Full-induction type intelligent digital interaction system based on cloud edge computing driving

Abstract

The invention discloses a full-induction type intelligent digital interaction system based on cloud edge computing driving. The full-induction type intelligent digital interaction system comprises the following steps: step S1, voice input and understanding; s2, environment perception and analysis; s3, generating response content; s4, matching the response voice with the action; s3, converting the response content in the step S3 into response voice, and synchronously generating corresponding action data; the data character image on the display screen is driven to perform corresponding action through the action data; and S5, continuously collecting user interaction data by the system, and continuously optimizing response content and an interaction mode by the system through big data analysis and a machine learning algorithm. According to the method, through the steps of voice input and understanding, environment perception and analysis, personalized learning optimization and the like, the system can understand user requirements, analyze the situation where the user is located and intelligently generate personalized response content, and therefore the interaction experience of the user and the system is improved.

Description

Technical Field

[0001] The present invention belongs to the field of voice interaction technology, and specifically relates to a fully-sensing intelligent digital interaction system driven by cloud-edge computing. Background Art

[0002] With the development of artificial intelligence and the rise of intelligent interaction methods, the combination of digital virtual people with life and entertainment has become a major trend. Based on the smart school uniform project of smart campus, we have derived a fully-sensing intelligent digital human interaction system driven by cloud-edge computers. It aims to be combined with smart school uniforms to escort the smart campus, match each student according to the chip in the smart school uniform, and identify each student's identity information. Create a personalized, anthropomorphic, fully-sensing digital virtual human image.

[0003] In today’s world of digital humans and virtual assistants, most existing systems rely on a single computing model, typically centralized cloud computing or processing power limited to edge devices. This leads to several key issues:

[0004] 1. Limitations of response speed and real-time interaction: Relying on a single cloud computing model may affect user experience due to network latency, while models that rely solely on edge computing are limited by local processing capabilities and cannot fully utilize cloud resources.

[0005] 2. Insufficient environmental adaptability: Most systems lack effective environmental perception capabilities and are unable to make intelligent responses and suggestions based on the current context, which reduces the naturalness and effectiveness of user interactions.

[0006] 3. Single interaction method: Many existing solutions only support limited forms of interaction (such as voice or touch screen) and lack multimodal interaction mechanisms, which limits the user's interaction freedom and experience quality.

[0007] 4. Limited personalized services and learning capabilities: Current digital humans often adopt general response strategies and lack the ability to deeply personalize learning and adapt to user needs. Summary of the invention

[0008] The purpose of the present invention is to provide a fully-sensing intelligent digital interaction system driven by cloud-edge computing to solve the problems of slow response speed, insufficient environmental adaptability and single interaction mode in digital interaction in the prior art proposed in the background technology.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0010] The fully inductive intelligent digital interactive system driven by cloud-edge computing includes the following steps:

[0011] Step S1, voice input and understanding: by collecting the user's voice input, and processing the collected voice data through natural language processing, the user's language is analyzed and the user's needs are understood;

[0012] Step S2, environmental perception and analysis: obtaining surrounding environment information through cameras and sensors; combining the collected environmental information with eye tracking and image recognition algorithms to analyze the user's situation and prepare for providing personalized responses later;

[0013] Step S3, generating a response content; based on the voice data obtained in step S1 and the environmental information obtained in step S2, and based on the system's preset rules or learning history, intelligently generating a response content for the user's query;

[0014] Step S4, matching the answering voice with the action; converting the answering content in step S3 into the answering voice, and synchronously generating corresponding action data; and driving the data character image on the display screen to make corresponding actions through the action data;

[0015] Step S5, personalized learning optimization; the system continuously collects user interaction data. Through big data analysis and machine learning algorithms, the system can continuously optimize the response content and interaction methods to provide more personalized services.

[0016] According to the above technical solution, in step S1, natural language processing includes the following steps:

[0017] Step S101, text preprocessing, cleaning text data;

[0018] Step S102, word embedding is performed again to map the words into a vector space: each word is converted into a vector representation of a fixed length using a pre-trained word vector model, and then feature extraction is performed for subsequent model training;

[0019] Step S103, use the Qwen2 7b model for training, and finally deploy the model to actual applications.

