Intelligent agent generation method, learning machine and electronic equipment

Abstract

The invention provides an intelligent agent generation method, a learning machine and electronic equipment. The generation method comprises the following steps: acquiring user input data and learning behavior data; processing the user input data to obtain a first feature vector, and processing the learning behavior data to obtain a second feature vector; fusing the first feature vector and the second feature vector to obtain a fused feature vector; inputting the fusion feature vector into a generator of a generative adversarial network to obtain a virtual image; and endowing the virtual image with an interaction behavior by adopting a machine learning algorithm according to the user input data and the learning behavior data to obtain an intelligent agent. According to the method, the fusion feature vector is obtained based on the input data and the learning behavior data, the virtual image is generated through the adversarial network generator, the virtual image meeting the emotional requirements of children can be generated in combination with the interests and characters of the children, and the interactive behaviors are generated based on the user input data and the learning behavior data, so that the actual requirements of the children can be better met.

Description

Technical Field

[0001] The present invention relates to the field of children's learning technology, and in particular to a method for generating an intelligent body, a learning machine and an electronic device. Background Art

[0002] In children's learning machines, virtual images are usually set up to interact with children, thereby achieving the purpose of assisting learning. However, in the existing technology, the generation of virtual images mainly has the following problems: 1. Image fixation: Existing systems mostly use preset intelligent body images, such as fixed cartoon characters, which lack flexibility and personalization, and the interactive experience is single: users cannot participate in the creation process of the intelligent body image, and it is difficult to meet children's needs for personalized interactive partners; 2. Lack of emotional connection: The preset image lacks relevance to children's interests and personality traits, making it difficult to establish a deep emotional connection; 3. Disconnection between learning behavior: The intelligent body image is separated from the child's learning behavior data and cannot be dynamically adjusted according to learning performance. Summary of the Invention

[0003] In order to overcome the above technical defects, the present invention provides a method for generating an intelligent agent, a learning machine and an electronic device, which can generate an intelligent agent based on the actual needs of children.

[0004] In order to solve the above problems, the present invention is implemented according to the following technical solutions:

[0005] A method for generating an intelligent agent, comprising the steps of:

[0006] Obtain user input data and learning behavior data;

[0007] Processing user input data to obtain a first eigenvector, and processing learning behavior data to obtain a second eigenvector;

[0008] Fusing the first eigenvector and the second eigenvector to obtain a fused eigenvector;

[0009] The fused feature vector is input into the generator of the generative adversarial network to obtain a virtual image;

[0010] Using machine learning algorithms, interactive behaviors are given to the virtual image based on user input data and learning behavior data to obtain an intelligent entity.

[0011] As a further improvement of the present invention, the step of obtaining user input data and learning behavior data includes:

[0012] According to the user's input, obtain user input data;

[0013] Perform format conversion, cleaning, and noise reduction on user input data;

[0014] Extracting feature information and / or text information from user input data;

[0015] Connect to the learning platform to obtain users' learning behavior data during the learning process;

[0016] Perform data cleaning on learning behavior data;

[0017] Normalize the learning behavior data.

[0018] As a further improvement of the present invention, the steps of processing the user input data to obtain the first feature vector and processing the learning behavior data to obtain the second feature vector include:

[0019] If the user input data is image data, the convolutional neural network encodes the feature information to obtain the first feature vector;

[0020] If the user input data is text digital, the text information is converted into a vector, and the vector is processed using a neural network to obtain the first eigenvector;

[0021] The normalized learning behavior data are spliced ​​to obtain a second eigenvector, or a neural network is used to perform nonlinear transformation on the learning behavior data to extract the second eigenvector.

[0022] As a further improvement of the present invention, the step of fusing the first eigenvector and the second eigenvector to obtain a fused eigenvector includes:

[0023] Calculate the attention weights of the first eigenvector and the second eigenvector respectively;

[0024] According to the attention weight, the first eigenvector and the second eigenvector are weighted and summed to obtain a fused eigenvector.

[0025] As a further improvement of the present invention, after obtaining the fused feature vector, the method further includes the following steps:

[0026] Perform nonlinear transformation and dimensionality reduction on the fused feature vector.

[0027] As a further improvement of the present invention, after obtaining the virtual image, the method further includes the following steps:

[0028] The discriminator of the adversarial network is used to compare the virtual image with the real image sample to obtain the comparison result, and the generator adjusts the virtual image according to the comparison result.

