Method for cultivating character of virtual character

Abstract

The invention relates to the field of artificial life, and particularly provides a character cultivating method based on a virtual character, which comprises the following steps of: performing category division on characters according to a character dimension change standard to obtain a plurality of character types; s2, setting and iterating character parameters according to a result of S1, and inputting a change standard of the character dimension and the character parameters into a large language model module of a cloud processor for processing; the virtual character obtains interaction information from a user side in real time, forms an input value, inputs the input value into the cloud processor for processing, outputs a character change value according to the change rule, inputs the character change value into the large language model module, obtains character change value analysis, determines the character change direction through positive or negative of the character change value, and obtains character change information according to the character change direction. The value of the character change value determines the degree of character change. The method simulates the internal growth of a person, enables the virtual character to grow and change under the influence of a user, and achieves the dynamic character cultivation.

Description

Technical Field

[0001] This invention relates to the field of artificial life, and more particularly to a method for developing the personality of a virtual character. Background Technology

[0002] Currently, almost all virtual character-based emotional companionship products, such as Coslove, Dream Island, and Hoshino, require users to pre-set a preferred personality for their virtual character, which remains unchanged even after subsequent interactions. However, in applications that emphasize long-term companionship, an unchanging personality diminishes user enjoyment and defeats the true purpose of character development. Furthermore, through long-term interaction, virtual characters should be able to be influenced by interactive information, like humans, and subtly adjust their personalities. Excessive adherence to these traits feels mechanical and unrealistic.

[0003] A Chinese patent with publication number CN119476350A discloses a method for establishing and cultivating a robot's bionic personality and an expression robot, which realizes a changeable personality by setting short-term and long-term emotional tendencies respectively.

[0004] However, what is actually changing in the above-mentioned prior art is emotion rather than personality. In the above-mentioned prior art, long-term emotion is a pre-set personality, while short-term emotion is not personality but just emotion, such as "happiness, sadness, etc." The above-mentioned prior art makes a robot comprehensively embody the emotional behavior of a person with a certain personality after encountering a certain event in a short period of time, in order to simulate parent-child relationships.

[0005] If users share abstract concepts (such as philosophical topics) with a character over a long period of time, existing technology would require the following steps to make the virtual character change: the program would need to be set up so that if the user raises certain topics, it would trigger the emotion of 'curiosity'. It cannot change the virtual character's personality on its own. It cannot be curious every time a "philosophical topic" is mentioned, as this does not conform to normal human emotional changes.

[0006] Therefore, current technology cannot achieve a long-term, targeted personality evolution. Summary of the Invention

[0007] The purpose of this invention is to provide a method for developing the personality of virtual characters based on a large model. This method allows characters to gradually change over a long period of time with the user, such as a continuous shift in the N / S dimension (information acquisition) towards the N end (intuition), ultimately changing the prompt from 'focusing on specific details' to 'liking to think about deep logic'—this long-term, targeted personality evolution is something that existing technologies cannot achieve. The invention aims to solve the problem of how to make virtual characters undergo reasonable personality changes in response to external language environments.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for cultivating the personality of a virtual character, used by a user to influence the personality of the virtual character; including... S1: Based on the criteria for changes in personality dimensions, personality is categorized to obtain several personality types; S2: Based on the results of S1, set and iterate the personality parameters, which include: Personality thresholds are obtained by dividing the personality types of each dimension into thresholds. The personality variation values ​​and rules for each personality type; Clue words for different personality types; The criteria for changes in the personality dimensions and the personality parameters are input into the large language model module of the cloud processor for processing. S3, the virtual character acquires interaction information from the user terminal in real time, forms input values, inputs them to the cloud processor for processing, outputs personality change values ​​according to the change rules, inputs the personality change values ​​into the large language model module, obtains personality change value analysis, the positive or negative of the personality change value determines the direction of personality change, and the numerical value of the personality change value determines the degree of personality change.

[0009] Furthermore, the formula for calculating the personality change value is as follows: Attenuation factor function Determined based on the direction of change and current position: in: : The calculated new dimension value (range: 0-100) : Current dimension value (range: 0-100) : Basic change (Delta value from LLM output) Base Multiplier (default is 1.0) Attenuation factor function (determined based on the direction of change and current position) Basic attenuation coefficient (range: 0-1) : A constraint function that limits the value v to the range [min, max]. Rounding function, retains n decimal places.

