A Gamification Creation System and Method Based on Machine Learning

Abstract

This invention relates to the field of machine learning computing systems, and more particularly to a machine learning-based gamified creation system and method, including a sensor data acquisition module, a gamified data recognition module, a machine learning creation module, a game creation output module, and a real-time feedback optimization module. This invention generates personalized game characters based on gamified data, supports style transfer and detail enhancement, introduces network latency monitoring and device computing power margin assessment, dynamically adjusts creation parameters, reduces model resolution or enables local caching to ensure smooth creation, and improves the efficiency of virtual character gamification creation. It identifies highly popular public characters through traffic analysis, corrects output results using similarity coefficients to avoid overlap between characters and public characters, and improves the efficiency and effectiveness of virtual character gamification creation by forming a "data-creation-feedback" closed loop.

Description

Technical Field

[0001] This invention relates to the field of machine learning computing system technology, and in particular to a gamified creation system and method based on machine learning. Background Technology

[0002] In terms of computational power generation in gamified creation fields such as the metaverse that require virtual avatars, although the introduction of distributed rendering architecture and neural radiation field technology has significantly improved rendering efficiency in recent years, and AI-driven reinforcement learning algorithms have further realized intelligent optimization of rendering parameters, combined with NVIDIA DLSS technology, reducing computational power requirements while ensuring image quality, there is still a significant delay when capturing facial expressions with low computational power. Moreover, existing solutions mostly use the WebRTC protocol or fog computing deployment, but there is still a delay of more than 20ms. While emerging non-invasive brain-computer interface technology directly converts neural signals into virtual expressions, breaking through the latency bottleneck of traditional sensors and providing a new path for real-time interaction, it still has a large computational power requirement. In the field of public image identification, although blockchain NFT technology has been used for the unique authentication of virtual avatars, and biometric watermarking and multimodal behavior recognition technology have enhanced the security of identity verification, after gamifying virtual avatars according to user needs, it is still impossible to distinguish public images while satisfying users' creative freedom.

[0003] Chinese Patent Publication No. CN116882482A discloses a method and apparatus for training a virtual avatar generation model and generating virtual avatars, involving artificial intelligence technologies such as computer vision, augmented reality, virtual reality, and deep learning, and applicable to scenarios such as metaverse and digital humans. However, this solution does not match the computing power required for the virtual avatar creation process, nor can it optimize the expression capture delay problem in the virtual avatar generation process. Summary of the Invention

[0004] To address this, the present invention provides a gamification creation system and method based on machine learning, which overcomes the problems of low efficiency and poor effect in the gamification creation of virtual characters caused by mismatched computing power, network latency, and lack of identification and differentiation of public images in the prior art.

[0005] To achieve the above objectives, in one aspect, the present invention provides a gamified creation system based on machine learning, comprising:

[0006] The sensor data acquisition module is used to acquire the source data of the creation through a group of sensor nodes;

[0007] The gamification data recognition module is used to identify the source data of the creation based on the data recognition model to obtain gamified data and non-gamified data. It is also used to adjust the acquisition process of the source data of the creation based on the number of non-gamified data from the same source.

[0008] The machine learning creation module is used to create game characters based on gamification data, monitor network latency, optimize the creation process of game characters in real time based on network latency, and adjust the real-time optimization process of game character creation based on the device's computing power margin.

[0009] The game creation output module is used to output game characters, identify public images based on the number of times game characters are accessed, correct the output results of game characters based on the public similarity coefficient of game characters, and optimize the identification process of public images based on the basic subject proportion of game characters.

[0010] The real-time feedback optimization module is used to optimize the acquisition process of source data in real time based on the frequency of gamified data points.

[0011] Furthermore, the gamified data recognition module sets the loss function of the data recognition model to a binary cross-entropy function, where N is the total number of samples in the data recognition test set. i To identify the sample sequence number in the data identification test set, y i To identify the actual category of the i-th sample in the test set, for gamified data y... i =1, for non-gamified data y i =0, Let be the prediction probability of the data recognition model, representing the prediction probability that the i-th sample is gamified data.

[0012] Furthermore, the gamified data identification module acquires the sensor node identifiers of non-gamified data, counts the non-gamified data with the same sensor node identifiers within the clustering monitoring period to obtain the number n of non-gamified data from the same source, compares the number n of non-gamified data from the same source with the preset number n0 of non-gamified data from the same source, and judges the clustering status of the non-gamified data based on the comparison result.

[0013] Furthermore, the machine learning creation module preprocesses the gamified data to obtain preprocessed gamified data, performs feature selection on the preprocessed gamified data to obtain gamified feature data, and selects a machine learning model for the gamified data according to the modeling requirements to obtain an adaptive creation model.