[0020] According to the above technical solution, in step S2, the surrounding environment information is obtained through the camera and the sensor as follows:

[0021] The user's facial image is captured through the camera, the user's facial center coordinates are calculated and mapped to the corresponding screen coordinates, and the digital human's face is driven to face the user's facial center coordinates based on the orientation of this coordinate and the digital human's coordinates.

[0022] According to the above technical solution, firstly, the MediaPipe Face Detection module is used to perform face detection. After the face is detected, the MediaPipe Face Mesh module is used to obtain 468 3D facial feature points, and the (x, y, z) coordinates of each feature point are obtained. Then, the average position of all feature points is calculated to determine the center point of the face.

[0023] After obtaining the coordinates of the center point of the face, it is necessary to map it from the camera coordinate system to the screen coordinate system: With the known camera resolution and screen resolution, the camera coordinates can be accurately mapped to the screen coordinates using the proportional conversion formula:

[0024] sx=(cx*sw) / cw and sy=(cy*sh) / ch

[0025] Where (cx, cy) is the position of the face center in the camera coordinate system, (sw, sh) is the screen resolution, and (sx, sy) is the mapped screen coordinates;

[0026] After being mapped to the screen coordinates, the user's position on the screen is determined based on the coordinates; through the Live2D API, the digital human's facial rotation angle is set to achieve facial orientation control.

[0027] According to the above technical solution, in step S3, the response content generated is specifically:

[0028] Step S301, obtaining the user's intention according to the text features processed in step S1;

[0029] Step S302, the system analyzes the current environment, including the user's historical queries and context information;

[0030] Step S303, the system generates a response to the user's query according to preset rules or historical learning, and finally presents the generated response to the user;

[0031] Specifically, the system first retrieves the preset rule base, which is used to directly map specific queries to preset answer templates or results. If the query does not completely match the preset rules, the system turns to the historical learning model and uses the deep learning algorithm Transformer to use the multi-head self-attention mechanism to capture the dependencies in the input sequence. Then, by stacking multiple self-attention layers, it captures contextual information at different levels and understands the structure and semantics of the sentence. Finally, the model generates a vector representation of the entire input sequence for generating a response.

[0032] The algorithm is used to retrieve relevant information from the knowledge base, and the system generates response content based on this information. Finally, the system presents the generated response content to the user in the form of text and voice to achieve intelligent interaction.

[0033] According to the above technical solution, in step S4, the answering voice and the action matching are specifically:

[0034] First, the system uses the TTS algorithm to generate the corresponding speech by analyzing the voice information, intonation and emotional characteristics of the response content generated in step S3;

[0035] While generating the response voice, the system also needs to generate action data that matches the response content: first, the generated voice content is subjected to semantic and sentiment analysis to extract the emotional information and key semantics in the voice; based on this information, the system uses preset rules or trained models to generate corresponding facial expression and body action data;

[0036] After the motion data is generated, the system uses Live2D technology to map the motion data onto the digital character image.

[0037] According to the above technical solution, step S5, personalized learning optimization is specifically as follows:

[0038] In order to achieve personalized learning optimization of the system, the system needs to continuously collect user interaction data, including user questions, feedback and interaction time; the data is stored in log files or databases; the collected data includes text data and behavioral data;

[0039] The collected data is cleaned to remove noise, duplicate data and invalid data, and data features are extracted, including extracting keywords and emotions from text data, and extracting frequency and duration of use from behavioral data.

[0040] According to the above technical solution, in step S5, the interaction method also includes gestures and touch screen interaction.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The method of the present invention, through the steps of voice input and understanding, environmental perception and analysis, and personalized learning optimization, the system can understand user needs, analyze the user's situation, and intelligently generate personalized response content, thereby improving the interactive experience between the user and the system. The system can intelligently generate response content by integrating natural language processing, image recognition, machine learning and other technologies, and continuously optimize services based on the user's historical data to achieve more intelligent and personalized interactive services. Through the steps of environmental perception and analysis and matching of response voice and action, the system can obtain user environmental information in real time, generate response content, and realize real-time interaction through action data, making the interaction more vivid and rich. Through the personalized learning optimization steps, the system continuously collects and analyzes user interaction data, continuously optimizes the response content and interaction method, provides users with more personalized services, and continuously improves the intelligence level of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the process of the present invention;

[0044] Figure 2 This is a system block diagram of the present invention. DETAILED DESCRIPTION

[0045] 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.

