Assessment method and system combining language model and virtual technology

Abstract

The invention discloses an assessment method and system combining a language model and a virtual technology, and the method comprises the following steps: drilling a police and people dialogue process, analyzing the drilling language output of a subject through a first language model, and obtaining a drilling analysis result; obtaining a drill score of the subject based on the drill analysis result; when the drill score of the subject meets a preset score requirement, obtaining an assessment permission; based on the examination permission, combining a virtual technology to construct police and people dialogue scene questions; the subject is assessed based on the police and people dialogue scene questions, and assessment language output is obtained; and analyzing the assessment language output by adopting a second language model to obtain an assessment evaluation report of the subject. According to the method and the system, the language model is combined with the virtual technology, so that the more advanced and efficient assessment method and system are constructed, and the assessment quality and effect are improved.

Description

Technical Field

[0001] The invention belongs to the field of virtual reality technology, and in particular relates to an assessment method and system combining a language model with virtual technology. Background Art

[0002] As the cutting-edge development direction of the new generation of information technology, virtual reality (VR) technology is gradually penetrating into various industries and fields. These technologies provide users with a new interactive experience by simulating real environments or augmenting real-life scenes. With the improvement of computer graphics processing capabilities and the continuous advancement of display technology, the realism and immersion of virtual environments have been significantly improved, making virtual technology show great application potential in education, training, entertainment and other fields.

[0003] In traditional assessment methods, it is often difficult to fully and realistically simulate actual work or learning scenarios due to factors such as venues, equipment, and resources. Virtual technology can break these limitations and provide a highly realistic virtual environment for assessment, thereby more accurately assessing the abilities and qualities of the subjects. In addition, traditional assessment methods often require manual scoring and evaluation, which are highly subjective and inefficient. AI technology can achieve automated evaluation, avoid interference from human factors, and improve the fairness and accuracy of the evaluation. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides an assessment method and system combining a language model with virtual technology, which assesses the comprehensive quality of the subjects and improves the subjects' police work ability by creating a virtual reality scene.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] An assessment method combining a language model and virtual technology includes the following steps:

[0007] Practice the police-civilian dialogue process and use the first language model to analyze the subjects' practice language output to obtain the practice analysis results;

[0008] Based on the drill analysis results, obtaining a subject's drill score;

[0009] When the subject's exercise score meets the preset score requirements, the subject will obtain the assessment permission;

[0010] Based on the aforementioned assessment permission, combined with virtual technology, a scenario topic of police-civilian dialogue is constructed;

[0011] Assessing the subjects based on the police-civilian dialogue scenario topic to obtain assessment language output;

[0012] The second language model is used to analyze the assessment language output to obtain the subject's assessment evaluation report.

[0013] Preferably, the method for obtaining the drill analysis result is:

[0014] Collect and classify daily police work issues and establish a problem database;

[0015] Allowing the subject to conduct a police-civilian dialogue drill based on the question database to obtain the subject's drill language output;

[0016] Using a tree-like long short-term memory network, extracting syntactic structure information outputted by the training language, and obtaining a syntactic structure information label;

[0017] Inputting the syntactic structure information into a first BERT language model to obtain a sentiment classification probability value and keyword information;

[0018] Based on the syntactic structure information label, the sentiment classification probability value and the keyword information, obtaining a loss function of the syntactic structure information;

[0019] Calculating the loss value of the syntactic structure information using a loss function, and optimizing the first BERT language model parameters based on the loss value to obtain an optimal first BERT language model;

[0020] Based on the optimal first BERT language model, the training analysis result is obtained.

[0021] Preferably, the drill analysis results include analysis results of the subject's intention recognition ability, communication clarity analysis results, communication logical coherence analysis results, and regulatory application accuracy analysis results.

[0022] Preferably, the method for constructing the topic of the police-civilian dialogue scenario is:

[0023] Extract test questions based on the test question bank to obtain test questions;

[0024] Creating a three-dimensional model of a scene and a three-dimensional model of a virtual character that meets the test questions;

[0025] Importing the scene three-dimensional model and the virtual character three-dimensional model into a virtual reality development platform, assembling the models, and obtaining a virtual scene of police-civilian dialogue;

[0026] Establishing behavior logic for the three-dimensional model of the virtual character based on the assessment topic, and setting triggers and interaction points;

[0027] The police-civilian dialogue virtual scene and the three-dimensional model of the virtual character after establishing the behavioral logic are deployed to the VR device to complete the construction of the police-civilian dialogue scene topic.