[0014] Furthermore, the machine learning creation module sends ICMP requests to the target address, measures the round-trip time of the ICMP requests, obtains the current network latency A, and compares the current network latency A with the maximum network latency Amax and the minimum network latency Amin. Amax ≥ 150ms and Amin ≤ 50ms. Based on the comparison results, the network latency is judged, and the creation process of the game character is optimized in real time based on the judgment results.

[0015] Furthermore, the machine learning creation module obtains the current real-time GPU utilization rate z, calculates the device computing power margin β based on the current real-time GPU utilization rate z and the theoretical maximum value zmax, compares the device computing power margin β with the preset device computing power margin β0, judges the current device computing power margin based on the comparison result, and adjusts the real-time optimization process of the game character creation process based on the judgment result.

[0016] Furthermore, the game creation output module obtains the game image from the machine learning creation module and outputs the game image. The game creation output module obtains the number of visits K of the network image through network retrieval, compares the number of visits K with the preset number of visits K0, where K0≥10000 times, and judges the category of the network image based on the comparison result.

[0017] Furthermore, the game creation output module inputs the labeled feature data and gamification feature data into the similarity comparison model, outputs the public similarity coefficient Q of the game character, and compares the public similarity coefficient Q of the game character with the preset similarity coefficient Q0, where Q0≥70%. Based on the comparison result, the degree of public similarity of the game character is judged, and the output result of the game character is corrected based on the judgment result.

[0018] Furthermore, the game creation output module inputs the game character into the basic subject proportion model and outputs the basic subject proportion rate. The basic subject proportion rate includes the basic body proportion r1 and the decoration proportion r2. The ratio of r1 to r2 is set to 100%. The basic body proportion r1 is compared with the preset body proportion r0. r0 ≥ 65%. The actual situation of the game character is judged based on the comparison result. The public image recognition process is optimized based on the judgment result.

[0019] On the other hand, the present invention also provides a gamification creation method based on machine learning, comprising:

[0020] Step S1: Acquire source data for creation through a group of sensor nodes;

[0021] Step S2, the gamification data recognition module, identifies the source data of the creation based on the data recognition model to obtain gamified data and non-gamified data, and also adjusts the acquisition process of the source data of the creation based on the number of non-gamified data from the same source;

[0022] Step S3 involves creating game characters based on gamification data, monitoring network latency, optimizing the creation process of game characters in real time based on network latency, optimizing the acquisition process of creation source data based on real-time optimization methods, and adjusting the real-time optimization process of game character creation based on device computing power margin.

[0023] Step S4 involves outputting the game character, identifying the public character based on the number of times the game character is accessed, correcting the output result of the game character based on the public similarity coefficient of the game character, and optimizing the identification process of the public character based on the basic subject ratio of the game character.

[0024] Step S5: Optimize the acquisition process of source data in real time based on the frequency of gamified data points.

[0025] Compared with existing technologies, the beneficial effects of this invention are as follows: The system uses a data identification model and non-gamified data, and dynamically adjusts the collection strategy based on the number of non-gamified data entries from the same source to reduce redundant data interference and improve data quality. The system generates personalized game characters based on gamified data through a machine learning creation module, supporting style transfer and detail enhancement. It introduces network latency monitoring and device computing power margin assessment, dynamically adjusts creation parameters, reduces model resolution, or enables local caching to ensure smooth creation and improve the efficiency of virtual character gamification. The system also identifies high-popularity public characters through traffic analysis, corrects output results with similarity coefficients to avoid overlap between characters and public characters, optimizes public character identification logic based on the proportion of basic subjects to avoid excessive homogenization, preserves creator style, and improves the creation effect of virtual character gamification. Furthermore, the system monitors the frequency of gamified data points and optimizes data collection strategies in real time, forming a "data-creation-feedback" closed loop to improve the creation efficiency and effect of virtual character gamification. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the structure of the machine learning-based gamified creation system in this embodiment;

[0027] Figure 2 This is a flowchart illustrating the gamified creation method based on machine learning in this embodiment. Detailed Implementation

[0028] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0029] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0030] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0031] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0032] Please see Figure 1 As shown, this is a schematic diagram of the structure of the machine learning-based gamified creation system of this embodiment. The system includes:

[0033] The sensor data acquisition module is used to acquire the source data of the creation through a group of sensor nodes;

[0034] The gamification data recognition module is used to identify the source data of the creation based on the data recognition model to obtain gamified data and non-gamified data. It is also used to adjust the acquisition process of the source data of the creation based on the number of non-gamified data from the same source. The gamification data recognition module is connected to the sensor data acquisition module.

[0035] The machine learning creation module is used to create game characters based on gamification data, monitor network latency, optimize the creation process of game characters in real time based on network latency, and adjust the real-time optimization process of game character creation based on the device's computing power margin. The machine learning creation module is connected to the gamification data recognition module.

