VR display system based on AI

Abstract

The invention discloses a VR display system based on AI. The VR display system comprises a VR display layer, an AI interaction layer, a data management layer and a user interface layer. The VR display layer is responsible for generating a 3D environment by using a VR technology; the AI interaction layer comprises an intention definition evaluation module, an inquiry module, a personalized recommendation module and a knowledge base, the intention definition evaluation module evaluates the definition of an intention in a user query statement according to common questions and answers provided by the knowledge base and positive and negative sample statements, and when the definition does not meet the requirement, the personalized recommendation module performs personalized recommendation; the inquiry module actively inquires related questions through a user interaction interface, the personalized recommendation module provides personalized real estate or construction project recommendation for clients according to user demands and preferences mined by the data management layer and data provided by the knowledge base, and the knowledge base is responsible for storing common questions and answers and positive and negative sample statements. The intention definition evaluation module transmits an evaluation result to the inquiry module, and the personalized recommendation module optimizes a recommendation algorithm by using information in a knowledge base.

Description

Technical Field

[0001] The present invention relates to the technical field of VR display, and in particular to a VR display system based on AI. Background Art

[0002] With the rapid development of technology, especially the continuous maturity of virtual reality (VR) and artificial intelligence (AI) technologies, these technologies are gradually penetrating into various industries, and the real estate industry is no exception. The VR display system based on AI can provide home buyers with an immersive experience, enabling them to freely explore every corner of the real estate project in a virtual environment, so as to more intuitively understand information such as the house type, decoration style, and surrounding environment of the real estate. This experience is not only more intuitive but also more convenient, greatly improving the satisfaction of home buyers.

[0003] In the prior art, assume that when a user uses a VR display system based on AI and wants to view the natural light condition of a certain room. Due to the diversity of natural language, the user may express this need in various ways, but the user's intention is to view the effect of the room under natural light. However, the expression input by the user may be vague or unclear, resulting in the system being unable to accurately understand the customer's needs. It may cause cognitive confusion, resulting in the inability to provide the view expected by the user or providing a wrong answer. Therefore, a VR display system based on AI is proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the drawbacks existing in the prior art that the VR display system cannot accurately understand the customer's needs, resulting in cognitive confusion, and thus being unable to provide the view expected by the user or providing a wrong answer, and to propose a VR display system based on AI.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A VR display system based on AI includes a VR display layer, an AI interaction layer, a data management layer, and a user interface layer;

[0007] The VR display layer is responsible for generating a 3D environment using VR technology, where users can freely explore and interact. The VR display layer simulates real estate or building spaces through 3D modeling and rendering technologies to provide an immersive experience;

[0008] The AI interaction layer includes an intent clarity evaluation module, an inquiry module, a personalized recommendation module, and a knowledge base. The intent clarity evaluation module evaluates the clarity of the intent in the user's query statement based on the common questions and answers and positive and negative sample statements provided by the knowledge base. When the clarity does not meet the requirements, the inquiry module actively asks relevant questions through the user interface. The relevant questions should be generated based on the understanding of the user input and the analysis of the context information, aiming to guide the user to provide more information or clarify their needs. The personalized recommendation module provides personalized real estate or construction project recommendations to customers based on the user needs and preferences mined by the data management layer and the data provided by the knowledge base. The knowledge base is responsible for storing common questions and answers and positive and negative sample statements. The intent clarity evaluation module passes the evaluation results to the inquiry module, and the personalized recommendation module uses the information in the knowledge base to optimize the recommendation algorithm;

[0009] The data management layer is responsible for storing and managing data of real estate or construction projects, including 3D models, pictures, videos, text descriptions, etc. At the same time, it mines user needs and preferences to provide data support for the AI interaction layer;

[0010] The user interface layer provides a user interaction interface and interaction methods, including VR device control, information display, operation prompts, etc., to ensure that users can easily get started and enjoy an immersive experience.