[0028] Preferably, the method of analyzing the test language output using the second language model is:

[0029] Based on the assessment language output, collecting the subject's reaction time and extracting the subject's reaction characteristics;

[0030] Performing syntactic structure analysis and keyword extraction on the test language output based on the second BERT language model to extract the subject's language features;

[0031] Based on the reaction characteristics and the language characteristics, a comprehensive quality score is given to the subject;

[0032] Based on the comprehensive quality score results, the subjects were classified and their police work ability deficiencies were analyzed;

[0033] Based on the grade classification and the police work ability deficiencies, an assessment and evaluation report of the subject is obtained.

[0034] The present invention also provides an assessment system combining a language model and virtual technology, wherein the method is characterized in that it comprises: a drill module, a drill scoring module, an assessment permission module, an assessment module, and an assessment analysis module;

[0035] The drill module is used to drill the police-civilian dialogue process and use the first language model to analyze the test subject's drill language output to obtain a drill analysis result;

[0036] The drill scoring module is used to obtain the subject's drill score based on the drill analysis result;

[0037] The assessment permission module is used to obtain the assessment permission when the subject's exercise score meets the preset score requirement;

[0038] The assessment module is used to construct a police-civilian dialogue scenario topic based on the assessment license and in combination with virtual technology; assess the subject based on the police-civilian dialogue scenario topic to obtain assessment language output;

[0039] The assessment analysis module is used to analyze the assessment language output using a second language model to obtain an assessment evaluation report of the subject.

[0040] Preferably, the drill module includes a drill language output acquisition unit, a syntactic information extraction unit, a syntactic information analysis unit, a loss function calculation unit, and a model optimization unit;

[0041] The language output acquisition unit is used to classify and collect daily police work problems and establish a problem database; enable the subject to conduct a police-civilian dialogue drill based on the problem database to obtain the subject's drill language output;

[0042] The syntactic information extraction unit is used to extract the syntactic structure information output by the training language by using a tree-like long short-term memory network, and obtain the syntactic structure information label;

[0043] The syntactic information analysis unit is used to input the syntactic structure information into a first BERT language model to obtain a sentiment classification probability value and keyword information;

[0044] The loss function calculation unit is used to obtain the loss function of the syntactic structure information based on the syntactic structure information label, the sentiment classification probability value and the keyword information;

[0045] The model optimization unit is used to calculate the loss value of the syntactic structure information using a loss function, and optimize the first BERT language model parameters based on the loss value to obtain an optimal first BERT language model; based on the optimal first BERT language model, obtain the drill analysis result.

[0046] Preferably, the assessment module includes: a test question extraction unit, a three-dimensional model creation unit, a model assembly unit, a behavior logic design unit and a deployment unit;

[0047] The test question extraction unit is used to extract test questions based on the test question bank to obtain test questions;

[0048] The three-dimensional model creation unit is used to create a three-dimensional model of a scene and a three-dimensional model of a virtual character that meets the test questions;

[0049] The model assembly unit is used to import the scene three-dimensional model and the virtual character three-dimensional model into the virtual reality development platform to assemble the models and obtain a virtual scene of police-civilian dialogue;

[0050] The behavior logic design unit is used to establish behavior logic for the virtual character three-dimensional model based on the assessment topic, and set triggers and interaction points;