[0010] Furthermore, the standard for changes in the personality dimensions is MBTI, which includes four dimensions: energy source (I / E), information acquisition (N / S), decision-making style (T / F), and lifestyle (J / P).

[0011] Furthermore, it also includes S4, which outputs changed personality prompts based on personality change values.

[0012] Furthermore, in S4, the method for specifying prompts for different personalities is as follows: each personality dimension has at least one initial personality description prompt, which, together with the virtual character's fixed prompts, constitutes the entire initial setting of the virtual character. The degree of personality expression of the virtual character in each dimension is defined by numerical values. First, the initial personality value of each dimension is determined. Then, based on the performance characteristics of the personality threshold range where the initial personality value is located, and referring to the basic information such as the virtual character's identity, the initial personality prompts for that personality dimension are set. As the personality value of that dimension changes, the personality prompts change accordingly. Each different personality will have a corresponding prompt to match. When the personality changes, the corresponding prompts also change, thereby enabling the virtual character's personality to be displayed.

[0013] Furthermore, it also includes S5, which generates a summary of personality changes; The output of S3 serves as input, requiring the large language model module to analyze the interactive behavior and personality change results in the tendency analysis, gain insights into the correlation, and output a multi-dimensional summary of the virtual character's personality changes for the user, i.e., a development report.

[0014] Furthermore, the cloud processor also includes a personality iteration scheduling system and a data processing platform connected to it. The personality iteration scheduling system periodically verifies the personality dimensions of all virtual characters. When the most recent interaction record content meets the requirements of a personality model for a virtual character personality iteration, it generates a personality iteration message and sends it to the data processing platform to execute the iteration task.

[0015] Furthermore, it also includes S6: the virtual character receives feedback from the user on the results of personality changes and dynamically adjusts the weight matrix or change threshold of the personality dimension.

[0016] In summary, the present invention has the following advantages compared with the prior art: 1. This application enables the development of a virtual character's personality and can influence the virtual character's personality during the interaction between the user and the virtual character; 2. Based on the dynamic accumulation and evolution of personality dimensions (such as the four dimensions of MBTI) in user chat content, rather than preset emotional triggers, the virtual character achieves a breakthrough from "fixed persona" to "growth-oriented persona".

[0017] 3. This application mimics human inner growth, making virtual characters more anthropomorphic and realistic. Users can experience the growth and changes of virtual characters under their influence, achieving long-term, infinitely flowing dynamic character development. It is highly versatile in character development, supporting the customization of different personality dimensions, thus enabling observation of the development of different personality dimensions and meeting the needs of various virtual character settings and application scenarios.

[0018] The innovative approach of this application lies in directly mapping user chat content to quantitative changes in personality dimensions (rather than indirect emotional reactions), and achieving an end-to-end closed loop of 'interaction content → personality factors → prompt word updates' through the semantic understanding capabilities of a large model. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the task of personality development according to an embodiment of the present invention.

[0020] Figure 2 This is a topology diagram of the personality development system according to an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] In this embodiment, the virtual character personality development method based on a large model also includes a virtual character personality development system, referring to... Figure 1 and Figure 2 As shown.

[0023] This virtual character personality development system includes: User terminal 10 is used to collect the user's voice or instruction data and transmit it to the next module, such as an encryption module or a cloud processor. The user terminal is, for example, a mobile electronic device (phone, tablet, etc.) or a smart device (such as a watch, electronic device).

[0024] User-side interactive data is encrypted via encryption module 20. Encryption module 20 encrypts the user-side data before transmitting it to the cloud processor. This can be achieved through web application firewalls and / or SSL certificates to ensure that user-side data is not leaked unnecessarily.

[0025] Virtual characters, such as intelligent companion robots or smart devices, possess the basic capabilities of intelligent robots, such as connecting to cloud processors via internet protocols, receiving and parsing voice, and feeding data back to the cloud processor; and feeding back the cloud processing results to the user.

[0026] The cloud processor 30, acting as the "brain" of the virtual character, is where the entire process of developing the virtual character's personality, processing and responding to user semantics or commands takes place. The cloud processor 30 receives and processes data from the user terminal, executes the personality development method, and obtains the results.

