Digital human character quantitative evaluation method based on large model

Abstract

The invention relates to a digital human character personality quantitative evaluation method based on a large model. The method comprises the following steps: constructing a digital human large model; performing psychological interpretation on the digital human roles in the digital human big model to obtain psychological interpretation data; performing fine adjustment on the open source large model based on psychological interpretation data to obtain a psychological expert large model; according to a character setting system of the digital person, a targeted character question set is manually designed, so that the digital person roles in the digital person big model are replied according to the character question set, and a character reply result is output; using the psychological expert big model to generate a personal question set for character combinations and personal questions with different characteristics, so that digital human roles in the digital human big model reply according to the personal question set and output a personal reply result; the character answer result and the person answer result are scored, and the score is higher when the answer is closer to a standard answer. According to the invention, the character characteristics of the digital person and the maintenance of the person settings of the digital person in the dialogue can be automatically and quantitatively evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of digital human evaluation, and in particular to a method for quantitatively evaluating the personality and setting of a digital human based on a large model. Background Art

[0002] The current evaluation of digital human personality and character is mainly based on manual subjective judgment. On the one hand, the evaluation requires the evaluator to have a high level of psychology-related knowledge, and the evaluation process is highly subjective and lacks consistency; on the other hand, manual evaluation is inefficient.

[0003] At the same time, a small number of evaluation methods based on small models lack comprehensive analysis capabilities based on large-scale data and deep learning technology support, and it is difficult to effectively capture the personality characteristics and deviations of digital humans. In addition, the industry lacks a unified quantitative indicator system, and the evaluation results are difficult to compare and reproduce. They cannot provide clear guidance for the optimal design of digital humans, and their applicability is poor.

[0004] Therefore, it is necessary to provide a quantitative evaluation method for digital human personality based on a large model, which can automatically and quantitatively evaluate the personality characteristics of digital humans and the maintenance of digital humans' personality in conversations. Summary of the invention

[0005] The purpose of the present invention is to provide a method for quantitatively evaluating the personality and setting of a digital human based on a large model, which can automatically and quantitatively evaluate the personality characteristics of a digital human and the maintenance of the digital human's setting during a conversation.

[0006] In order to solve the problems existing in the prior art, the present invention provides a method for quantitatively evaluating the personality and setting of a digital person based on a large model, comprising the following steps:

[0007] Build a digital human model;

[0008] Conduct psychological interpretation on the digital human characters in the digital human large model to obtain psychological interpretation data;

[0009] Fine-tuning the open source big model based on the psychological interpretation data to obtain the psychological expert big model; fine-tuning refers to the process of further training the pre-trained model to adapt it to a specific task or data set. Fine-tuning enables the model to perform specific tasks better by using a smaller new data set for training based on the pre-trained model, without the need to train the entire model from scratch. This method not only saves time and computing resources, but also effectively improves the performance of the model on specific tasks.

[0010] According to the digital human personality setting system, a targeted personality question set is manually designed, so that the digital human character in the digital human large model can answer the personality question set and output the personality answer result;

[0011] Use the psychological expert model to generate a set of personality questions for different personality combinations and personality settings, so that the digital human character in the digital human model can answer the personality question set and output the personality answer result;

[0012] The personality and character answer results will be scored, and the closer they are to the standard answer, the higher the score will be.

[0013] Optionally, in the method for quantitatively evaluating the personality and setting of a digital human based on a macro-model, the method for constructing a macro-model of a digital human is as follows:

[0014] Define a set of role descriptions, wherein the set includes multiple attributes, each attribute includes multiple values, and the attributes include name, age, occupation, personality traits, background story, and special skills;

[0015] Define each attribute as a seed dataset;

[0016] Feed various sub-datasets into the big model;

[0017] The large model randomly selects a value from the seed data set to form a set of initial character settings;

[0018] Form a digital human character based on the initial character settings;

[0019] By analogy, multiple digital human characters are constructed to form a large digital human model.

[0020] Optionally, in the quantitative assessment method for digital human personality based on the big model, the personality question set is set around five dimensions: openness, conscientiousness, extroversion, agreeableness, and emotional stability.

[0021] Optionally, in the method for quantitatively evaluating the personality and setting of a digital human based on a large model, the personality question set is set as follows:

[0022] Design a prompt that includes instructions, personality test questions, and format requirements for digital human output.