[0011] The above technical solution further includes:

[0012] Furthermore, the VR display layer includes a 3D modeling module, a rendering module, a tracking module, and an identification module. The 3D modeling module is responsible for constructing a 3D model of the real estate or construction space. The three-dimensional model accurately reflects the structure, layout, and details of the actual space, including the dimensions of the rooms, the positions of the doors and windows, the placement of the furniture, etc. The rendering module is responsible for converting the 3D model into a visual image and rendering it into the picture seen by the user in real time. The rendering module receives the 3D model data from the 3D modeling module. The tracking module captures the user's head and body movements through the sensors on the VR device and uses the captured data to adjust the perspective and position in the virtual environment.

[0013] Furthermore, the intent clarity evaluation module evaluates the clarity of the intent in the user's query statement based on the common questions and answers and positive and negative sample statements provided by the knowledge base. The specific steps are as follows:

[0014] Semantic Analysis and Entity Recognition: Using natural language processing (NLP) technology, perform semantic analysis and entity recognition on the user's query statement. The semantic analysis is responsible for parsing the syntactic structure and semantic relationships of the user's query statement, identifying components such as the subject, predicate, and object of the sentence. The entity recognition is responsible for identifying entities related to the real estate and construction fields from the user's query statement, such as building names, housing types, etc.;

[0015] Intent Matching and Clarity Evaluation: Match the semantic analysis results of the user's query statement with the questions in the knowledge base, and evaluate the clarity of the user's intent based on the matching degree and the similarity of positive and negative sample statements. Use cosine similarity to calculate the similarity between the user's query statement and the questions in the knowledge base. At the same time, train a classification model based on positive and negative sample statements to evaluate the clarity of the user's intent. The classification model outputs a probability value indicating the clarity degree of the user's query statement intent. According to different weights assigned to the cosine similarity and the results of the classification model, perform weighted summation on the cosine similarity and the results of the classification model to obtain the final clarity evaluation result;

[0016] Result Output and Feedback: Present the evaluation result to the user or system administrator, and optimize the model according to the user's feedback.

[0017] Furthermore, when using cosine similarity to calculate the similarity between the user's query statement and the questions in the knowledge base, convert the user's query statement and the questions in the knowledge base into vector representations through TF-IDF. The similarity calculation formula is:

[0018]

[0019] where, is the feature vector of the user's query statement, is the feature vector of the question in the knowledge base;

[0020] Sort the matching degrees of the user's query statement and all questions in the knowledge base according to the cosine similarity value. The closer the similarity value is to 1, the more similar the two vectors are, that is, the more consistent the intent of the user's query statement and the knowledge base question.

[0021] Furthermore, the specific steps for training a classification model based on positive and negative sample statements:

[0022] Collect positive and negative sample statements: Positive samples: Collect the statements in the user's query statements with clear and definite intents. These statements should be able to directly correspond to a certain question or intent in the knowledge base; Negative samples: Collect the statements in the user's query statements with fuzzy or unclear intents. These statements may contain ambiguities, incomplete information, or cannot directly correspond to the questions in the knowledge base;

[0023] Labeling data: Label the collected positive and negative sample sentences to clarify whether the intention of each sentence is clear, and give corresponding labels (such as 1 for clear and 0 for unclear);

[0024] Feature extraction: convert positive and negative sample sentences into feature vectors;

[0025] Divide the data set: divide the labeled positive and negative sample sentences into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model.

[0026] Training process: The classification model is trained using the training set data. During the training process, the classification model learns how to distinguish between sentences with clear intent and sentences with ambiguous intent based on the feature vectors and labels of the positive and negative sample sentences. The classification model prediction function is:

[0027] Among them, x is the feature vector, θ is the model parameter, and h θ (x) represents the probability that the sentence intent is clear. The goal of the training process is to find the optimal θ so that the loss function of the model on the training set (such as cross entropy loss) is minimized;

[0028] Evaluate model performance: Use the test set data to evaluate the performance of the model, and use the confusion matrix to intuitively understand the performance of the model on positive and negative samples;

[0029] Optimize the model: Based on the evaluation results, adjust the model parameters or select a more appropriate model for training.