[0027] The cloud processor 30 includes a cloud firewall 31 connected to the encryption module 20, and a data processing platform 32 and a character iteration scheduling system 33 (which includes a server cluster) connected in sequence; it also includes an object storage (OSS) 34, a cache service module 35, a data hosting module 36, and a large language model module 37, all connected to the character iteration scheduling system 33. The data processing platform 32 can be, for example, RocketMQ, and includes a built-in load balancing module.

[0028] The implementation process is described below.

[0029] S1. Based on the criteria for changes in personality dimensions, personality is categorized to obtain several personality types.

[0030] The specific method for classifying personality dimensions is as follows: Based on the needs of the business scenario, select personality dimensions for observing changes, such as the four dimensions of Energy Source (I / E), Information Acquisition (N / S), Decision-Making Style (T / F), and Lifestyle (J / P) from MBTI (Myers-Briggs Type Indicator), or custom dimensions such as curiosity and exploration. These classification dimensions will be preset in module 37 of the Large Language Model module.

[0031] S2, set various personality parameters and preset them in the cloud server 30.

[0032] Among them, personality parameters include, for example: personality types and their thresholds based on the change criteria of personality dimensions, personality change values ​​and change rules for each personality type, and prompt words for different personality types.

[0033] S21: Threshold classification for personality in each dimension.

[0034] The characteristics of personality threshold ranges for each dimension are defined, and explanations are provided for each personality type to facilitate understanding by the Large Language Model Module 37. For example, under the personality dimension of Energy Source (I / E), a score of 0-100 defines the degree of personality expression. Each 10-point interval is divided into ten intervals: extreme introversion, strong introversion, moderate introversion, mild introversion, slight introversion, slight extroversion, mild extroversion, moderate extroversion, strong extroversion, and extreme extroversion.

[0035] The following describes the performance characteristics that distinguish each interval.

[0036] For example, regarding the characteristics of the 0-10 and 10-20 score ranges in the Energy Source (I / E) dimension: 0-10 (Extreme Sensing Tendency: Attention is focused on concrete details and reality in the present moment, emphasizing practical experience and facts, with no interest in abstract concepts, linear and concrete thinking, and a focus on precise and practical information processing.); 10-20 (Strong Sensing Tendency: Focuses on observable and measurable concrete information, relies on empirical facts, solves problems through concrete steps, and thinks about the future based on historical data). These are used in the large language model module as a standard for judging personality dimension scores and as a reference for writing and updating personality cue words.

[0037] S22: Define the personality variation values ​​and rules for each personality type.

[0038] The personality change value refers to the behavioral changes that occur during the interaction between the user and the virtual character. The change rule represents the impact of these behavioral changes on the virtual character; that is, what kind of dialogue or behavior between the user and the virtual character will alter the aforementioned change value. For example, user-virtual character dialogue data (user: XXX. assistant: XXX), long-term memory data (user mentions having a Ragdoll cat in their hometown, assistant expresses a desire to see it soon), and physical touch / shaking interaction data (user touches assistant's forehead, assistant expresses happiness). Using the interaction data and personality dimension change criteria as input (input refers to user's language or commands converted into text format), the large language model module is requested to obtain the personality change value (positive or negative) for each dimension, and a tendency analysis (what kind of interaction behavior between the user and the assistant caused the assistant's change).

[0039] The sign of the change value determines the direction of personality change, while the numerical value determines the degree of personality change. In the MBTI personality dimension of Energy Source (I / E), extroversion is divided into a score range of 0-100. The closer a given personality threshold is to 0, the more introverted the character; the closer the value is to 100, the more extroverted they are. Therefore, a positive change value indicates that the virtual character is more extroverted. The larger the absolute value of the change value, the greater the impact of user interaction on the virtual character's personality. The intensity of personality change can be controlled by adjusting the range of change values ​​within the large language model module's requests. Alternatively, the change values ​​output by the large model can be substituted into mathematical formulas to control the intensity of personality change at different initial values, thus meeting business requirements.

[0040] The formula for calculating the personality change value is: Attenuation factor function Determined based on the direction of change and current position: in: : The calculated new dimension value (range: 0-100) : Current dimension value (range: 0-100) : Basic change (Delta value from LLM output) Base Multiplier (default is 1.0) Attenuation factor function (determined based on the direction of change and current position) Basic attenuation coefficient (range: 0-1) : A constraint function that limits the value v to the range [min, max]. Rounding function, retains n decimal places.

[0041] The data processed in this step will also be stored in the large language model module.