[0023] Optionally, in the method for quantitatively evaluating the personality and setting of a digital human based on a large model, the setting problem set includes self-cognition problems, chat partner cognition problems, memory problems, personality problems and characteristic setting problems.

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] (1) Improved objectivity: This aspect uses a fine-tuned large model of psychological experts to replace manual evaluation, which significantly reduces the subjectivity in the evaluation and ensures the scientificity and consistency of the evaluation results.

[0026] (2) Improved accuracy: Through deep learning model analysis based on large-scale data, it is possible to fully capture the personality traits of digital humans and deviations from the human setting during conversations, making the evaluation results more accurate.

[0027] (3) Enhanced efficiency. The present invention realizes a high degree of automation in the evaluation process and can complete the personality and character evaluation of a large number of digital people in a short period of time, greatly improving the evaluation efficiency.

[0028] (4) Wide applicability: The present invention is applicable to a variety of personality setting systems (such as the Big Five personality, MBTI, etc.) and different application scenarios (such as virtual anchors, smart assistants, game characters, etc.), and has high flexibility and scalability.

[0029] (5) Optimization support: Through quantitative evaluation results, this aspect can provide clear guidance for the personalized design, optimization and adjustment of digital humans, and the improvement of intelligent interactive performance, thus promoting the innovation and development of digital human technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flowchart of an evaluation method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The specific implementation of the present invention will be described in more detail below in conjunction with the schematic diagram. The advantages and features of the present invention will become clearer based on the following description. It should be noted that the drawings are all in a very simplified form and are not in exact proportions, and are only used to facilitate and clearly assist in explaining the purpose of the embodiments of the present invention.

[0032] Hereinafter, if the method described herein includes a series of steps, the order in which the steps are presented herein is not necessarily the only order in which the steps may be performed, and some of the steps described may be omitted and / or some other steps not described herein may be added to the method.

[0033] At present, the evaluation of digital human personality and personality is mainly based on manual subjective judgment. On the one hand, the evaluation requires the evaluator to have a high level of psychology knowledge, and the evaluation process is highly subjective and lacks consistency; on the other hand, manual evaluation is inefficient. At the same time, a small number of evaluation methods based on small models lack comprehensive analysis capabilities based on large-scale data and deep learning technology support, and it is difficult to effectively capture the personality characteristics and personality deviations of digital humans. In addition, there is a lack of a unified quantitative indicator system in the industry, and the evaluation results are difficult to compare and reproduce, and cannot provide clear guidance for the optimal design of digital humans, and the applicability is poor.

[0034] In order to solve the problems existing in the prior art, the present invention provides a method for quantitatively evaluating the personality and setting of a digital person based on a large model, such as Figure 1The quantitative evaluation method comprises the following steps:

[0035] S1: Build a digital human model as follows:

[0036] S11: Define a set of role descriptions, which includes multiple attributes, each of which includes multiple values, which are generated by a large model or code script. For example, the attribute of name includes values ​​such as Zhang San, Li Si, Wang Wu, etc.

[0037] Preferably, the attributes include name, age, occupation, personality traits, background story, and special skills;

[0038] S12: define each attribute as a seed dataset;

[0039] S13: Input various sub-datasets into a large model, i.e., a large language model;

[0040] S14: the large model randomly selects a value from the seed data set to form a set of initial role settings;

[0041] S15: forming a digital human character according to the initial character setting;

[0042] S16: By analogy, multiple digital human characters are constructed to form a large digital human model.

[0043] S2: Conduct psychological interpretation of the digital human characters in the digital human big model and obtain psychological interpretation data to ensure the scientificity and authority of the data generated by the psychological expert big model.

[0044] S3: Based on the psychological interpretation data, the open source big model is fine-tuned to obtain the psychological expert big model. The psychological expert big model of the present invention has the ability to master both psychological knowledge and character setting knowledge, and it can deeply understand personality traits and their psychological basis. At the same time, it can identify the stability and consistency of the digital human personality in different scenarios, and provide technical support for the subsequent quantitative evaluation of the digital human personality traits and personality.

[0045] S4: Psychological test: According to the digital human personality setting system, a targeted personality question set is manually designed. The personality question set can flexibly increase dimensions or adjust question content according to needs. The digital human character in the digital human model answers according to the personality question set and outputs the personality answer result;

[0046] Preferably, the personality question set is set around the five dimensions of openness, conscientiousness, extraversion, agreeableness, and emotional stability.