[0029] As a further improvement of the present invention, after obtaining the intelligent agent, the generator adjusts the virtual image according to the comparison result, including:

[0030] Define the loss function of the generator;

[0031] Forward propagation: Input the fused feature vector into the generator to obtain a virtual image, and input the virtual image into the discriminator to obtain a probability value;

[0032] Back propagation: Generate a loss function based on the probability value, and use the back propagation algorithm to calculate the gradient of the generator based on the loss function;

[0033] Update parameters: Use optimization algorithms to update the parameters of the generator according to the gradient;

[0034] Repeat the steps of forward propagation, backpropagation, and parameter update until the virtual image obtained by the generator meets the requirements.

[0035] As a further improvement of the present invention, after obtaining the intelligent agent, the method further includes the following steps:

[0036] Save the virtual image and corresponding interactive behavior.

[0037] The present invention also provides a learning machine for implementing the above-mentioned generation method, comprising:

[0038] A user input module, used for allowing users to input user input data;

[0039] An AI generation module is configured to process user input data to obtain a first eigenvector, process learning behavior data to obtain a second eigenvector, fuse the first eigenvector and the second eigenvector to obtain a fused eigenvector, and input the fused eigenvector into a generator of a generative adversarial network to obtain a virtual image. The AI ​​generation module is further configured to employ a machine learning algorithm to impart interactive behaviors to the virtual image based on the user input data and the learning behavior data to obtain an intelligent agent.

[0040] An interaction module, for displaying the intelligent agent and interactive behaviors;

[0041] The storage module is used to store input data, agents and corresponding interaction behaviors.

[0042] The present invention also provides an electronic device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the above-mentioned generation method.

[0043] Compared with the existing technology, the beneficial effects of the present invention are as follows: a fused feature vector is obtained based on input data and learning behavior data, and a virtual image is generated through a generator of an adversarial network. It can combine children's interests and personality characteristics to generate a virtual image that meets their emotional needs. At the same time, interactive behaviors are also generated based on user input data and learning behavior data. In the process of interacting with users, it can better meet the actual needs of children and thus increase children's interest in learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the generation method of the present invention;

[0045] Figure 2 Schematic diagram of the structure of the learning machine of the present invention. DETAILED DESCRIPTION

[0046] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0047] The present invention provides a method for generating an intelligent agent, such as Figure 1 As shown, the steps include:

[0048] S1. Obtain user input data and learning behavior data. Through this step, the user can input user input data through the user interface according to his or her personal preferences and customize the generation of the virtual image. At the same time, the user's preference for the virtual image can also be inferred based on the learning behavior data.

[0049] First, based on the user's input, obtain the user input data, which may include: material data, template data, and property settings:

[0050] Material data: Users upload their own prepared images, videos, and other materials to the system through the upload interface. These materials may include information about the facial features, clothing styles, and accessories that children desire for their intelligent agents to possess. For example, a user might upload a picture of a favorite cartoon character and hope that the intelligent agent's facial image resembles that character.

[0051] Template data: The system provides a variety of preset agent image templates, such as "cute elf", etc. Users can choose the template that suits their needs. The template contains a set of basic image features and style information.

[0052] Attribute Settings: Users can set the agent's personality traits (e.g., cheerful, quiet, optimistic, attentive), language habits (e.g., humorous, concise, catchphrases, speaking speed), and interests (e.g., painting, music). These attributes will serve as an important basis for subsequently generating the agent's image and behavior patterns.

[0053] The system converts, cleans, and de-noises user-input data. Because material data, template data, and attribute settings are different types of data, pre-processing is required. For example, user-uploaded images and video materials of different formats are converted to a system-supported format. Common image formats (such as JPEG and PNG) are converted to standard formats for internal processing to facilitate subsequent feature extraction operations. Uploaded materials are cleaned to remove noise and irrelevant information. For example, for image materials, background interference elements are removed; for video materials, noise and unclear parts are eliminated.

[0054] Extract feature information and / or text information from user input data; for image materials, use computer vision technology to extract facial features (such as eye shape, nose outline), color and other feature information; for text-based attribute settings, use natural language processing technology to extract keywords and semantic information.

[0055] Connect with the learning platform to obtain the user's learning behavior data during the learning process, such as learning time, learning content preferences, answer accuracy, attention span, etc. These data can be obtained by connecting with learning software, online learning platforms, etc., or directly recorded in the learning module within the system.

[0056] Perform data cleaning on learning behavior data to remove outliers and erroneous data, such as abnormal records of learning time due to system failures.

