality function based on the rationality score.
  12. 12. The method of any of claims 1-11, wherein after sending the at least one target future scene to the client for presentation, further comprising: And generating guiding information based on the at least one target future scene, and sending the guiding information to the client so that the client guides a user to perform dialogue interaction or task simulation in the at least one target future scene based on the guiding information.
  13. 13. The method of any one of claim 1 to 12, wherein, The plurality of memory elements includes a plurality of words in time, place, person, thing, event, adjective, emotion, feeling, action.
  14. 14. The method of any one of claim 1 to 13, wherein, The user-related data includes one or more of text content, picture content, audio content, and video content.
  15. 15. A system for generating a future scene, comprising: a server side, deployed with a generating device adapted to perform the method of any of claims 1-14, the generating device comprising: The acquisition module is suitable for acquiring user-related data, wherein the user-related data comprises one or more of user instant input content, user historical behavior data and user multidimensional image data; an extraction module adapted to extract a plurality of memory elements from the user-related data using at least one large language model; the determining module is suitable for determining and generating preference parameters according to a preset mapping relation or algorithm model; A generation module adapted to generate one or more future scenes according to the generation preference parameters and the plurality of memory elements using at least one large language model, wherein the future scenes comprise one or more of text scenes, picture scenes, audio scenes, video scenes, virtual reality scenes, and augmented reality scenes; The evaluation module is suitable for carrying out quality evaluation on the one or more future scenes, determining at least one target future scene according to a quality evaluation result, and sending the at least one target future scene to the client for display so as to intervene in the emotion of the user; The optimization module is suitable for optimizing the preset mapping relation or algorithm model so as to optimize the generated preference parameters; The client is in communication connection with the server and is suitable for displaying the at least one target future scene and guiding a user to perform dialogue interaction or task simulation in the at least one target future scene so as to intervene in emotion of the user.
  16. 16. A computing device, comprising: at least one processor, and A memory storing program instructions, wherein the program instructions are configured to be adapted to be processed by the at least one processor, the program instructions comprising instructions for processing the method of any of claims 1-14.
  17. 17. A computer program product comprising computer program instructions which, when executed by a processor, implement the method of any of claims 1-14.

Description

Method, system and computing device for generating future scene for emotional intervention Technical Field The invention relates to the technical field of artificial intelligence, in particular to a method for generating a future scene for emotional intervention, a system for generating the future scene and computing equipment. Background The time travel of MTT (MENTAL TIME TRAVEL ) theory is a bi-directional time latitude, and when reviewing the past, we not just know what happens in the past, but can see the scene at the time again through the mind, experience the feeling and emotion at the time. In looking for the future, we construct and simulate scenes in the mind that may occur in the future based on past memory and experience. When we walk around the scene which may happen in the future, corresponding decision judgment is made, and corresponding feeling and emotion are experienced. This helps us to feel more safe, more aggressive, and more desirable. If the MTT ability of an individual is too weak or lost, the individual will be trapped in a 'permanent present' -perhaps with the current confused, and cannot get the feeling of mastery and safety for the future, cannot imagine the positive possible result in the future, suffer from depression, and may travel out of control, always without voluntarily recall the past negative memory, and have a catastrophic preview of the future, suffer from anxiety. To enhance the active emotion of the user, cognitive behavioral class mobile applications such as Moodgym, woebot (early version) and various punch diary APPs have been developed in the prior art. According to prior art schemes, based on CBT theory, it is believed that emotion is not caused directly by an event, but by our opinion (cognition) of an event. The question of the therapist in the consulting room is digitized by the APP, forcing the user to call the prefrontal cortex (rational brain) to examine the fear or anxiety created by the amygdala (emotional brain), thus breaking the spiral of negative thinking. Recording emotion diaries is essentially an "externalization" process, converting chaotic emotions into words or data, and can increase psychological distance and reduce overwhelming feeling of the emotions. The greatest disadvantage of the above solution is poor compliance, digital APP cannot replace supervision and supervision by a human psychological consultant, and users are easily tired and give up after a period of use. Secondly, these APPs all use fixed CBT scripts and logic trees, which cannot solve the complex emotion problem of the user. In addition, if the cognitive ability of the user itself is low, cognitive distortion, recording cannot be completed or negative reinforcement is generated. In the prior art, a scheme is also provided for realizing psychological support of a user by adopting a dialogue type AI and a large model. AI simulates unconditional active concerns (Unconditional Positive Regard) and co-estrus responses through Natural Language Processing (NLP) techniques, simulating the treatment consortium of "consult relationship" which is one of the most critical factors in psychological consultation, and letting users feel "heard" and "admitted", thereby releasing oxytocin and reducing stress levels. While AI guides the user through typing to constantly teach his own experiences. The traumatic or stress events are organized into coherent linguistic stories, which helps the brain integrate fragmented memory, reducing invasive thinking. However, a drawback of this approach is that AI may have illusion problems and even give harmful advice. Meanwhile, due to the limited memory of large models, in long-term chaperones, AI may forget the key details mentioned earlier by the user for several months, resulting in an "emotional connection" instant break. Furthermore, since AI is a probabilistic model, it does not really care about the user, and for sensitive users, the feeling of autism may double against it once it is realized that the counterpart is only making mathematical predictions. In view of this, there is a need for a method of generating future scenes for emotional intervention to solve the problems presented in the above-mentioned solutions. Disclosure of Invention To this end, the present invention provides a method and a system for generating future scenes for emotional intervention to solve or at least alleviate the above-presented problems. According to one aspect of the invention, a method for generating future scenes for emotional intervention is provided, which comprises the steps of acquiring user related data, wherein the user related data comprise one or more of user immediate input content, user historical behavior data and user multidimensional image data, extracting a plurality of memory elements from the user related data by utilizing at least one large language model, determining generation preference parameters according to a preset mapping relation or algori