[0047] The method of setting up a personality question set is as follows: design a prompt, which contains instructions, personality test questions, and format requirements for digital human output.

[0048] S5: Dialogue test: Use the psychological expert model to generate a set of personality questions for different personality combinations and personality settings, so that the digital human character in the digital human model can answer the personality question set and output the personality answer result;

[0049] The character setting question set includes self-recognition questions, chat partner recognition questions, memory questions, personality questions, and special character setting questions. The character setting question set is different from the personality question set. The content of the character setting question set revolves around scenes in daily life, such as "What would you do if you encounter a traffic jam?", "Would you get angry if you encountered a traffic jam?", etc., and the digital human is asked to answer the questions in the generated character setting question set.

[0050] S6: Score the personality response results and the persona response results. The closer to the standard answer, the higher the score. Specifically, use data analysis methods to calculate the score of each personality response result, and judge whether the score is consistent with the initial personality setting of the digital person (i.e., the standard answer). The psychological expert model scores the persona response results. The more consistent the answer is with the personality and persona setting (i.e., the standard answer), the higher the score. In this way, the stability and consistency of the persona of the digital person in the conversation is further evaluated.

[0051] In summary, compared with the prior art, the present invention has the following advantages:

[0052] (1) Improved objectivity: This aspect uses a fine-tuned large model of psychological experts to replace manual evaluation, which significantly reduces the subjectivity in the evaluation and ensures the scientificity and consistency of the evaluation results.

[0053] (2) Improved accuracy: Through deep learning model analysis based on large-scale data, it is possible to fully capture the personality traits of digital humans and deviations from the human setting during conversations, making the evaluation results more accurate.

[0054] (3) Enhanced efficiency. The present invention realizes a high degree of automation in the evaluation process and can complete the personality and character evaluation of a large number of digital people in a short period of time, greatly improving the evaluation efficiency.

[0055] (4) Wide applicability: The present invention is applicable to a variety of personality setting systems (such as the Big Five personality, MBTI, etc.) and different application scenarios (such as virtual anchors, smart assistants, game characters, etc.), and has high flexibility and scalability.

[0056] (5) Optimization support: Through quantitative evaluation results, this aspect can provide clear guidance for the personalized design, optimization and adjustment of digital humans, and the improvement of intelligent interactive performance, thus promoting the innovation and development of digital human technology.

[0057] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A quantitative evaluation method for digital human personality based on a large model, characterized by: The following steps are involved: Build a digital human model; Conduct psychological interpretation on the digital human characters in the digital human large model to obtain psychological interpretation data; Fine-tune the open source big model based on the psychological interpretation data to obtain the psychological expert big model; According to the digital human personality setting system, a targeted personality question set is manually designed, so that the digital human character in the digital human large model can answer the personality question set and output the personality answer result; Use the psychological expert model to generate a set of personality questions for different personality combinations and personality settings, so that the digital human character in the digital human model can answer the personality question set and output the personality answer result; The personality and character answer results will be scored, and the closer they are to the standard answer, the higher the score will be.

2. The method for quantitatively evaluating the personality and setting of a digital human based on a large model as claimed in claim 1, characterized in that: The way to build a digital human model is as follows: Define a set of role descriptions, wherein the set includes multiple attributes, each attribute includes multiple values, and the attributes include name, age, occupation, personality traits, background story, and special skills; Define each attribute as a seed dataset; Feed various sub-datasets into the big model; The large model randomly selects a value from the seed data set to form a set of initial character settings; Form a digital human character based on the initial character settings; By analogy, multiple digital human characters are constructed to form a large digital human model.

3. The method for quantitatively evaluating the personality and character of a digital person based on a large model as claimed in claim 1, characterized in that: The personality question set is structured around five dimensions: openness, conscientiousness, extraversion, agreeableness, and emotional stability.

4. The method for quantitatively evaluating the personality and setting of a digital human based on a large model as claimed in claim 3, characterized in that: Here's how to set up a personality question set: Design a prompt that includes instructions, personality test questions, and format requirements for digital human output.

5. The method for quantitatively evaluating the personality and character of a digital human based on a large model as claimed in claim 1, characterized in that: The character setting question set includes self-perception questions, chat partner perception questions, memory questions, personality questions, and special character setting questions.

Images