[0046] Embodiment 1

[0047] like Figure 1 As shown in the figure, the fully inductive intelligent digital interactive system driven by cloud-edge computing includes the following steps:

[0048] Step S1, voice input and understanding: by collecting the user's voice input, and processing the collected voice data through natural language processing, the user's language is analyzed and the user's needs are understood;

[0049] Step S2, environmental perception and analysis: obtaining surrounding environment information through cameras and sensors; combining the collected environmental information with eye tracking and image recognition algorithms to analyze the user's situation and prepare for providing personalized responses later;

[0050] Step S3, generating a response content; based on the voice data obtained in step S1 and the environmental information obtained in step S2, and based on the system's preset rules or learning history, intelligently generating a response content for the user's query;

[0051] Step S4, matching the answering voice with the action; converting the answering content in step S3 into the answering voice, and synchronously generating corresponding action data; and driving the data character image on the display screen to make corresponding actions through the action data;

[0052] Step S5, personalized learning optimization; the system continuously collects user interaction data. Through big data analysis and machine learning algorithms, the system can continuously optimize the response content and interaction methods to provide more personalized services.

[0053] In order to achieve personalized learning optimization of the system, the system needs to continuously collect user interaction data, including user questions, feedback, and interaction time. These data can be stored in log files or databases. The collected data includes text data (such as user input and feedback content) and behavioral data (such as usage time and interaction time). The collected data is cleaned to remove noise, duplicate data, and invalid data, and useful features are extracted, for example, keywords and emotions are extracted from text data, and usage frequency and duration are extracted from behavioral data. The system can use the ALS algorithm for model training, which is a matrix decomposition algorithm for recommendation systems. ALS learns user preferences and behavior patterns by alternately optimizing the user feature matrix and the project feature matrix to minimize the error between the predicted score and the actual score. Specifically, ALS represents user interaction data as a user-item matrix, and decomposes it into two low-dimensional matrices through matrix decomposition, representing user features and project features, respectively. Then, by alternately minimizing the loss function of each matrix, the feature matrix is ​​continuously optimized until convergence. This method can effectively process sparse matrices and is suitable for large-scale data sets. By regularly updating and optimizing the model, retraining and adjusting it based on new user data, the system can ensure that the model adapts to changes in user needs and improves the accuracy and robustness of the model. Based on the model output, the system can dynamically adjust the response content to provide more accurate and personalized answers.

[0054] The method of the present invention, through the steps of voice input and understanding, environmental perception and analysis, and personalized learning optimization, the system can understand user needs, analyze the user's situation, and intelligently generate personalized response content, thereby improving the interactive experience between the user and the system. The system can intelligently generate response content by integrating natural language processing, image recognition, machine learning and other technologies, and continuously optimize services based on the user's historical data to achieve more intelligent and personalized interactive services. Through the steps of environmental perception and analysis and matching of response voice and action, the system can obtain user environmental information in real time, generate response content, and realize real-time interaction through action data, making the interaction more vivid and rich. Through the personalized learning optimization steps, the system continuously collects and analyzes user interaction data, continuously optimizes the response content and interaction method, provides users with more personalized services, and continuously improves the intelligence level of the system.

[0055] Furthermore, the core of the system generating response content based on preset rules or historical learning lies in the intelligent matching and generation mechanism.

[0056] The system will first retrieve the preset rule base. These rules are based on business logic, domain knowledge and expert experience, and are used to directly map specific queries to preset answer templates or results (for example, in the field of education, answer templates will be formulated for specific questions: questions related to "What should I do if I encounter bullying in school?" will provide specific solutions and the school's person in charge and contact information; at the same time, it will integrate the industry's professional knowledge and experience to provide standard answers to common questions.). If the query does not completely match the preset rules, the system will turn to the historical learning model, which is trained based on a large amount of historical conversation data. The specific training process is similar to the natural language processing in step S1. The model converts text data into word embedding representations, with each word corresponding to a vector. The deep learning algorithm Transformer uses the multi-head self-attention mechanism Multi-Head Self-Attention to capture dependencies in the input sequence. Then, by stacking multiple self-attention layers, different levels of context information are captured to understand the structure and semantics of the sentence. Each self-attention layer is followed by a feedforward neural network layer to transform the feature representation. Finally, the model generates a vector representation representing the entire input sequence for generating a response.