[0030] Furthermore, a similarity threshold is set. When the similarity between the user's query statement and a question in the knowledge base exceeds the threshold, the user's intention is considered to be clear and unambiguous; otherwise, it is considered that the clarity does not meet the requirements; when the clarity does not meet the requirements, the inquiry module guides the user to provide more information or clarify his or her needs by actively asking related questions.

[0031] Furthermore, the data management layer mines user needs and preferences, including the following steps:

[0032] Data collection and preprocessing: Collect interaction data between users and the VR display system from multiple channels, including user browsing history, click behavior, dwell time, search keywords, etc., clean and preprocess the interaction data, remove invalid and redundant information, and ensure data quality;

[0033] Feature extraction and selection: Extract features that reflect user needs and preferences from preprocessed data;

[0034] User Profile Construction: Based on the extracted features, a user profile is constructed. A user profile is an abstract representation of user needs and preferences, which can include multiple dimensions such as user age, gender, occupation, income level, housing purchase intention, etc.;

[0035] User Needs and Preferences Mining: Analyze the user profile through a decision tree to mine the potential needs and preferences of users. The generation process of the decision tree is expressed as:

[0036]

[0037] where D is the data set, A is the feature set, A * is the optimal feature, and StopCondition(D) is the stopping condition (such as the number of samples in the node is less than the threshold, the purity reaches the requirement, etc.);

[0038] Result Application and Optimization: Apply the mined user needs and preferences to the AI interaction layer.

[0039] Furthermore, extract features that can reflect user needs and preferences from the preprocessed data, use the random forest algorithm to evaluate the importance of the extracted features, and according to the feature importance evaluation results, screen out the feature points that reflect user needs and preferences. Evaluate the importance of features by calculating the information gain or reduction amount of each feature when splitting the decision tree nodes. For feature A, the information gain formula is: where H(D) is the entropy of the data set D, and H(D v ) is the entropy of the subset after partitioning according to different values of feature A, and |D v | and |D| are the sizes of subset D v and the entire set D, respectively.

[0040] The present invention has the following beneficial effects:

[0041] In the present invention, the intention clarity evaluation module evaluates the clarity of the intention in the user query statement according to the common questions and answers and positive and negative sample statements provided by the knowledge base. When the system detects that the user intention is unclear or ambiguous, the inquiry module actively asks relevant questions to further clarify the user's intention, gradually narrow the scope of the user's needs, and ensure that the operation finally understood and executed is what the user really wants. In this way, the system can more deeply understand the user's needs and provide more appropriate responses and services. Brief Description of the Drawings

[0042] Figure 1 It is a system block diagram of a VR display system based on AI proposed by the present invention. Detailed Embodiments

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0044] Please refer to Figure 1 As shown, the present invention is a VR display system based on AI, including a VR display layer, an AI interaction layer, a data management layer, and a user interface layer;

[0045] The VR display layer is responsible for generating a 3D environment using VR technology, where users can freely explore and interact. The VR display layer simulates real estate or building spaces through 3D modeling and rendering technologies to provide an immersive experience;

[0046] The AI interaction layer includes an intention clarity evaluation module, an inquiry module, a personalized recommendation module, and a knowledge base. The intention clarity evaluation module evaluates the clarity of the intention in the user's query statement based on the common questions and answers and positive and negative sample sentences provided by the knowledge base. When the clarity does not meet the requirements, the inquiry module actively asks relevant questions through the user interface. The personalized recommendation module provides personalized real estate or building project recommendations to customers based on the user needs and preferences mined by the data management layer and the data provided by the knowledge base. The knowledge base is responsible for storing common questions and answers and positive and negative sample sentences. The intention clarity evaluation module transmits the evaluation results to the inquiry module, and the personalized recommendation module uses the information in the knowledge base to optimize the recommendation algorithm;

[0047] The data management layer is responsible for storing and managing data of real estate or building projects, including 3D models, pictures, videos, text descriptions, etc. At the same time, it mines user needs and preferences to provide data support for the AI interaction layer;

[0048] The user interface layer provides user interfaces and interaction methods, including VR device control, information display, operation prompts, etc., to ensure that users can easily get started and enjoy an immersive experience.