[0051] The deployment unit is used to deploy the virtual scene of police-civilian dialogue and the three-dimensional model of the virtual character after establishing the behavior logic to the VR device to complete the construction of the police-civilian dialogue scene topic.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows: through virtual reality (VR) technology, the subjects can interact with the virtual environment in real time. This interactive method is more vivid and intuitive than traditional paper-and-pencil tests or oral defense, and helps to more comprehensively evaluate the comprehensive quality of the subjects. Artificial intelligence (AI) technology can automatically collect and analyze data during the assessment process and quickly generate assessment reports, greatly improving the efficiency of the assessment. At the same time, the AI ​​assessment process is not affected by human factors, ensuring the fairness of the assessment. The combination of virtual technology and AI makes the design of assessment content more flexible and diverse. The difficulty, content and form of the assessment can be quickly adjusted according to actual needs to meet the assessment requirements of different levels and needs. The present invention also uses a language model to analyze the drill and assessment language output to objectively evaluate the overall quality of the subjects. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0054] Figure 1 The present invention is a flowchart of an assessment method combining a language model and virtual technology according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Embodiment 1

[0058] like Figure 1 As shown, an assessment method combining a language model and virtual technology includes the following steps:

[0059] S1: Practice the police-civilian dialogue process and use the first language model to analyze the subjects' practice language output to obtain the practice analysis results;

[0060] A further implementation method is that the method for obtaining the drill analysis result is:

[0061] Collect and classify daily police work issues and establish a problem database;

[0062] The subjects are made to conduct police-civilian dialogue drills based on the question database to obtain the subjects' drill language output; the syntactic structure information of the drill language output is extracted using a tree-like long short-term memory network, and a syntactic structure information label is obtained; in this embodiment, the syntactic structure information label includes a hard label and a soft label, the hard label represents the real emotional label of the sentence, and the soft label represents the emotional probability distribution of the sentence. Specifically, the syntactic structure information of the existing data set is learned using a tree-like long short-term memory network to encode a sentence component structure tree; next, the drill language output is parsed into a syntactic tree structure using a Stanford Parser, and then the hidden representation of each node on the syntactic tree is calculated using a tree-like long short-term memory network with frozen parameters and the soft label of each node is obtained to obtain a "text-soft label" sample, so that the training sample (drill language output) has a syntactic structure.

[0063] The syntactic structure information is input into the first BERT language model to obtain the sentiment classification probability value and keyword information; specifically, preset identifiers are added before and after the training samples, and the first BERT language model is input, the output of the language model is encoded, and the Softmax classifier is input to obtain the sentiment classification probability value (hard label) of the training samples.

[0064] Based on the syntactic structure information label (here referred to as the soft label), the sentiment classification probability value (hard label) and the keyword information, the loss function of the syntactic structure information is obtained; in this embodiment, for samples with soft labels, the weighted cross entropy loss is used to calculate the soft label loss item; for samples with hard labels, the weighted cross entropy loss is also used to calculate the hard label loss item; the pairwise sorting loss function is used to calculate the keyword loss item; the preset loss weights are used to weight the soft label loss item, the hard label loss item and the keyword loss item to obtain the final loss function.

[0065] The loss function is used to calculate the loss value of the syntactic structure information, and the parameters of the first BERT language model are optimized based on the loss value to obtain the optimal first BERT language model; based on the final loss function, the total loss value is obtained, and the parameters of the first BERT language model are adjusted through the BP neural network to improve the accuracy of the language model for sentiment classification. In this embodiment, the core of BERT is composed of multiple stacked Transformer encoder layers. Each encoder layer contains the following key components:

[0066] Multi-head self-attention mechanism: allows the model to focus on tokens at different positions when processing a sequence and calculate the attention weights between tokens to capture the dependencies in the input sequence.

[0067] Feedforward Neural Network: The output of the self-attention mechanism is further transformed to extract higher-level features.

[0068] Residual connection and layer normalization: used to improve the training stability and effect of the model, and help alleviate the problems of gradient disappearance and gradient explosion.

[0069] Based on the optimal first BERT language model, the drill analysis results are obtained.

[0070] Example of emotion classification for subjects:

[0071] (1) In an emergency, how do you calm down an emotional reporter who is complaining about a family dispute?

[0072] (2) When faced with a consultant who does not understand the legal procedures, how can you explain the relevant legal provisions clearly and patiently?

[0073] (3) When handling a traffic accident, how can one communicate effectively with both parties involved, collect key information, and maintain order at the scene?

[0074] By analyzing the subjects' practice language output, we can obtain the subjects' emotional categories (positive, negative, etc.) in different practice test questions and whether the key words are used incorrectly.