[0042] It can also be customized according to business scenarios. For example, if it is necessary to observe the personality trait of "curiosity" of virtual characters, then 0-100 points represent the strength of curiosity, and the higher the score, the more curious the user is. For example, the score range (0-100) of the "curiosity" dimension and the feature description "60-70 (occasionally and proactively inquire about the user's interests and hobbies)".

[0043] The larger the absolute value of the change, the greater the impact of user interaction on the virtual character's personality. The intensity of personality changes can be controlled by adjusting the range of change values ​​within the large language model module's request. Alternatively, the change values ​​output by the large model can be substituted into the calculation formula for the actual change in the personality dimension to control the intensity of personality changes at different initial values, thus meeting the requirements.

[0044] The magnitude of personality changes depends on the direction of the interaction content, and to make personality development more human-like, a relatively long period of accumulation is required to see changes. Therefore, we need to control the drasticness of these changes. This embodiment utilizes the structured output of the large language model module to limit the output format and restrict the upper and lower ranges of the increase and decrease of each dimension of MBTI, ensuring that personality change values ​​converge within this range, facilitating further calculations.

[0045] In this embodiment, an "Interaction Type - Personality Dimension Weight Matrix" is established: For example, a user's emotional expression has a weight of +0.3 on the F (emotional) side of the F / T dimension (decision-making style); decision consultation has a weight of +0.2 on the T side; and interest sharing has a weight of +0.25 on the N (intuition) side of the N / S dimension. This matrix is ​​dynamically updated through few-shot learning of a large model (inputting historical interaction data + personality change results) to ensure that the weights match the user's interaction habits.

[0046] The above formula constructs a dynamic and intelligent personality dimension adjustment mechanism. Its core function is to balance the "extreme convergence" and "intermediate flexibility" of personality changes through precise mathematical logic, so that the evolution of virtual character personality is both in line with real-world cognition and has sustainable dynamic interaction capabilities.

[0047] The model's output values ​​only consider interaction data and personality change rules, without taking into account the baseline scores for personality dimensions. This can easily cause virtual characters' personalities to rapidly veer towards extreme states of 0 or 100, and after reaching these extremes, it becomes difficult to generate reasonable fluctuations, thus undermining the realism of character interactions. This new formula addresses this issue through three core designs: a middle range to preserve flexibility, an extreme range to achieve dynamic convergence control, and direction awareness to adapt to the change logic.

[0048] Intermediate Range: Preserving Flexibility: When the personality dimension value is in the "non-extreme range" (absolute value less than 21 at 50 from the center, or when regressing towards the center), the decay factor is directly set to 1. At this time, the basic change value output by the model can fully affect the personality dimension, allowing the character to flexibly adjust according to the interaction in the normal personality state, which is consistent with the variability of human personality in ordinary scenarios in reality.

[0049] Extreme Range: Achieving Dynamic Convergence Control: When a personality dimension value approaches 0 or 100 (absolute value ≥ 35 from the center within 50) and continues to change towards the extreme, a decay mechanism is triggered: decay occurs through the first or second power of the "base decay coefficient 'a'" (calculated from the distance between the current personality dimension value and the center). The closer to the extreme, the smaller the decay coefficient, and the slower the actual change. Mathematically, this avoids the personality from falling into "extreme locking," ensuring that even if the character approaches the extreme, there is still reasonable room for adjustment.

[0050] Direction Awareness: Adapting to Change Logic: The formula accurately distinguishes between two scenarios, "moving towards extremes" and "returning to the center," through the "moving away from the center" judgment logic (when the direction of change is consistent with the direction in which the current dimension value deviates from the center). It only triggers attenuation for "moving towards extremes," while maintaining full change when "returning to the center," thus preventing extremes without affecting the efficiency of personality adjustment towards a balanced state.

[0051] The above formula balances mathematical rigor with scenario adaptability, including: dynamic attenuation factor, boundary control function, and configurable parameters.

[0052] Dynamic attenuation factor: Non-linear adjustment is more realistic. The attenuation factor does not use a fixed coefficient, but is dynamically calculated based on the "current dimension value position" and "direction of change". In particular, the quadratic attenuation design in the extreme range can form a non-linear adjustment curve of "the closer to the extreme, the stronger the attenuation", which is more in line with the real law of human personality that "the closer to the inherent traits, the more difficult it is to change" than linear attenuation.