[0036] The game creation output module is used to output game characters, identify public images based on the number of visits to game characters, correct the output results of game characters based on the public similarity coefficient of game characters, and optimize the identification process of public images based on the basic subject proportion of game characters. The game creation output module is connected to the machine learning creation module.

[0037] The real-time feedback optimization module is used to optimize the acquisition process of creation source data in real time based on the frequency of gamified data points. The real-time feedback optimization module is connected to the game creation output module.

[0038] Specifically, the system is installed in the gamified creation terminal for virtual avatars. It acquires source data through a cluster of sensor nodes on the terminal and creates and outputs game avatars based on this data. The system uses a data recognition model to distinguish between game avatars and non-game data, dynamically adjusting the acquisition strategy based on the number of similar non-game data entries to reduce redundant data interference and improve data quality. The system uses a machine learning creation module to generate personalized game avatars based on game data, supporting style transfer and detail enhancement. It incorporates network latency monitoring and device computing power margin assessment to dynamically adjust creation parameters, reducing model resolution or enabling local caching to ensure smooth creation and improve the efficiency of virtual avatar gamification. The system also identifies highly popular public avatars through traffic analysis, corrects output results using similarity coefficients to avoid overlap between avatars and public avatars, optimizes public avatar recognition logic based on the proportion of basic subjects to avoid excessive homogenization, preserves the creator's style, and enhances the gamified creation effect of virtual avatars. Furthermore, the system monitors the frequency of game data points and optimizes data acquisition strategies in real time, forming a "data-creation-feedback" closed loop to improve the efficiency and effectiveness of virtual avatar gamification.

[0039] Specifically, the sensor data acquisition module connects to the sensor node group via a hybrid connection method and acquires the creation source data at a preset acquisition frequency P0 using an event-driven approach. The sensor node group refers to a collection of sensor devices distributed across various locations on the wearable device. This embodiment does not limit the location of the sensor node group; those skilled in the art can freely set it according to actual conditions, as long as it meets the requirement of sufficient data acquisition from various locations on the wearable device. For example, it can be placed on the head of the wearable device, with an infrared imaging sensor facing the wearer's face. The creation source data refers to the raw data collected by the sensor node group for creating game characters. The creation source data is labeled with the sensor node source, which serves as the sensor node identifier. This embodiment does not limit the labeling method for the sensor node identifier; those skilled in the art can label the creation source data identifier according to actual conditions. For example, the sensor device identification code can be added as a suffix to the creation source data. This embodiment does not limit the specific type of creation source data; those skilled in the art can freely set it according to actual conditions. For example, the creation source data can be set to include visual data, audio data, motion data, environmental data, and wearer behavior data. The data refers to images and videos collected by visual sensors to capture the wearer's facial expressions, movements, and environmental scenes. The audio data refers to sounds collected by microphones to analyze the wearer's voice commands and environmental sound effects. The motion data refers to wearer motion information collected by sensors such as accelerometers and gyroscopes to capture the wearer's movements or postures. The environmental data refers to environmental information collected by sensors such as temperature, humidity, and light to adjust the performance of game scenes or characters. The wearer behavior data refers to wearer interaction data collected by interaction sensors, such as clicks and swipes, to analyze wearer habits and preferences. The hybrid connection method refers to a connection scheme that combines wireless and wired connections, selected based on node characteristics. This embodiment does not limit the content of the hybrid connection method; those skilled in the art can freely set it according to actual conditions, as long as it meets the needs for acquiring the source data. For example, Wi-Fi can be used for wireless connection, and Ethernet can be used for wired connection. The event-driven method refers to each sensor node in the sensor node group actively sending data to the sensor data acquisition module when it detects a specific event. For example, when there are fluctuations in the wearer's facial expression data in the visual data, the visual data can be sent to the sensor data acquisition module.

[0040] Specifically, the gamified data recognition module selects a convolutional neural network (CNN) model as the basic framework for the data recognition model. It divides the data recognition training dataset into three parts: 70% as the training set, 15% as the validation set, and 15% as the test set. The CNN model is trained on the training set, and its weights are updated using backpropagation. After each epoch, the CNN model is validated using the validation set to obtain a validation loss value and a validation accuracy. When the validation loss value and the validation accuracy meet preset validation conditions, the CNN model is tested on the test set to obtain a validation test accuracy. When the validation test accuracy reaches a preset accuracy, the CNN model is output as the data recognition model. The gamified data recognition module converts the source data into image format and inputs it into the data recognition model, obtaining the data recognition results output by the model. The data recognition results include gamified and non-gamified data.