[0075] A further implementation method is that the drill analysis results include analysis results of the subject's intention recognition ability, communication clarity analysis results, communication logical coherence analysis results, and regulatory application accuracy analysis results.

[0076] S2: Based on the drill analysis results, obtain the subject's drill score; score the subject in each drill analysis category to obtain the drill score.

[0077] S3: When the subject's exercise score meets the preset score requirements, he / she will obtain assessment permission; the conditions for obtaining assessment permission also include that the total exercise time meets the preset exercise time.

[0078] S4: Based on the assessment permission and combined with virtual technology, construct the police-civilian dialogue scenario topic;

[0079] A further implementation method is to use virtual technology to simulate various complex police situations and public reactions in the police-civilian dialogue assessment, so as to more accurately evaluate the subject's response ability. The method of constructing the police-civilian dialogue scenario topic is:

[0080] Extract test questions based on the test question bank to obtain test questions;

[0081] Create 3D scene models and 3D virtual character models that meet the test questions;

[0082] Import the scene 3D model and the virtual character 3D model into the virtual reality development platform, assemble the models, and obtain the virtual scene of police-civilian dialogue;

[0083] Establish behavioral logic for the 3D model of the virtual character based on the assessment questions, and set triggers and interaction points;

[0084] The virtual scene of police-civilian dialogue and the three-dimensional model of the virtual character after establishing the behavioral logic are deployed to the VR device to complete the construction of the police-civilian dialogue scenario topic.

[0085] S5: Assess the subjects based on the police-civilian dialogue scenario questions and obtain the assessment language output;

[0086] S6: Use the second language model to analyze the assessment language output and obtain the subject's assessment evaluation report.

[0087] A further implementation method is that the method of using the second language model to analyze and assess language output is:

[0088] Based on the assessment language output, the subject's reaction time is collected and the subject's reaction characteristics are extracted; in this embodiment, a timing software is used to record the entire process from the subject receiving the test questions to the subject making a response, and a reaction time series is obtained; the reaction time series is cleaned, and the subject's reaction time and its corresponding assessment language output are timestamped and aligned.

[0089] Then, reaction feature extraction is performed. Basic statistics such as the average reaction time, the shortest reaction time, and the longest reaction time of each subject are calculated to obtain the first reaction feature. The distribution of the subjects' reaction speed is analyzed, such as whether it presents a normal distribution or other types of distribution patterns, to obtain the second reaction feature. Analyze whether the reaction time of the subjects when an error occurs is different from the reaction time when the answer is correct, to obtain the third reaction feature. Study the relationship between the complexity of the assessment language output (such as sentence length, vocabulary difficulty, contextual understanding requirements, etc.) and the reaction time of the subjects to obtain the fourth reaction feature. Perform a time series analysis on the above four reaction features to obtain the trend of changes in the reaction time of the subjects between different assessment questions.

[0090] Based on the second BERT language model, the syntactic structure analysis and keyword extraction of the test language output are performed to extract the subject's language features; (the second BERT language model uses the same network architecture as the first BERT language model)

[0091] Combined with the reaction characteristics and language characteristics, the subjects were scored for comprehensive quality; in this embodiment, the reaction characteristics and language characteristics of the geometric subjects were used to evaluate their cognitive load when processing language outputs with different emotional tendencies.

[0092] Based on the comprehensive quality score results, the subjects are divided into grades and their police work ability deficiencies are analyzed; in this embodiment, it also includes exploring the differences in reaction time and reaction characteristics between different subjects, and analyzing the reasons behind them (such as age, gender, language ability, etc.).

[0093] Based on the classification and deficiencies of police work capabilities, we obtain the assessment and evaluation reports of the subjects. We provide feedback to the subjects or relevant personnel so that they can understand their strengths and weaknesses and carry out targeted training or improvement accordingly.