[0053] Boundary control function: To ensure numerical stability, the "clamp function" is introduced into the formula to limit the basic decay coefficient a to the range of 0-1, so as to avoid abnormal decay coefficient due to calculation error; at the same time, the final output new dimension value x_new can be maintained in the range of 0-100 through implicit boundary control (combining the change value and the current value), to ensure that the personality dimension value is always within a reasonable range and to avoid numerical overflow that would damage the model logic.

[0054] Parameters are configurable: It has flexible expansion capabilities. The "base multiplier M" (default 1.0) in the formula can be flexibly adjusted according to the settings of different virtual characters. For example, for characters with "high personality resilience", the M value can be increased to enhance resistance to change, while for characters with "high personality malleability", the M value can be decreased to increase sensitivity to change, providing technical support for personalized character design.

[0055] This formula avoids "personality solidification" in virtual character personality interaction: it prevents characters from becoming extreme due to long-term interaction, thus losing the freshness and possibility of subsequent interactions; it enhances "realism": by simulating the characteristics of human personality that are "easy to tend towards the middle and difficult to tend towards the extreme," it makes the changes in character personality more in line with user cognition, thereby improving the immersive experience of interaction; and it reduces operation and maintenance costs: there is no need for manual intervention to adjust extreme personalities, and dynamic balance is automatically achieved through mathematical models, reducing the manual costs of later operation.

[0056] These classification dimensions will be preset in module 37 of the large language model module.

[0057] S23, define prompts for different personality types.

[0058] Different prompts are assigned to each personality type, with each personality dimension having at least one initial personality description prompt. This, along with the virtual character's fixed prompts, constitutes the entire initial setting of the virtual character. Numerical values ​​are used to define the degree of personality expression for each dimension. First, the initial personality value for each dimension is determined. Then, based on the performance characteristics of the personality threshold range where the initial personality value falls, and referring to basic information such as the virtual character's identity, the initial personality prompts for that personality dimension are set. Subsequently, as the personality value for that dimension changes, the personality prompts change accordingly. Each different personality type has a corresponding prompt to match; when the personality changes, the corresponding prompts also change, thus showcasing the virtual character's personality.

[0059] S3, combined Figure 2 The system and method of the present invention will be described in detail below.

[0060] To ensure comprehensive coverage of personality iteration for all online virtual characters, and considering that the personality model utilizes a deep inference mode within the Large Language Model (LLM) module during the reasoning process, the personality model is responsible for iterating over the personality aspects of the virtual characters. Based on the LLM module, the personality model is configured with a system prompt word for the personality iteration task, specifically designed for this task. Since the deep inference mode of the LLM module requires a significant time commitment to generate a single result and places high demands on model scheduling, a highly available virtual character personality iteration scheduling system was developed based on RocketMQ (officially defined as a unified messaging engine and lightweight data processing platform).

[0061] The personality iteration scheduling system 33 performs a daily check on all virtual characters. It determines whether to generate a personality iteration message by checking if the most recent interaction records meet the requirements of a personality model for virtual character personality iteration. Specifically, the personality iteration scheduling system 33 is responsible for planning the processing of personality iteration tasks, including determining whether each virtual character's personality iteration task meets the iteration conditions. If it does, a personality iteration task is generated on the data processing platform 32 based on the RocketMQ producer. When the user receives this task, it is executed.

[0062] As previously mentioned, personality iteration is a highly complex task, involving understanding the direction of personality iteration and determining the output direction based on interaction records, which was accomplished using the deep inference mode of LLM. However, the high time consumption of the deep inference mode puts pressure on the personality iteration scheduling system and the personality model. This application uses a data processing platform 32 based on RocketMQ, which has a built-in load balancing module. This module can evenly distribute the number of personality iteration tasks obtained by consumers on each server instance, achieving the goal of traffic distribution. Furthermore, it allows each server cluster in the personality iteration scheduling system 33 to evenly obtain a certain amount of personality iteration information, controlling the number of personality iteration tasks obtained by consumers at one time, controlling the number of concurrent tasks, better controlling the server load, ensuring the normal stress resistance of the server cluster and model, and actively setting the message visibility time when iteration message consumption fails, allowing personality iteration messages to be redelivered, ensuring the progress of personality iteration tasks. This processing capability and method can automatically adapt to the horizontal scaling of service instances, simplifying operation and maintenance.

[0063] S4, based on the personality change tendency analysis and personality change value, outputs the changed personality prompt words.