[0049] In one embodiment, the VR display layer includes a 3D modeling module, a rendering module, a tracking module, and an identification module. The 3D modeling module is responsible for constructing a 3D model of a property or building space. The three-dimensional model accurately reflects the structure, layout, and details of the actual space, including the dimensions of rooms, the positions of doors and windows, the placement of furniture, etc. The rendering module is responsible for converting the 3D model into a visual image and rendering it in real time into the picture seen by the user. The rendering module receives the 3D model data from the 3D modeling module. The tracking module captures the user's head and body movements through sensors on the VR device and uses the captured data to adjust the perspective and position in the virtual environment.

[0050] In one embodiment, the intention clarity evaluation module evaluates the clarity of the intention in the user's query statement based on the common questions and answers provided by the knowledge base and positive and negative sample statements. The specific steps are as follows:

[0051] Semantic analysis and entity recognition: Using natural language processing (NLP) technology, perform semantic analysis and entity recognition on the user's query statement. The semantic analysis is responsible for parsing the grammatical structure and semantic relationships of the user's query statement, identifying components such as the subject, predicate, and object of the sentence. The entity recognition is responsible for identifying entities related to the real estate and construction fields from the user's query statement, such as property names, housing types, etc.;

[0052] Intention matching and clarity evaluation: Match the semantic analysis results of the user's query statement with the questions in the knowledge base, and evaluate the clarity of the user's intention based on the matching degree and the similarity of positive and negative sample statements. Use cosine similarity to calculate the similarity between the user's query statement and the questions in the knowledge base. At the same time, train a classification model based on positive and negative sample statements to evaluate the clarity of the user's intention. The classification model outputs a probability value indicating the clarity degree of the intention of the user's query statement. Assign different weights to the results of cosine similarity and the classification model, and perform weighted summation on the results of cosine similarity and the classification model according to the weights to obtain the final clarity evaluation result;

[0053] Result output and feedback: Present the evaluation result to the user or system administrator and optimize the model according to the user's feedback;

[0054] Suppose the user enters a query statement: "I want to see the floor plan of the three-bedroom, two-living-room of this property."

[0055] Semantic analysis: Parse that the subject of the sentence is "I", the predicate is "want to see", and the object is "the floor plan of the three-bedroom, two-living-room of this property".

[0056] Entity recognition: Identify the entities as "property" and "the floor plan of the three-bedroom, two-living-room".

[0057] Intention Matching and Clarity Assessment: The user's query statement is matched with the questions in the knowledge base, and it is found that it highly matches the intention of "querying the house type diagram". At the same time, based on the similarity assessment of positive and negative sample statements, it is considered that the user's intention is clear and definite.

[0058] Result Output: The system shows the user the three-bedroom and two-living-room house type diagram of the property and gives the evaluation result: "Your query intention is clear and definite, and the required house type diagram has been shown to you."

[0059] In one embodiment, the cosine similarity is used to calculate the similarity between the user's query statement and the questions in the knowledge base. The user's query statement and the questions in the knowledge base are respectively converted into vector representations through TF-IDF. The similarity calculation formula is:

[0060]

[0061] Where, is the feature vector of the user's query statement, is the feature vector of the question in the knowledge base;

[0062] The matching degrees of the user's query statement with all questions in the knowledge base are sorted according to the cosine similarity value. The closer the similarity value is to 1, the more similar the two vectors are, that is, the more consistent the intention of the user's query statement and the knowledge base question.