[0094] Embodiment 2

[0095] The present invention also provides an assessment system and method combining a language model and virtual technology, characterized in that it comprises: a drill module, a drill scoring module, an assessment permission module, an assessment module and an assessment analysis module;

[0096] The drill module is used to drill the police-civilian dialogue process and use the first language model to analyze the subject's drill language output to obtain the drill analysis results;

[0097] A drill scoring module, used to obtain a subject's drill score based on the drill analysis results;

[0098] The assessment permission module is used to obtain the assessment permission when the subject's exercise score meets the preset score requirements;

[0099] The assessment module is used to construct police-civilian dialogue scenario topics based on the assessment permission and combined with virtual technology; the subjects are assessed based on the police-civilian dialogue scenario topics to obtain assessment language output;

[0100] The assessment analysis module is used to analyze the assessment language output using a second language model to obtain the subject's assessment evaluation report.

[0101] A further implementation method is that the drill module includes a drill language output acquisition unit, a syntactic information extraction unit, a syntactic information analysis unit, a loss function calculation unit, and a model optimization unit;

[0102] The language output acquisition unit is used to classify and collect daily police work problems and establish a problem database; enable the subjects to conduct police-civilian dialogue drills based on the problem database and obtain the subjects' drill language output;

[0103] A syntactic information extraction unit, used to extract syntactic structure information outputted by the drill language using a tree-like long short-term memory network, and obtain a syntactic structure information label;

[0104] A syntactic information analysis unit, used to input the syntactic structure information into the first BERT language model to obtain a sentiment classification probability value and keyword information;

[0105] A loss function calculation unit, used to obtain a loss function of syntactic structure information based on syntactic structure information labels, sentiment classification probability values, and keyword information;

[0106] The model optimization unit is used to calculate the loss value of the syntactic structure information using the loss function, and optimize the first BERT language model parameters based on the loss value to obtain the optimal first BERT language model; based on the optimal first BERT language model, obtain the drill analysis result.

[0107] A further implementation method is that the assessment module includes: a test question extraction unit, a three-dimensional model creation unit, a model assembly unit, a behavior logic design unit, and a deployment unit;

[0108] A test question extraction unit, used to extract test questions based on the test question bank to obtain test questions;

[0109] A 3D model creation unit, used to create a 3D model of a scene and a 3D model of a virtual character that meets the test questions;

[0110] A model assembly unit is used to import the scene three-dimensional model and the virtual character three-dimensional model into the virtual reality development platform to assemble the models and obtain a virtual scene of police-civilian dialogue;

[0111] The behavior logic design unit is used to establish the behavior logic for the virtual character 3D model based on the assessment questions, and to set triggers and interaction points;

[0112] The deployment unit is used to deploy the virtual scene of police-civilian dialogue and the three-dimensional model of the virtual character after the behavioral logic is established to the VR device, so as to complete the construction of the police-civilian dialogue scene topic.

[0113] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. An assessment method combining language model and virtual technology, characterized in that: The following steps are involved: Practice the police-civilian dialogue process and use the first language model to analyze the subjects' practice language output to obtain the practice analysis results; Based on the drill analysis results, obtaining a subject's drill score; When the subject's exercise score meets the preset score requirements, the subject will obtain the assessment permission; Based on the aforementioned assessment permission, combined with virtual technology, a scenario topic of police-civilian dialogue is constructed; Assessing the subjects based on the police-civilian dialogue scenario topic to obtain assessment language output; The second language model is used to analyze the assessment language output to obtain the subject's assessment evaluation report.

2. The assessment method combining language model and virtual technology according to claim 1, characterized in that: The method for obtaining the drill analysis result is: Collect and classify daily police work issues and establish a problem database; Allowing the subject to conduct a police-civilian dialogue drill based on the question database to obtain the subject's drill language output; Using a tree-like long short-term memory network, extracting syntactic structure information outputted by the training language, and obtaining a syntactic structure information label; Inputting the syntactic structure information into a first BERT language model to obtain a sentiment classification probability value and keyword information; Based on the syntactic structure information label, the sentiment classification probability value and the keyword information, obtaining a loss function of the syntactic structure information; Calculating the loss value of the syntactic structure information using a loss function, and optimizing the first BERT language model parameters based on the loss value to obtain an optimal first BERT language model; Based on the optimal first BERT language model, the training analysis result is obtained.

3. The assessment method combining language model and virtual technology according to claim 2, characterized in that: The drill analysis results include analysis results of the subjects' intention recognition ability, communication clarity analysis results, communication logical coherence analysis results, and regulatory application accuracy analysis results.