[0041] Specifically, the data recognition learning sample dataset refers to a learning dataset stored in the form of source data and data recognition results, used to train a convolutional neural network model. The epoch refers to the process by which the convolutional neural network model completes one forward and backward propagation on the data recognition training set. The validation loss value refers to the loss value of the loss function in the convolutional neural network model when the data recognition validation set is input into the model for validation. The validation accuracy refers to the ratio of the number of data recognition results output by the convolutional neural network model that match the data recognition results in the data recognition validation set to the total number of samples in the data recognition validation set. The preset verification condition refers to the fact that the verification loss value does not decrease and the verification accuracy does not improve for 5 consecutive epochs. The recognition test accuracy refers to the ratio of the number of data test results that match the data recognition results in the data recognition test set after the data recognition test set is input into the convolutional neural network model to the total number of samples in the data recognition test set. The preset accuracy refers to the preset value of the recognition test accuracy of the convolutional neural network model to determine whether it has reached the output standard. For example, the preset accuracy can be set to 98%. The gamified data refers to the data in the creation source data used to create game characters. The non-gamified data refers to the data in the creation source data that is not used to create game characters.

[0042] It is understood that this embodiment does not limit the method of converting the source data of creation into image form. Those skilled in the art can limit it according to the actual situation, as long as the feature expression requirements of the source data of creation are met. For example, it can be set to extract features from visual data through convolutional layers, adjust the image in the visual data to a fixed size, such as 224x224, and perform normalization processing, convert audio data into Mel-frequency cepstral coefficients (MFCC) and treat it as an image to input into a CNN, reconstruct motion data, environmental data, and wearer behavior data into images according to time series, or use a one-dimensional convolutional network for processing, and process the environmental data through a simple fully connected layer, extract the features of wearer behavior data using feature engineering, and combine them into the convolutional neural network model.

[0043] Specifically, the gamified data recognition module sets the loss function of the data recognition model to a binary cross-entropy function. N is the total number of samples in the data recognition test set. i To identify the sample sequence number in the data identification test set, y i To identify the actual category of the i-th sample in the test set, for gamified data y... i =1, for non-gamified data y i =0, Let be the prediction probability of the data recognition model, representing the prediction probability that the i-th sample is gamified data.

[0044] Specifically, the gamified data identification module acquires the sensor node identifiers of non-gamified data, counts the non-gamified data with the same sensor node identifiers within the clustering monitoring period to obtain the number n of homogeneous non-gamified data, and compares the number n with a preset number n0 of homogeneous non-gamified data. Based on the comparison result, the clustering status of the non-gamified data is determined, wherein:

[0045] When n≤n0, the gamified data recognition module determines that the clustering state of the non-gamified data is not clustered;

[0046] When n > n0, the gamified data recognition module determines that the clustering state of the non-gamified data has reached clustering.

[0047] The gamified data recognition module adjusts the acquisition process of creation source data based on the clustering status of non-gamified data, wherein:

[0048] If the clustering status of the non-gamified data is "not clustered", the gamified data identification module will not adjust the acquisition process of the source data.

[0049] If the clustering state of the non-gamified data is not clustered, the gamified data identification module adjusts the acquisition process of the source data of the creation. According to the clustering adjustment coefficient α, the preset acquisition frequency P0 of the sensor node corresponding to the non-gamified data is adjusted. The adjusted preset acquisition frequency is Pa0. Pa0 = α × P0 is set, and 1 < α < 1.2 is set.

[0050] Specifically, the preset acquisition frequency P0 refers to the frequency at which the sensor data acquisition module acquires the source data of the creation through an event-driven method. This embodiment does not specifically limit the value of the preset acquisition frequency P0; those skilled in the art can freely set it according to actual conditions, as long as it meets the requirement of sufficient acquisition of the source data of the creation. For example, the preset acquisition frequency P0 can be set to 1 second / time. The clustering monitoring period refers to whether the clustering state of the non-gamified data has reached the preset clustering period duration. For example, the clustering monitoring period can be set to 48 hours. The same sensor node identifier refers to the sensor node identifier having the same character expression content. This embodiment does not include clustering monitoring... The counting method for non-gamified data with the same sensor node identifier within a period is specifically defined. Those skilled in the art can freely set it according to the actual situation, as long as it meets the counting requirements for non-gamified data with the same sensor node identifier. For example, it can be set to count non-gamified data with the same character expression content of sensor node identifiers as one. The preset number of homogeneous non-gamified data entries refers to a preset value that reflects the clustering state of the non-gamified data and the number of homogeneous non-gamified data entries that have reached the clustering. For example, the preset number of homogeneous non-gamified data entries n0 = 30 can be set. The sensor node corresponding to the non-gamified data refers to the sensor node that is the source of the acquisition of the non-gamified data.