[0057] The algorithm retrieves relevant information from the knowledge base and synthesizes this information to generate relevant and natural responses. Finally, the system presents these generated responses to users in the form of text and voice, achieving intelligent interaction.

[0058] In this embodiment, a specific implementation is provided.

[0059] like Figure 2 As shown, in the present invention, it aims to implement an interaction system applicable to various scenarios in a smart campus through intelligent digital human technology. Provide an efficient and natural user interaction experience on low-cost edge devices, while ensuring the instant update of content and a personalized learning mechanism. The present invention mainly consists of the following steps:

[0060] Step 1: Voice input and understanding

[0061] Condition: Collect the voice input of users (students, parents or visitors).

[0062] Function: Through natural language processing (NLP) technology, parse the user's language and understand their intentions and needs. The steps of natural language processing include text preprocessing, which cleans, tokenizes and removes stop words from the text data; then word embedding, which maps words to a vector space; subsequently feature extraction, which extracts useful text features; then select an appropriate model and train it, such as a recurrent neural network or Transformer; then evaluate and optimize the model performance; finally deploy the model to actual applications.

[0063] Specifically, text preprocessing includes deleting errors, inconsistencies and irrelevant information in the corrected data. For example, removing noise, correcting spelling mistakes, unifying text formats, etc. Then tokenize the text, splitting the continuous text into meaningful lexical units. For example, "I left a message for my dad" will be split into "I", "gave", "dad", "left a message", "already". At the same time, stop words are removed based on tokenization. These words usually do not help much in understanding the meaning of the sentence, such as "already", "of", "is", "in", etc.

[0064] Specifically, mapping words to a vector space is as follows: Use the pre-trained word vector model bge-large-zh-v1.5 to convert each word into a fixed-length vector representation, and these vectors can capture the semantic relationships between words. Subsequently, feature extraction is carried out to extract meaningful features from the text, such as word frequency, part-of-speech tagging, dependency syntax, etc., for subsequent model (such as using the Qwen2 7b model) training.

[0065] Step 2: Environment perception and analysis

[0066] Condition: Use cameras and sensors to obtain surrounding environment information (such as location, light, etc.).

[0067] Through Mediapipe technology and Live2D technology, real-time face capture and driving the digital human face toward the user can be achieved. First, face detection is performed using the Mediapipe Face Detection module, which is based on a lightweight deep learning model and can achieve real-time performance on mobile devices. After the face is detected, the Mediapipe Face Mesh module is used to obtain 468 3D facial feature points, which can be used to accurately calculate the center point of the face. Usually, the center point of the face is determined by calculating the average position of specific feature points (such as eyes, nose, and mouth), or directly using the tip of the nose or the midpoint between the eyes as the center point. After obtaining the coordinates of the center point of the face, it needs to be mapped from the camera coordinate system to the screen coordinate system. With the known camera resolution and screen resolution, the camera coordinates can be accurately mapped to the screen coordinates using the proportional conversion formula: sx = (cx*sw) / cw and sy = (cy*sh) / ch, where (cx,cy) is the position of the face center in the camera coordinate system, (sw,sh) is the screen resolution, and (sx,sy) is the mapped screen coordinates. After mapping to the screen coordinates, the user's position on the screen can be determined based on these coordinates. In order to make the digital human's face face the user, it is necessary to calculate the rotation angle of the digital human's face so that it faces the calculated screen coordinates. The specific implementation relies on Live2D technology. Through the Live2D API, the digital human's facial rotation angle is set to achieve facial orientation control. In this way, the real-time detection and tracking of the user's facial center point can be achieved, and the digital human's face can be driven to face the user.

[0068] Function: Combines gaze tracking and image recognition algorithms to analyze the user's situation and prepare for subsequent personalized responses. Captures facial images through the front camera of the Android device, calculates the coordinates of the user's face center and maps them to the corresponding screen coordinates, and drives the face of the digital human to face the user's face center coordinates based on the orientation of this coordinate and the coordinates of the digital human.