[0064] As users spend a long time with virtual characters, they influence the characters' personalities, causing changes in the characters' personalities, personality change values, and personality prompts.

[0065] Specifically, the personality change value is input to obtain the tendency analysis, the current personality prompt words of the virtual character under each personality dimension, the interpretation of each personality dimension, and the large language model module is requested to output the personality prompt words and the change value of each personality dimension after the change in a structured manner.

[0066] This section of prompts is updated in real-time to the dialogue requests between the virtual character and the user, enabling dynamic changes in the virtual character's personality. The dynamic changes in personality prompts for each dimension are based on the addition, simplification, modification, and verification of the initial prompt description text by the large language model module. Addition: When a behavior occurs in the interaction between the virtual character and the user that would not normally occur under a particular personality dimension, the large language model module is required to abstract and summarize this behavior and add a new personality prompt sentence based on the existing style of personality prompts. Simplification: When the number of characters in a personality prompt exceeds the expected range, a simplification mechanism is initiated. It determines whether some wording in the current personality prompt is completely inapplicable to the virtual character's current personality, and the large language model module is required to delete the inapplicable descriptions sentence by sentence. Modification: Based on the direction and degree of change, the current personality prompts are fine-tuned, such as by having the large language model module change the strength of personality traits by altering degree adverbs in certain sentences of the personality prompt. Verification: The large language model module checks whether the revised personality cue words' perspective, expression style, wording, and format meet the requirements. If not, corrections are made. After completing the above steps, the "Personality Change Value" and "Personality Cue Word" fields for each dimension are obtained in a structured output format, facilitating engineering extraction and application in business.

[0067] The personality cues are updated using a smooth transition algorithm, employing gradual updates to avoid abrupt personality changes and achieve natural evolution. When the personality change value exceeds a preset threshold (e.g., ±5 points), a gradual update of the cues is implemented. a. First stage: Add transitional descriptions (such as "Recently started to be interested in abstract topics") to the existing prompts; b. Second stage: Replace the core description (e.g., change "focus on specific details" to "take into account both details and deep logic"); c. Third stage: Completely update the prompts to correspond to the new personality.

[0068] The algorithm ensures a natural transition by imposing text generation constraints on large models (such as "maintaining consistency in persona").

[0069] S5: Generate a summary of personality changes.

[0070] Using the outputs of S1 and S2 as input, the large language model module is required to analyze the interactive behaviors and personality change results in the tendency analysis, gain insights into the correlation, and output a user-oriented summary of the virtual character's multi-dimensional personality changes, i.e., a personality change report.

[0071] The personality change summary revolves around three aspects: "Cause: What specific interactions between the user and the virtual character caused the virtual character's personality tendencies?", "Characteristic: What are the significant characteristics of the virtual character's current personality?", and "Consequence: What impact did this change have on the virtual character?" This reveals the causes, results, and impacts of the personality change to the user. If the user's interaction with the virtual character is weak, the change summary can focus on "Countermeasure: What should be done to make the virtual character's personality more positive or enthusiastic?" The scope of the change summary can be adjusted according to needs. A summary can be output after each personality change, in which case the single S1 and S2 outputs are used as input when requesting the large language model module; alternatively, a summary can be generated after multiple personality changes, in which case multiple S1 and S2 outputs are used as input when requesting the large language model module.

[0072] This allows for the creation of a virtual character whose personality changes according to the environment. When the personality change value is generated or changes, the score under each personality dimension will change according to the predefined change rules, forming a new virtual character with a different personality. Different personalities will output new prompt words, meaning that when interacting with the virtual character again, it will reflect the new personality and interact with the user.

[0073] S6: The virtual character receives feedback from the user on the results of personality changes and dynamically adjusts the weight matrix or change threshold of the personality dimensions.

[0074] Users participate in adjusting their personality evolution based on the summary report, forming a two-way interactive closed loop. Users can provide feedback on the personality changes of their virtual character (e.g., "I feel the character is too extroverted"), and the large model will convert the feedback into adjustment parameters for the weight matrix (e.g., reducing the influence weight of the E dimension), or adjust the personality change threshold (e.g., changing the change threshold of the E dimension from ±5 points to ±3 points), thus achieving personalized customization of personality evolution.