[0051] Specifically, the machine learning creation module preprocesses the gamified data to obtain preprocessed gamified data, performs feature selection on the preprocessed gamified data to obtain gamified feature data, and selects a machine learning model for the gamified data according to modeling requirements to obtain an adaptive creation model, wherein:

[0052] When the modeling requirement is to create a new game character, the machine learning creation module selects a generative adversarial network model as the adaptive creation model.

[0053] When the modeling requirement is to reconstruct the game character, the machine learning creation module selects the variational autoencoder model as the adaptive creation model.

[0054] The machine learning creation module inputs gamification feature data into the adaptive creation model for creation, and obtains the game image output by the adaptive creation model.

[0055] Specifically, this embodiment does not limit the method of preprocessing the gamification data. Those skilled in the art can freely set it according to the actual situation, as long as it meets the preprocessing requirements of the gamification data. For example, in this embodiment, the gamification data is converted into a unified JSON format to obtain unified format gamification data. The unified format gamification data is then input into a multimodal learning framework to obtain fused feature gamification data. The fused feature gamification data is used as preprocessed gamification data. In this embodiment, feature selection is performed on the preprocessed gamification data according to the adaptive creation model, wherein:

[0056] When the machine learning creation module selects a generative adversarial network model as the adaptive creation model, this embodiment converts the non-image features in the preprocessed gamified data into image channels to obtain image channel gamified data. The image channel gamified data is then aligned using the Cross-Modal Attention CLIP style and weighted feature extraction is performed to obtain gamified feature data.

[0057] When the machine learning creation module selects the variational autoencoder model as the adaptive creation model, this embodiment performs semantic group encoding on the preprocessed gamified data to obtain semantic group encoded gamified data, and performs sensitive attribute isolation on the semantic group encoded gamified data to obtain gamified feature data.

[0058] The modeling requirements refer to the requirements for constructing game characters, including creating new game characters and reconstructing game characters. This embodiment does not limit the method of obtaining modeling requirements; those skilled in the art can freely set them according to actual conditions, as long as they meet the actual requirements for reflecting modeling requirements. For example, the modeling requirements can be obtained through the user's operation path on the interface. The machine learning model for gamified data refers to the basic architecture model of the adaptive creation model. The adaptive creation model is a machine learning model that takes gamified feature data as input and game characters as output, including generative adversarial network (GAN) models and variational autoencoder (VAE) models. This embodiment does not limit the content of the GAN model, as long as it meets the requirements for generating game characters. For example, the gamified feature data can be normalized as an RGB image and then input into the GAN model for training and game character generation. This embodiment does not limit the content of the VAE model, as long as it meets the requirements for generating game characters. For example, the gamified feature data can be processed by Gaussian distribution and then input into the GAN model for training and game character generation.

[0059] Specifically, the machine learning creation module sends ICMP requests to the target address, measures the round-trip time of the ICMP requests to obtain the current network latency A, and compares the current network latency A with the maximum network latency Amax and the minimum network latency Amin. Amax ≥ 150ms, Amin ≤ 50ms. Based on the comparison results, the network latency is assessed, and the creation process of the game character is optimized in real time based on the assessment results.

[0060] When A < Amin, the machine learning creation module determines the network latency as excellent and does not perform real-time optimization of the game character creation process;

[0061] When Amin≤A≤Amax, the machine learning creation module determines the network latency as good and does not perform real-time optimization of the game character creation process;

[0062] When A > Amax, the machine learning creation module determines that the network latency is network congestion and optimizes the creation process of the game character in real time. It extracts historical gamification feature data of historical game characters, and uses the gamification feature that appears most frequently in the historical gamification feature data as creation habit data to supplement the current gamification feature data input into the adaptive creation model, and obtains the game character output by the adaptive creation model. The proportion of the creation habit data a1 input into the adaptive creation model is 30%, and the proportion of the gamification feature data b1 is 70%.

[0063] Specifically, the ICMP request refers to a data packet type in the Internet Control Message Protocol used for network diagnostics and status reporting, used to monitor network latency. The maximum network latency Amax refers to the maximum preset value used to judge network latency, and the minimum network latency Amin refers to the minimum preset value used to judge network latency. The network latency status refers to the quality of network latency, including network latency status as excellent, network latency status as good, and network latency status as congested. The historical game character refers to a historically created game character, and the historical gamification feature data refers to the gamification feature data of the historical game character.

[0064] Specifically, the machine learning creation module monitors the current network latency to obtain the current network situation, and optimizes the creation process of game characters in real time based on the current network situation. When optimizing the creation process of game characters in real time, creation habit data is added as a supplement to the gamification feature data and input into the adaptive creation model. Based on the user's historical habits, game characters are quickly produced, reducing the waiting time for users to create game characters.