[0057] Normalize learning behavior data. Normalize different types of learning behavior data to have the same dimensions and range to facilitate subsequent analysis and integration. For example, normalize data such as learning time and answer accuracy to the range [0, 1].

[0058] S2. Process the user input data to obtain a first eigenvector, and process the learning behavior data to obtain a second eigenvector;

[0059] If the user input data is image data, the convolutional neural network encodes the feature information to obtain the first feature vector. This means that the feature information extracted in the previous step (such as facial features and color features) is further encoded using a convolutional neural network (CNN). For example, a pre-trained ResNet model can be used to input image features into some layers of the ResNet to obtain a fixed-length feature vector. This vector can more effectively represent the semantic information of the image.

[0060] If the user input data is text, the text information is converted into vectors and processed using a neural network to obtain the first feature vector. Specifically, for keywords and semantic information extracted from text information such as attribute settings, word embedding (such as Word2Vec and GloVe) is used to convert the text into a vector representation. These vectors are then processed using a recurrent neural network (RNN) or a long short-term memory network (LSTM) to obtain the overall first feature vector of the text.

[0061] The normalized learning behavior data, such as learning time, answering accuracy, etc., are numerical data in themselves. The normalized learning behavior data can be spliced ​​to obtain the second eigenvector, or a weighted connection neural network can be used to perform nonlinear transformation on the learning behavior data to extract the second eigenvector.

[0062] S3, fusing the first eigenvector and the second eigenvector to obtain a fused eigenvector, specifically comprising the following steps:

[0063] Calculating attention weights: The first feature vector (the first feature vector is: image feature vector and / or text feature vector) and the second feature vector are input into the attention mechanism module. First, the importance weight of each feature vector is calculated. For example, a multi-layer perceptron (MLP) can be used to calculate the similarity score between each feature vector and a learnable query vector. The softmax function is then used to convert these scores into a probability distribution, which serves as the attention weight.

[0064] Weighted summation: Based on the attention weights, the first and second eigenvectors are weighted and summed to produce a fused eigenvector. Specifically, each eigenvector is multiplied by its corresponding attention weight and then summed to produce a fused eigenvector. This fused vector combines information from different data types and weights them according to their importance.

[0065] Perform a nonlinear transformation on the fused feature vector, such as using the ReLU activation function, to increase the model's expressive power. If the fused feature vector is high in dimensionality, dimensionality reduction can be performed using methods such as principal component analysis (PCA) or autoencoders to reduce computational effort and avoid overfitting. This process can be performed on the original fused feature vector or on the fused feature vector after nonlinear transformation. If the original fused feature vector is high in dimensionality, resulting in computational complexity and a high risk of overfitting, direct dimensionality reduction can simplify the data structure. For example, principal component analysis (PCA) can identify the main components, reducing the dimensionality while preserving key information. If the feature vector remains high in dimensionality after nonlinear transformation and you wish to preserve the nonlinear relationships and local features of the data, dimensionality reduction of the transformed vector is more appropriate. For example, t-SNE (T-distributed stochastic neighbor embedding) can reveal the complex internal structure of the data in a low-dimensional space.

[0066] It should be noted that the data preprocessing stage emphasizes the complete process of obtaining feature vectors from the original image. This involves first extracting relatively scattered and primitive feature information such as color, texture, and shape, then encoding it to convert it into a numerical feature vector that computers can efficiently process. Image feature encoding in the data fusion stage, on the other hand, is a re-encoding operation performed on top of preprocessing to meet fusion requirements. For example, during preprocessing, edge detection algorithms are used to extract image edge features, which are then converted into vectors using a specific encoding method. The fusion stage may further encode the encoded vectors using a convolutional neural network (CNN) to better integrate them with other data feature vectors, making their feature representation more suitable for the fusion task.

[0067] S4. Input the fused feature vector into the generator of the generative adversarial network to obtain a virtual image. The virtual image can be a two-dimensional image or a three-dimensional image. For example, if the user sets the personality of the intelligent agent to be cheerful, and the child shows a preference for sports content in learning, the generator may generate an intelligent agent image wearing sportswear and smiling.