[0069] Step 3: Generate response content

[0070] Condition: Based on the user intent in step 1 and the environmental analysis results in step 2.

[0071] Function: Intelligently generate responses to user queries based on preset rules or learning history. The process of generating responses is usually divided into several key steps: first, the system needs to understand the user's intention, which involves natural language processing technology; then, the system analyzes the current environment, including the user's historical queries and contextual information; then, the system generates responses to user queries based on preset rules or historical learning, and finally presents the generated responses to the user.

[0072] Step 4: Match the answering voice with the action.

[0073] Condition: Convert the text content generated in step 3 into response speech and generate corresponding digital human action data simultaneously.

[0074] Function: Ensure that the animation of the digital human image (including lip shape, expression and body movements) is synchronized with the response voice, increasing the realism and naturalness of the interaction.

[0075] While generating the response voice, the system also needs to generate action data that matches the response content: first, the voice signal is converted into text form through automatic speech recognition technology ASR; then the converted text is processed by natural language, including word segmentation, part-of-speech tagging, named entity recognition, etc.; the meaning of the text is understood through semantic analysis, and the main information of the sentence is captured, including extracting keywords, identifying subject-verb-object structures, etc. In addition, the emotional category of the sentence is determined by establishing an emotional vocabulary containing emotional words, such as positive, negative or neutral. Based on this information, the system uses preset rules or trained models to generate corresponding facial expressions and body movement data.

[0076] Live2D is a 2D image-based technology that realizes the three-dimensional display of characters on the screen by converting static flat images into three-dimensional models that can be deformed in real time. Its core principle is to use a grid model composed of multiple grids and nodes, as well as perspective transformation and deformation technology to achieve three-dimensionalization, deformation and animation of images. Artists first draw the various parts of the character, including images of different angles and expressions, and then convert these images into three-dimensional models through the Live2D tool. Developers can add various animation effects to these models, such as blinking, smiling, moving the head, etc., and trigger these animations based on user input or program logic. Ultimately, Live2D models can be easily integrated into games, applications or network services to provide users with a more vivid and interactive experience.

[0077] Step 5: Multimodal interactive feedback

[0078] Conditions: In addition to voice interaction, gestures, touch screen and other interaction methods are also supported.

[0079] Function: Provide richer interaction methods to meet the operating habits and scenario requirements of different users.

[0080] Step 6: Personalized learning optimization.

[0081] Condition: The system continuously collects user interaction data, such as query history, feedback evaluation, etc.

[0082] Function: Through big data analysis and machine learning algorithms, the system can continuously optimize the response content and interaction methods to provide more personalized services. The system collects user data and historical interaction records through big data analysis, and combines machine learning algorithms to extract features and train models to predict user preferences and behaviors. Through real-time personalized recommendations and continuous collection of user feedback for model iteration, the system continuously optimizes the response content and interaction methods, thereby providing more personalized and highly satisfactory services.

[0083] The relationship between each step and its function:

[0084] Steps 1 and 2 are performed in parallel, together providing the system with comprehensive information about user intent and environmental context;

[0085] Step 3 generates accurate response content based on the information in the first two steps;

[0086] Step 4: Convert the text response into voice and generate matching digital human actions to provide users with a more natural interactive experience;

[0087] Step 5 provides additional interactive channels to enhance the applicability and convenience of the system;

[0088] Step six optimizes system performance through a continuous learning process to ensure continuous improvement and personalization of services.

[0089] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0090] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A fully inductive intelligent digital interactive system driven by cloud-edge computing, characterized by: The following steps are involved: Step S1, speech input and understanding; By collecting user voice input and processing the collected voice data through natural language processing, the user's language can be analyzed and the user's needs can be understood; Step S2, environmental perception and analysis: obtaining surrounding environmental information through cameras and sensors; combining the collected environmental information with eye tracking and image recognition algorithms to analyze the user's situation and prepare for subsequent personalized responses; Step S3, generating a response content; based on the voice data obtained in step S1 and the environmental information obtained in step S2, and based on the system's preset rules or learning history, intelligently generating a response content for the user's query; Step S4, matching the answering voice with the action; converting the answering content in step S3 into the answering voice, and synchronously generating corresponding action data; and driving the data character image on the display screen to make corresponding actions through the action data; Step S5, personalized learning optimization; the system continuously collects user interaction data. Through big data analysis and machine learning algorithms, the system can continuously optimize the response content and interaction methods to provide more personalized services.

2. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 1 is characterized in that: In step S1, natural language processing includes the following steps: Step S101, text preprocessing, cleaning text data; Step S102, word embedding is performed again to map the words into a vector space: each word is converted into a vector representation of a fixed length using a pre-trained word vector model, and then feature extraction is performed for subsequent model training; Step S103, use the Qwen2 7b model for training, and finally deploy the model to actual applications.

3. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 1 is characterized in that: In step S2, the surrounding environment information is obtained through the camera and the sensor as follows: The user's facial image is captured through the camera, the user's facial center coordinates are calculated and mapped to the corresponding screen coordinates, and the digital human's face is driven to face the user's facial center coordinates based on the orientation of this coordinate and the digital human's coordinates.

4. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 3 is characterized by: First, use the MediaPipe Face Detection module to perform face detection. After the face is detected, use the MediaPipe Face Mesh module to obtain 468 3D facial feature points, obtain the (x, y, z) coordinates of each feature point, and then calculate the average position of all feature points to determine the center point of the face. After obtaining the coordinates of the center point of the face, it is necessary to map it from the camera coordinate system to the screen coordinate system: With the known camera resolution and screen resolution, the camera coordinates can be accurately mapped to the screen coordinates using the proportional conversion formula: sx=(cx*sw) / cw and sy=(cy*sh) / ch Where (cx, cy) is the position of the face center in the camera coordinate system, (sw, sh) is the screen resolution, and (sx, sy) is the mapped screen coordinates; After being mapped to the screen coordinates, the user's position on the screen is determined based on the coordinates; through the Live2D API, the digital human's facial rotation angle is set to achieve facial orientation control.

5. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 1 is characterized in that: In step S3, the response content generated is specifically: Step S301, obtaining the user's intention according to the text features processed in step S1; Step S302, the system analyzes the current environment, including the user's historical queries and context information; Step S303, the system generates a response to the user's query according to preset rules or historical learning, and finally presents the generated response to the user; Specifically, the system first retrieves the preset rule base, which is used to directly map specific queries to preset answer templates or results. If the query does not completely match the preset rules, the system turns to the historical learning model and uses the deep learning algorithm Transformer to use the multi-head self-attention mechanism to capture the dependencies in the input sequence. Then, by stacking multiple self-attention layers, it captures contextual information at different levels and understands the structure and semantics of the sentence. Finally, the model generates a vector representation of the entire input sequence for generating a response. The algorithm is used to retrieve relevant information from the knowledge base, and the system generates response content based on this information. Finally, the system presents the generated response content to the user in the form of text and voice to achieve intelligent interaction.

6. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 1 is characterized in that: In step S4, the answering voice and the action matching are specifically as follows: First, the system uses the TTS algorithm to generate the corresponding speech by analyzing the voice information, intonation and emotional characteristics of the response content generated in step S3; While generating the response speech, the system also needs to generate action data that matches the response content: first, semantic and sentiment analysis is performed on the generated speech content to extract the emotional information and key semantics in the speech; Based on this information, the system uses preset rules or trained models to generate corresponding facial expression and body movement data; After the motion data is generated, the system uses Live2D technology to map the motion data onto the digital character image.

7. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 1 is characterized in that: Step S5, personalized learning optimization is specifically as follows: In order to achieve personalized learning optimization of the system, the system needs to continuously collect user interaction data, including user questions, feedback, and interaction time; Data is stored via log files or databases; The data collected includes textual data and behavioral data; The collected data is cleaned to remove noise, duplicate data and invalid data, and data features are extracted, including extracting keywords and emotions from text data, and extracting frequency and duration of use from behavioral data.

8. The fully inductive intelligent digital interactive system based on cloud-edge computing drive according to claim 1 is characterized in that: In step S5, the interaction method also includes gestures and touch screen interaction.

Images