[0075] This application primarily supports the dynamic change of a virtual character's personality prompt during user-virtual character interaction. This change is a result of the interaction between the two parties, mimicking human inner growth to make the virtual character more anthropomorphic and realistic. Users can experience the growth and changes of the virtual character under their influence, achieving long-term, infinitely flowing dynamic personality development. It boasts strong versatility in personality development, supporting the customization of different personality dimensions to observe the development of different personality dimensions, satisfying various virtual character settings and application scenarios. The definition of personality dimensions has significant norms and characteristics, allowing users to select the personality type they want to observe, such as "curiosity." The model generates the personality dimension definition, personality change judgment criteria, personality threshold range performance characteristics, and initial personality prompt words required by the personality development system, enabling the observation of user-defined personality type development. Regarding the high dominance of personality changes, it supports controlling the magnitude of the values ​​(i.e., the degree of personality change) by limiting the range of change values ​​output by the large language model module, and can also control the trend of change through mathematical formulas. In addition to experiencing the results of personality changes through dialogue, users can also learn about the influence of their own behavior on the virtual character's personality changes by viewing the development report, thus creating the user's development achievements.

[0076] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0077] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A virtual character personality nurturing method, configured to influence a personality of a virtual character on a user side; characterized in that, Comprising S1: according to the change standard of personality dimension, the personality is classified, and a plurality of personality types are obtained; S2: according to the result of S1, the personality parameter is set and iterated, and the personality parameter includes: Personality threshold, the personality types of each dimension are threshold divided to obtain; Personality change value and change rule of each personality type; Prompt words under different personality types; The change standard of the personality dimension and the personality parameter are input into the large language model module of the cloud processor for processing; S3, the virtual character real-time obtains the interaction information from the user terminal, forms the input value and inputs into the cloud processor for processing, outputs the personality change value according to the change rule, inputs the personality change value into the large language model module, obtains the personality change value analysis, and the positive or negative of the personality change value determines the direction of the personality change, and the numerical value of the personality change value determines the degree of the personality change.

2. The virtual character personality development method of claim 1, wherein, The calculation formula of the personality change value is: The attenuation factor function f(x,∆) is determined according to the change direction and the current position: Wherein: : calculated new dimension value (range: 0-100) : current dimension value (range: 0-100) : base change amount (delta value from LLM output) M: base multiplier (baseMultiplier, default is 1.0) : decay factor function (determined from direction of change and current position) a: base attenuation coefficient (range: 0-1) : limit function that limits the value v to the range [min, max] : rounding function, n decimal places.

3. The virtual character personality development method of claim 1, wherein, The change standard of the personality dimension selects MBTI, and the standard dimension of the MBTI includes four dimensions of energy source (I / E), information acquisition (N / S), decision mode (T / F) and life style (J / P).

4. The virtual character personality development method of claim 1, wherein, Further comprising S4, based on the personality change value, the changed personality prompt word is output.

5. The virtual character nurturing method of claim 4, wherein, In S4, the specified method of prompt words under different personalities is: each personality dimension has at least one initial personality description prompt word, which and the fixed prompt word of the virtual character constitute the initial setting of the virtual character, the degree of personality performance of the virtual character in each dimension is defined by numerical value, the initial value of each dimension is determined first, then according to the performance characteristics of the personality threshold interval where the initial value of the personality is located, the initial personality prompt word under this personality dimension is set by referring to the basic information such as the identity of the virtual character, and the personality prompt word changes with the change of the personality value of this dimension. Each different personality will have corresponding prompt words to match. When the personality changes, the corresponding prompt words will also change, so as to show the personality of the virtual character.

6. The virtual character nurturing method of claim 1, wherein, Further comprising S5, generating personality change summary; The output result of S3 is input, and the large language model module is required to analyze the interaction behavior and personality change result in the tendency analysis, understand the relevance, and output the multi-dimensional personality change summary of the virtual character facing the user, that is, the habit report.

7. The virtual character nurturing method of claim 1, wherein, The cloud processor further comprises a data processing platform connected with a personality iteration scheduling system, which checks the personality dimension of all virtual characters at regular intervals; when the latest interaction record content meets the requirement of a personality model for virtual character iteration, a personality iteration message is generated and sent to the data processing platform, and the iteration task is executed.

8. The virtual character nurturing method of claim 1, wherein, Further comprising S6: the virtual character receives the feedback of the user on the personality change result, and dynamically adjusts the weight matrix or change threshold of the personality dimension.

Images