[0068] S5. The adversarial network's discriminator compares the virtual image with real-world image samples to obtain a comparison result. The generator adjusts the virtual image based on the comparison result. The GAN's discriminator is used to evaluate the authenticity and personalization of the generated intelligent agent image. The discriminator compares the generated image with the real-world image sample and assigns a score. Based on the feedback from the discriminator, the generator continuously adjusts its own parameters to produce a more realistic and satisfactory image. The specific steps include the following:

[0069] Defining the generator's loss function: The goal of the generator is to generate an image of an agent that can deceive the discriminator. The cross-entropy loss function is usually used to measure the generator's loss. Suppose the image generated by the generator is G(z), where z is the fused feature vector, and the discriminator's output for the generated image is D(G(z)). The generator's loss function can be defined as:

[0070] LG=-log(D(G(z)))

[0071] The meaning of the above loss function is that the generator hopes that the discriminator will judge the image it generates as a real image, so the smaller the value of the loss function, the better.

[0072] Forward propagation: Input the fused feature vector z into the generator G to obtain the first virtual image G(z), and input the first virtual image G(z) into the discriminator D to obtain the probability value.

[0073] Backpropagation: Generate a loss function based on the probability value. Based on the generator's loss function LG, the backpropagation algorithm is used to calculate the gradient of the generator parameters. Specifically, using the chain rule, starting from the loss function, the gradient of each parameter is calculated step by step.

[0074] Update parameters: Use an optimization algorithm (such as stochastic gradient descent SGD, Adam, etc.) to update the parameters of the generator according to the calculated gradient. For example, in the Adam optimization algorithm, the parameter update formula is:

[0075] θG=θG-α·mt / (vt+∈)

[0076] Among them, θG is the parameter of the generator, α is the learning rate, mt and vt are the first-order moment and second-order moment estimates used in the Adam algorithm to adaptively adjust the learning rate, and ∈ is a small constant used to avoid division by zero errors.

[0077] The steps of forward propagation, backpropagation, and parameter updates are repeated until the virtual image generated by the generator can effectively deceive the discriminator, meaning that the discriminator cannot accurately distinguish between the generated image and the real image. Through continuous iteration of the generator and discriminator training process, the generated intelligent agent image reaches a satisfactory level. During this iteration, the system further adjusts the generator parameters based on real-time user feedback to improve the quality and personalization of the image.

[0078] Alternatively, after the virtual image is generated, it is provided for preview by the user, and the user provides feedback based on the current virtual image, and the generator optimizes the virtual image based on the feedback.

[0079] S6. Using a machine learning algorithm, the avatar is assigned interactive behaviors based on user input and learning behavior data, resulting in an intelligent agent. Specifically, based on the child's learning behavior data and user-defined attributes, the agent's interactive behaviors in different learning scenarios are generated. For example, if a child frequently encounters difficult problems in math, the agent might use encouragement and guidance to help the child solve the problem. If a child shows a strong interest in a particular subject, the agent might provide more relevant learning resources and expanded content.

[0080] Furthermore, during the interaction between the agent and the child, feedback data is collected in real time, such as the child's response to the agent's behavior and changes in learning outcomes. Based on this feedback, the agent's behavior patterns are optimized and adjusted to better suit the child's needs and learning habits.

[0081] S7. Save user input data, learning behavior data, avatar, and corresponding interaction behaviors. This data can be used as a historical record for subsequent analysis and comparison, and can also provide data support for further optimization of the intelligent agent. In addition, a learning report can be generated based on the interaction process.

[0082] The present invention integrates user input data and learning behavior data to generate an intelligent body image that is more in line with children's needs; it uses an adversarial network to generate high-resolution, personalized intelligent body images, which can support the dynamic generation of facial expressions and body movements; in addition, it can automatically adjust the parameters of the generation model based on user feedback and learning behavior data to improve the accuracy and personalization of image generation; according to children's learning behavior data, the intelligent body image is dynamically adjusted to enhance the learning incentive effect.

[0083] Through the agent generation method of this invention, the generated agent image and interactive behaviors can be applied to children's learning scenarios, such as learning software and online education platforms. The agent can interact with children, providing learning guidance, encouragement, and companionship, helping them improve their learning interest and learning outcomes. Simultaneously, data on interactions between children and the agent will continue to be collected, forming a closed-loop data processing process to continuously optimize the agent's image and behavior.

[0084] Based on the same inventive concept, the present invention also provides a learning machine, such as Figure 2As shown, the method for implementing the above-mentioned generation method as claimed in the claim includes: a user input module, an AI generation module, an interaction module and a storage module, wherein the user input module is used for the user to input user input data; the AI ​​generation module is used to process the user input data to obtain a first feature vector, process the learning behavior data to obtain a second feature vector, and fuse the first feature vector and the second feature vector to obtain a fused feature vector, and input the fused feature vector into the generator of the generative adversarial network to obtain a virtual image, and the AI ​​generation module is further used to use a machine learning algorithm to assign interactive behavior to the virtual image based on the user input data and the learning behavior data to obtain an intelligent agent; the interaction module is used to display the intelligent agent and interactive behavior; and the storage module is used to store input data, intelligent agents and corresponding interactive behaviors. The specific implementation process is not detailed here.