[0063] Vector Representation: User Query Statement Vector: User Vector = [I, want, to, see, the, three-bedroom, two-living-room, house type diagram];

[0064] Knowledge Base Question Vector: Question Vector = [Excuse, me, are, there, any, house type diagrams, that, can, be, viewed]. In practice, TF-IDF will be used to convert the text into a high-dimensional vector.

[0065] Calculate Cosine Similarity: The norms of the obtained user vector and question vector are 1 and 1 respectively, and their dot product is 0.6.

[0066] Then the cosine similarity is:

[0067] Matching Degree Sorting and Intention Clarity Assessment: Assuming that the set threshold is 0.5, since the cosine similarity between the user's query statement and the knowledge base question is 0.6, which exceeds the threshold, it is considered that the user's intention is clear and definite, that is, the user wants to view the three-bedroom and two-living-room house type diagram of the property.

[0068] In one embodiment, the specific steps of training a classification model based on positive and negative sample statements:

[0069] Collecting positive and negative sample statements: Positive samples: Collect statements in the user's query with clear and definite intentions, which should be directly corresponding to a certain question or intention in the knowledge base; Negative samples: Collect statements in the user's query with vague or unclear intentions, which may contain ambiguity, incomplete information or cannot be directly corresponding to the questions in the knowledge base;

[0070] Labeling data: Label the collected positive and negative sample statements to clarify whether the intention of each statement is clear and give corresponding labels (such as 1 for clear and 0 for unclear);

[0071] Feature extraction: Convert the positive and negative sample statements into feature vectors;

[0072] Dividing the dataset: Divide the labeled positive and negative sample statements into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model;

[0073] Training process: Use the training set data to train the classification model. During the training process, the classification model learns how to distinguish between statements with clear and vague intentions based on the feature vectors and labels of the positive and negative sample statements. The prediction function of the classification model is:

[0074] where x is the feature vector, θ is the model parameter, and h θ (x) represents the probability that the statement intention is clear. The goal of the training process is to find the optimal θ to minimize the loss function (such as cross-entropy loss) of the model on the training set;

[0075] Evaluating the model performance: Use the test set data to evaluate the performance of the model and intuitively understand the performance of the model on positive and negative samples through the confusion matrix;

[0076] Optimizing the model: According to the evaluation results, adjust the model parameters or select a more suitable model for training;

[0077] Suppose in an AI-based VR display system for real estate and construction, there are the following two user query statements:

[0078] Positive sample: "I want to see the floor plan of the three-bedroom and two-living-room of this property."

[0079] Negative sample: "How about this property?"

[0080] For these two statements, their feature vectors can be extracted respectively and the trained clarity evaluation model can be used for prediction. If the model can accurately judge the positive sample as having a clear intention and the negative sample as having a vague intention, it indicates that the model has good performance.

[0081] In one embodiment, a similarity threshold is set. When the similarity between the user's query statement and a certain question in the knowledge base exceeds this threshold, it is considered that the user's intention is clear and definite; otherwise, it is considered that the clarity does not meet the requirements. When the clarity does not meet the requirements, the inquiry module actively inquires about relevant questions to guide the user to provide more information or clarify their needs.

[0082] Suppose the user's query statement is: "What about the three-bedroom and two-living-room apartment type in this real estate project?" However, the system cannot accurately match relevant questions in the knowledge base, so it is determined that the clarity does not meet the requirements. At this time, the system can actively inquire: "Do you want to know the specific situation of the three-bedroom and two-living-room apartment type in this real estate project? Such as area, orientation, or decoration standard, etc." Or, the system can also ask more broadly: "Do you want to know the apartment type, price, or location information of the real estate project? Or information in other aspects?";

[0083] The system adjusts the subsequent processing strategy according to the user's answer. If the user provides more information or clarifies their needs, the system can recalculate the similarity and try to match questions in the knowledge base; if the user still does not provide enough information, the system can continue to inquire or provide other assistance.