[0065] Specifically, the machine learning creation module obtains the current real-time GPU utilization rate z, and calculates the device computing power margin β based on the current real-time GPU utilization rate z and the theoretical maximum value zmax. β is set to (1-z / zmax). The device computing power margin β is compared with a preset device computing power margin β0. Based on the comparison result, the current device computing power margin is judged, and the real-time optimization process of the game character creation process is adjusted according to the judgment result.

[0066] When β≥β0, the machine learning creation module determines that the current device's computing power margin is sufficient and does not adjust the real-time optimization process of the game character creation process;

[0067] When β < β0, the machine learning creation module determines that the current device's computing power margin is insufficient, and adjusts the real-time optimization process of the game character creation process according to the adjustment coefficient w = 1.3 - 0.3 × e -(β0-β) The maximum network latency Amax is adjusted, where e is the base of the natural logarithm, to obtain the adjusted maximum network latency Amax'. Amax' is set to W × Amax. The original maximum network latency Amax is replaced with the adjusted maximum network latency Amax', and the network latency is reassessed.

[0068] Specifically, the current real-time GPU utilization rate refers to the percentage of GPU usage under the current circumstances. The GPU refers to a processor specifically designed for efficient processing of graphics rendering and parallel computing tasks. The Chinese name for GPU is Graphics Processing Unit. The device computing power margin refers to the remaining percentage of GPU usage. The preset device computing power margin refers to a preset value used to judge the current device computing power margin. The current device computing power margin refers to a description of the remaining computing power margin of the device. The current device computing power margin includes situations where the current device computing power margin is insufficient and situations where the current device computing power margin is sufficient.

[0069] Specifically, the machine learning creation module compares the device's computing power margin β with a preset device computing power margin β0. If β is less than β0, an adjustment coefficient w = 1.3 - 0.3 × e is applied. -(β0-β) Adjust the maximum network latency Amax to increase the maximum network latency and avoid misjudging network latency due to insufficient computing power.

[0070] Specifically, the game creation output module acquires the game image from the machine learning creation module and outputs the game image.

[0071] Specifically, the game creation output module obtains the number of visits K of a web character through network retrieval, compares the number of visits K with a preset number of visits K0, where K0 ≥ 10000 times, and determines the category of the web character based on the comparison result, wherein:

[0072] When K < K0, the game creation output module determines that the category of the online character is a normal character;

[0073] When K≥K0, the game creation output module determines that the image category of the online image is a public image, and marks the feature data of the online image to obtain marked feature data.

[0074] Specifically, the number of visits refers to the total number of times a web image is clicked and accessed. A web image refers to a person or character appearing on the internet. The preset number of visits refers to a preset value used to determine the category of a web image. The category of a web image refers to the classification of web images. The category of a web image includes the category of a common image and the category of a public image. A public image refers to a well-known person or character. The feature data of a web image refers to the image parameters obtained by scanning the web image using 3D modeling software. The image parameters include facial organ parameters and face size parameters.

[0075] Specifically, the game creation output module categorizes online images through network retrieval to differentiate them from public images, making it easier for the game creation output module to correct the output game images.

[0076] Specifically, the game creation output module inputs the labeled feature data and gamification feature data into a similarity comparison model, outputs the public similarity coefficient Q of the game character, and compares the public similarity coefficient Q with a preset similarity coefficient Q0, where Q0 ≥ 70%. Based on the comparison result, the degree of public similarity of the game character is judged, and the output result of the game character is corrected based on the judgment result.

[0077] When Q < Q0, the game creation output module determines that the public similarity of the game character is not similar and does not correct the output result of the game character.

[0078] When Q≥Q0, the game creation output module determines that the similarity between the game image and the public is similar, corrects the output result of the game image, uses the labeled feature data as the error training set to retrain the adaptive creation model, outputs the corrected game image, and the game creation output module replaces the original game image with the corrected game image for output.

[0079] Specifically, the similarity comparison model refers to a neural network learning model that takes labeled feature data and gamified feature data as input and outputs a public similarity coefficient of game characters. The similarity comparison model is constructed using a similarity comparison model construction method, which includes:

[0080] The historical comparison training set was divided into a simulated training set (70%) and a simulated validation set (30%). A recurrent neural network (RNN) model was selected as the neural network architecture for the similarity comparison model. The Adam optimizer and cross-entropy loss function were used to train the RNN model. The simulated training set was loaded into the RNN model, and forward propagation was performed through the RNN model to calculate the output value of the similarity comparison model. The loss function value was calculated based on the output value and the true value of the RNN model. The gradient was calculated through the backpropagation algorithm, and the weights and biases of the RNN model were updated. The process of forward propagation, loss calculation, and backpropagation was repeated until the preset number of training rounds was reached. The accuracy of the RNN model was verified through the simulated validation set. The RNN model with an accuracy of 90% was output as the similarity comparison model.