4. The assessment method combining language model and virtual technology according to claim 1, characterized in that: The method of constructing the topic of the police-civilian dialogue scenario is: Extract test questions based on the test question bank to obtain test questions; Creating a three-dimensional model of a scene and a three-dimensional model of a virtual character that meets the test questions; Importing the scene three-dimensional model and the virtual character three-dimensional model into a virtual reality development platform, assembling the models, and obtaining a virtual scene of police-civilian dialogue; Establishing behavior logic for the three-dimensional model of the virtual character based on the assessment topic, and setting triggers and interaction points; The police-civilian dialogue virtual scene and the three-dimensional model of the virtual character after establishing the behavioral logic are deployed to the VR device to complete the construction of the police-civilian dialogue scene topic.

5. The assessment method combining language model and virtual technology according to claim 4, characterized in that: The method of analyzing the test language output using the second language model is: Based on the assessment language output, collecting the subject's reaction time and extracting the subject's reaction characteristics; Performing syntactic structure analysis and keyword extraction on the test language output based on the second BERT language model to extract the subject's language features; Based on the reaction characteristics and the language characteristics, a comprehensive quality score is given to the subject; Based on the comprehensive quality score results, the subjects were classified and their police work ability deficiencies were analyzed; Based on the grade classification and the police work ability deficiencies, an assessment and evaluation report of the subject is obtained.

6. An assessment system combining language model and virtual technology, applying the method described in any one of claims 1 to 5, characterized in that: include: Drill module, drill scoring module, assessment permission module, assessment module and assessment analysis module; The drill module is used to drill the police-civilian dialogue process and use the first language model to analyze the test subject's drill language output to obtain a drill analysis result; The drill scoring module is used to obtain the subject's drill score based on the drill analysis result; The assessment permission module is used to obtain the assessment permission when the subject's exercise score meets the preset score requirement; The assessment module is used to construct police-civilian dialogue scenario topics based on the assessment permission and in combination with virtual technology; Assessing the subjects based on the police-civilian dialogue scenario topic to obtain assessment language output; The assessment analysis module is used to analyze the assessment language output using a second language model to obtain an assessment evaluation report of the subject.

7. The assessment system combining language model and virtual technology according to claim 6, characterized in that: The training module includes a training language output acquisition unit, a syntactic information extraction unit, a syntactic information analysis unit, a loss function calculation unit and a model optimization unit; The language output acquisition unit is used to classify and collect daily police work problems and establish a problem database; enable the subject to conduct a police-civilian dialogue drill based on the problem database to obtain the subject's drill language output; The syntactic information extraction unit is used to extract the syntactic structure information output by the training language by using a tree-like long short-term memory network, and obtain the syntactic structure information label; The syntactic information analysis unit is used to input the syntactic structure information into a first BERT language model to obtain a sentiment classification probability value and keyword information; The loss function calculation unit is used to obtain the loss function of the syntactic structure information based on the syntactic structure information label, the sentiment classification probability value and the keyword information; The model optimization unit is used to calculate the loss value of the syntactic structure information using a loss function, and optimize the first BERT language model parameters based on the loss value to obtain an optimal first BERT language model; based on the optimal first BERT language model, obtain the drill analysis result.

8. The assessment system combining language model and virtual technology according to claim 6, characterized in that: The assessment module includes: a test question extraction unit, a three-dimensional model creation unit, a model assembly unit, a behavior logic design unit and a deployment unit; The test question extraction unit is used to extract test questions based on the test question bank to obtain test questions; The three-dimensional model creation unit is used to create a three-dimensional model of a scene and a three-dimensional model of a virtual character that meets the test questions; The model assembly unit is used to import the scene three-dimensional model and the virtual character three-dimensional model into the virtual reality development platform to assemble the models and obtain a virtual scene of police-civilian dialogue; The behavior logic design unit is used to establish behavior logic for the virtual character three-dimensional model based on the assessment topic, and set triggers and interaction points; The deployment unit is used to deploy the virtual scene of police-civilian dialogue and the three-dimensional model of the virtual character after establishing the behavior logic to the VR device to complete the construction of the police-civilian dialogue scene topic.

Images