[0084] In one embodiment, the data management layer mines user needs and preferences, including the following steps:

[0085] Data collection and preprocessing: Collect interaction data between users and the VR display system from multiple channels, including user browsing records, click behaviors, stay times, search keywords, etc. Clean and preprocess the interaction data to remove invalid and redundant information and ensure data quality;

[0086] Feature extraction and selection: Extract features from the preprocessed data that can reflect user needs and preferences;

[0087] User portrait construction: Based on the extracted features, construct a user portrait. The user portrait is an abstract representation of user needs and preferences and can include multiple dimensions such as the user's age, gender, occupation, income level, and housing purchase intention;

[0088] Mining of user needs and preferences: Analyze the user portrait through a decision tree to mine the user's potential needs and preferences. The generation process of the decision tree is expressed as:

[0089]

[0090] where D is the data set, A is the feature set, A * is the optimal feature, and StopCondition(D) is the stop condition (such as the number of samples in the node is less than the threshold, the purity reaches the requirement, etc.);

[0091] Result Application and Optimization: Apply the mined user requirements and preferences to the AI interaction layer.

[0092] In one embodiment, extract features from the preprocessed data that can reflect user requirements and preferences, use the random forest algorithm to evaluate the importance of the extracted features, and according to the feature importance evaluation results, screen out the feature points that reflect user requirements and preferences. Evaluate the importance of features by calculating the information gain or reduction amount when each feature splits at the decision tree node. For feature A, the information gain formula is: where H(D) is the entropy of dataset D, and H(D v ) is the entropy of the subset after partitioning according to different values of feature A, and |D v | and |D| are the sizes of subset D v and the entire set D, respectively.

[0093] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based VR display system, characterized in that, It includes a VR display layer, an AI interaction layer, a data management layer, and a user interface layer; The VR display layer is responsible for generating a 3D environment using VR technology. The VR display layer simulates a real estate or building space through 3D modeling and rendering techniques; The AI interaction layer includes an intention clarity evaluation module, an inquiry module, a personalized recommendation module, and a knowledge base. The intention clarity evaluation module evaluates the clarity of the intention in the user's query statement based on the common questions and answers provided by the knowledge base and positive and negative sample statements. When the clarity does not meet the requirements, the inquiry module actively asks relevant questions through the user interface. The personalized recommendation module provides personalized real estate or building project recommendations to customers according to the user needs and preferences mined by the data management layer and the data provided by the knowledge base. The knowledge base is responsible for storing common questions and answers as well as positive and negative sample statements. The intention clarity evaluation module passes the evaluation result to the inquiry module, and the personalized recommendation module uses the information in the knowledge base to optimize the recommendation algorithm; The data management layer is responsible for storing and managing data of real estate or building projects. At the same time, it mines user needs and preferences; The user interface layer provides a user interface and interaction methods.

2. The AI-based VR display system according to claim 1, wherein, The VR display layer includes a 3D modeling module, a rendering module, a tracking module, and an identification module. The 3D modeling module is responsible for building a 3D model of the real estate or building space. The rendering module is responsible for converting the 3D model into a visual image and rendering it into the picture seen by the user in real time. The rendering module receives the 3D model data from the 3D modeling module. The tracking module captures the user's head and body movements through sensors on the VR device and uses the captured data to adjust the perspective and position in the virtual environment.

3. An AI-based VR display system according to claim 1, characterized in that The intention clarity evaluation module evaluates the clarity of the intention in the user's query statement based on the common questions and answers provided by the knowledge base and positive and negative sample statements. The specific steps are as follows: Semantic analysis and entity recognition: Using natural language processing technology, perform semantic analysis and entity recognition on the user's query statement. The semantic analysis is responsible for parsing the syntactic structure and semantic relationship of the user's query statement and identifying the components of the sentence. The entity recognition is responsible for identifying the entities related to the real estate and construction fields from the user's query statement; Intention matching and clarity evaluation: Match the semantic analysis result of the user's query statement with the questions in the knowledge base, and evaluate the clarity of the user's intention according to the matching degree and the similarity of positive and negative sample statements. Use cosine similarity to calculate the similarity between the user's query statement and the questions in the knowledge base. At the same time, train a classification model based on positive and negative sample statements to evaluate the clarity of the user's intention. The classification model outputs a probability value indicating the clarity degree of the intention of the user's query statement. According to different weights assigned to the cosine similarity and the result of the classification model, perform weighted summation on the cosine similarity and the result of the classification model to obtain the final clarity evaluation result; Result output and feedback: Present the evaluation result to the user or system administrator and optimize the model according to the user feedback.