[0081] The historical comparison training set refers to the training set used to train the similarity comparison model. The historical comparison training set is obtained by analyzing historical marker feature data and historical gamification feature data. The public similarity coefficient refers to the specific numerical value of the similarity between the game character and the public character. The preset similarity coefficient refers to the preset value for judging the public similarity of the game character. The public similarity of the game character refers to the degree of similarity between the game character and the public character. The public similarity of the game character includes the public similarity of the game character being dissimilar and the public similarity of the game character being similar.

[0082] Specifically, the game creation output module compares the public similarity coefficient of the game image output by the similarity comparison model with the preset similarity coefficient. If Q is greater than Q0, the output result of the game image is corrected to avoid disputes caused by the overlap between the game image and the public image.

[0083] Specifically, the game creation output module inputs the game character into a basic subject proportion model and outputs a basic subject proportion rate. This basic subject proportion rate includes a basic body proportion r1 and a decoration proportion r2, with r1 + r2 = 100%. The basic body proportion r1 is compared with a preset body proportion r0, where r0 ≥ 65%. Based on the comparison result, the actual situation of the game character is judged, and the public image recognition process is optimized based on the judgment result.

[0084] When r1≤r0, the game creation output module determines that the public game image is not real and does not optimize the public image recognition process;

[0085] When r1 > r0, the game creation output module determines that the game character is real and optimizes the public character recognition process, based on the recognition optimization coefficient γ = 1 + 0.4 × e-(r1-r0) The preset access volume K0 is optimized, where e is the base of the natural logarithm, to obtain the optimized preset access volume K0'. The optimized preset access volume K0' is set to K0 × γ, and the preset access volume K0 is replaced with the optimized preset access volume K0'.

[0086] Specifically, the basic subject proportion model refers to a neural network learning model that takes a game character as input and outputs a basic subject proportion rate. The basic subject proportion model is constructed using a basic subject proportion model construction method, which includes:

[0087] The basic subject proportion analysis dataset is divided into three parts: 70% as the analysis training set, 15% as the analysis validation set, and 15% as the analysis test set. A convolutional neural network (CNN) model is trained based on this dataset. The weights of the CNN model are updated using a backpropagation algorithm. After each epoch, the CNN model is validated using the analysis validation set to obtain the recognition validation loss value and the recognition validation accuracy. When the validation loss value and the validation accuracy meet preset validation conditions, the CNN model is tested using the analysis test set to obtain the analysis test accuracy. When the analysis test accuracy reaches a preset accuracy, the CNN model is output as the basic subject proportion analysis model.

[0088] The basic subject proportion analysis dataset refers to the learning dataset used to train the basic subject proportion analysis model. An epoch refers to the process of the convolutional neural network model completing one forward and backward propagation on the data recognition training set. The validation loss value refers to the loss value of the loss function in the convolutional neural network model when the data recognition validation set is input into the model for validation. The validation accuracy refers to the ratio of the number of output analysis results consistent with the analysis results in the validation set to the total number of samples in the validation set. The preset validation condition is that the validation loss value does not decrease and the validation accuracy does not improve for seven consecutive epochs. The analysis test accuracy refers to the accuracy of the analysis test set input into the model. The ratio of the number of test results output by the convolutional neural network model that match the data recognition results in the analytical test set to the total number of samples in the analytical test set. The preset accuracy rate refers to the preset value of the analytical accuracy rate for judging whether the convolutional neural network model has reached the output standard. For example, the preset accuracy rate can be set to 95%. The basic body proportion refers to the proportion of the body part of the game character to the overall part of the game character. The decoration proportion refers to the proportion of the decoration part of the game character to the overall part of the game character. The preset body proportion is a preset value used to judge the realism of the game character. The realism of the game character refers to the similarity between the game character and the real person. The realism of the game character includes the realism of the game character and the unrealism of the game character.

[0089] Specifically, by optimizing the preset visit volume K0 and reducing its value, the game image becomes closer to the real state when the basic body proportion is greater than the preset body proportion. The preset visit volume K0 should be reduced to make the public image judgment requirements more stringent and avoid the overlap between the game image and the public image.

[0090] Specifically, the real-time feedback optimization module acquires the gamification data point frequency c1 through a sensor node group, compares c1 with a preset gamification data point frequency c0, where c0 ≥ 7 times, determines the importance of the gamification data point based on the comparison result, and optimizes the acquisition process of the source data in real time based on the determination result.

[0091] When c1≥c0, the real-time feedback optimization module determines that the gamification data points are important and does not perform real-time optimization on the process of acquiring the source data.