[0085] Based on the same inventive concept, the present invention also provides an electronic device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the above-mentioned generation method.

[0086] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0087] The memory can be used to store the computer program or module, and the processor implements the various functions of the trajectory planning optimization method by running or executing the computer program or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0088] The above is a detailed introduction to the method for generating an intelligent body, storage medium and electronic device provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application. The above embodiments are only preferred embodiments for fully illustrating the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the scope of protection of the present invention.

Claims

1. A method for generating an intelligent agent, characterized in that: Including steps: Obtain user input data and learning behavior data; Processing user input data to obtain a first eigenvector, and processing learning behavior data to obtain a second eigenvector; Fusing the first eigenvector and the second eigenvector to obtain a fused eigenvector; The fused feature vector is input into the generator of the generative adversarial network to obtain a virtual image; Using machine learning algorithms, interactive behaviors are given to the virtual image based on user input data and learning behavior data to obtain an intelligent entity.

2. The generation method according to claim 1, characterized in that The step of obtaining user input data and learning behavior data includes: According to the user's input, obtain user input data; Perform format conversion, cleaning, and noise reduction on user input data; Extracting feature information and / or text information from user input data; Connect to the learning platform to obtain users' learning behavior data during the learning process; Perform data cleaning on learning behavior data; Normalize the learning behavior data.

3. The generation method according to claim 2, characterized in that The steps of processing the user input data to obtain the first feature vector and processing the learning behavior data to obtain the second feature vector include: If the user input data is image data, the convolutional neural network encodes the feature information to obtain the first feature vector; If the user input data is text digital, the text information is converted into a vector, and the vector is processed using a neural network to obtain the first eigenvector; The normalized learning behavior data are spliced ​​to obtain a second eigenvector, or a neural network is used to perform nonlinear transformation on the learning behavior data to extract the second eigenvector.

4. The generation method according to claim 3, characterized in that The step of fusing the first eigenvector and the second eigenvector to obtain a fused eigenvector includes: Calculate the attention weights of the first eigenvector and the second eigenvector respectively; According to the attention weight, the first eigenvector and the second eigenvector are weighted and summed to obtain a fused eigenvector.

5. The generation method according to claim 4, characterized in that After obtaining the fused feature vector, the following steps are also included: Perform nonlinear transformation and dimensionality reduction on the fused feature vector.

6. The generation method according to claim 1, characterized in that After obtaining the virtual image, the following steps are also included: The discriminator of the adversarial network is used to compare the virtual image with the real image sample to obtain the comparison result, and the generator adjusts the virtual image according to the comparison result.

7. The generation method according to claim 1, characterized in that After obtaining the intelligent agent, the generator adjusts the virtual image according to the comparison result, including: Define the loss function of the generator; Forward propagation: Input the fused feature vector into the generator to obtain a virtual image, and input the virtual image into the discriminator to obtain a probability value; Back propagation: Generate a loss function based on the probability value, and use the back propagation algorithm to calculate the gradient of the generator based on the loss function; Update parameters: Use optimization algorithms to update the parameters of the generator according to the gradient; Repeat the steps of forward propagation, backpropagation, and parameter update until the virtual image obtained by the generator meets the requirements.

8. The generation method according to claim 1, characterized in that After obtaining the agent, the following steps are also included: Save user input data, learning behavior data, virtual images, and corresponding interactive behaviors.

9. A learning machine, characterized in that Used to implement the generation method according to any one of claims 1 to 8, comprising: A user input module, used for allowing users to input user input data; An AI generation module is configured to process user input data to obtain a first eigenvector, process learning behavior data to obtain a second eigenvector, fuse the first eigenvector and the second eigenvector to obtain a fused eigenvector, and input the fused eigenvector into a generator of a generative adversarial network to obtain a virtual image. The AI ​​generation module is further configured to employ a machine learning algorithm to impart interactive behaviors to the virtual image based on the user input data and the learning behavior data to obtain an intelligent agent. An interaction module, for displaying the intelligent agent and interactive behaviors; The storage module is used to store input data, agents and corresponding interaction behaviors.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the generation method as described in any one of claims 1 to 8.

Images