4. An AI-based VR display system according to claim 3, wherein, The cosine similarity is used to calculate the similarity between the user's query statement and the questions in the knowledge base. The user's query statement and the questions in the knowledge base are respectively converted into vector representations through TF-IDF. The similarity calculation formula is as follows: Among them, is the feature vector of the user's query statement, is the feature vector of the questions in the knowledge base; Sort the matching degrees of the user's query statement and all questions in the knowledge base according to the cosine similarity value. The closer the similarity value is to 1, the more similar the two vectors are, that is, the more consistent the intention of the user's query statement and the knowledge base question.

5. An AI-based VR display system according to claim 3, wherein, The specific steps for training a classification model based on positive and negative sample statements are as follows: Collect positive and negative sample statements: Positive samples: Collect the statements with clear and definite intentions in the user's query statements; Negative samples: Collect the statements with vague or unclear intentions in the user's query statements; Label the data: Label the collected positive and negative sample statements to clarify whether the intention of each statement is clear and give the corresponding labels; Feature extraction: Convert the positive and negative sample statements into feature vectors; Divide the data set: Divide the labeled positive and negative sample statements into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the performance of the model; Training process: Use the training set data to train the classification model. During the training process, the classification model learns how to distinguish between clearly-intended and ambiguously-intended statements based on the feature vectors and labels of positive and negative sample statements. The prediction function of the classification model is: where x is the feature vector, θ is the model parameter, and h θ (x) represents the probability of clear statement intention; Evaluate the model performance: Use the test set data to evaluate the performance of the model, and intuitively understand the performance of the model on positive and negative samples through the confusion matrix; Optimize the model: According to the evaluation results, adjust the parameters of the model or select a model for training.

6. The AI-based VR display system according to claim 3, wherein Set a similarity threshold. When the similarity between the user's query statement and a certain question in the knowledge base exceeds this threshold, it is considered that the user's intention is clear and definite; otherwise, it is considered that the clarity does not meet the requirements. When the clarity does not meet the requirements, the inquiry module actively asks relevant questions to guide the user to provide more information or clarify their needs.

7. An AI-based VR display system according to claim 1, characterized in that, The data management layer mines the user's needs and preferences, including the following steps: Data collection and preprocessing: Collect the interaction data between the user and the VR display system, and clean and preprocess the interaction data; Feature extraction and selection: Extract the features that can reflect the user's needs and preferences from the preprocessed data; User portrait construction: Construct a user portrait based on the extracted features; User need and preference mining: Analyze the user portrait through a decision tree to mine the user's potential needs and preferences. The generation process of the decision tree is expressed as: where D is the dataset, A is the feature set, and A * is the optimal feature, and StopCondition(D) is the stopping condition; Result application and optimization: Apply the mined user needs and preferences to the AI interaction layer.

8. An AI-based VR display system according to claim 7, characterized in that, Extract the features that can reflect the user's needs and preferences from the preprocessed data, use the random forest algorithm to evaluate the importance of the extracted features, and filter out the feature points that reflect the user's needs and preferences according to the feature importance evaluation results. Evaluate the importance of the features by calculating the information gain or reduction amount when each feature splits the decision tree node. For feature A, the information gain formula is: Among them, H(D) is the entropy of the data set D, and H(D v ) is the entropy of the subset after partitioning according to different values of the feature A. |D v | and |D| are the sizes of the subset D v and the entire set D, respectively.

Images