[0092] When c1 < c0, the real-time feedback optimization module determines that the importance of the gamified data point is not important, and optimizes the acquisition process of the source data in real time, canceling the acquisition of the source data of the gamified data point.

[0093] Specifically, the frequency of gamified data points refers to the frequency at which data is acquired from gamified data points, the preset frequency of gamified data points refers to a preset value used to judge the importance of gamified data points, the importance of gamified data points refers to the degree of importance of gamified data points, and the importance of gamified data points includes gamified data points being important and gamified data points being unimportant.

[0094] Specifically, the real-time feedback optimization module compares the frequency of gamified data points c1 with the preset frequency of gamified data points c0 to obtain the importance of the gamified data points, and optimizes the acquisition process of the source data in real time according to the importance. If c1 is less than c0, the acquisition process of the source data is optimized in real time by canceling the acquisition of the source data for that gamified data point, thereby reducing the data collection of unimportant game points, reducing the generation time of game characters and the computing power burden on the GPU.

[0095] Please see Figure 2 As shown, this is a flowchart illustrating the gamification creation method based on machine learning in this embodiment. The method includes:

[0096] Step S1: Acquire source data for creation through a group of sensor nodes;

[0097] Step S2, the gamification data recognition module, identifies the source data of the creation based on the data recognition model to obtain gamified data and non-gamified data, and also adjusts the acquisition process of the source data of the creation based on the number of non-gamified data from the same source;

[0098] Step S3 involves creating game characters based on gamification data, monitoring network latency, optimizing the creation process of game characters in real time based on network latency, optimizing the acquisition process of creation source data based on real-time optimization methods, and adjusting the real-time optimization process of game character creation based on device computing power margin.

[0099] Step S4 involves outputting the game character, identifying the public character based on the number of times the game character is accessed, correcting the output result of the game character based on the public similarity coefficient of the game character, and optimizing the identification process of the public character based on the basic subject ratio of the game character.

[0100] Step S5: Optimize the acquisition process of source data in real time based on the frequency of gamified data points.

[0101] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A gamified creation system based on machine learning, characterized in that, include: The sensor data acquisition module is used to acquire the source data of the creation through a group of sensor nodes; The gamification data recognition module is used to identify the source data of the creation based on the data recognition model to obtain gamified data and non-gamified data. It is also used to adjust the acquisition process of the source data of the creation based on the number of non-gamified data from the same source. The machine learning creation module is used to create game characters based on gamification data, monitor network latency, optimize the creation process of game characters in real time based on network latency, and adjust the real-time optimization process of game character creation based on the device's computing power margin. The game creation output module is used to output game characters, identify public images based on the number of times game characters are accessed, correct the output results of game characters based on the public similarity coefficient of game characters, and optimize the identification process of public images based on the basic subject proportion of game characters. The real-time feedback optimization module is used to optimize the acquisition process of source data in real time based on the frequency of gamified data points; The gamified data recognition module sets the loss function of the data recognition model to the binary cross-entropy function. To identify the total number of samples in the test set. To identify the sample sequence number in the test set of data. For the data recognition test set, the first The actual category of each sample, for gamified data =1, for non-gamified data =0, Let be the predicted probability of the data recognition model, representing the probability of the data recognition model for the th... Each sample represents the predicted probability of gamified data; The gamified data identification module acquires the sensor node identifiers of non-gamified data and counts the non-gamified data with the same sensor node identifiers within the clustering monitoring period to obtain the number n of homogeneous non-gamified data. This number n is then compared with a preset number n0 of homogeneous non-gamified data. Based on the comparison result, the clustering status of the non-gamified data is determined, wherein: When n≤n0, the gamified data recognition module determines that the clustering state of the non-gamified data is not clustered; When n > n0, the gamified data recognition module determines that the clustering state of the non-gamified data has reached clustering. The gamified data recognition module adjusts the acquisition process of creation source data based on the clustering status of non-gamified data, wherein: If the clustering status of the non-gamified data is "not clustered", the gamified data identification module will not adjust the acquisition process of the source data. If the clustering state of the non-gamified data is not clustered, the gamified data identification module adjusts the acquisition process of the source data of the creation. According to the clustering adjustment coefficient α, the preset acquisition frequency P0 of the sensor node corresponding to the non-gamified data is adjusted. The adjusted preset acquisition frequency is Pa0, Pa0 = α × P0, and 1 < α < 1.

2.

2. The machine learning-based gamified creation system according to claim 1, characterized in that, The machine learning creation module preprocesses the gamified data to obtain preprocessed gamified data, performs feature selection on the preprocessed gamified data to obtain gamified feature data, and selects a machine learning model for the gamified data according to the modeling requirements to obtain an adaptive creation model.

3. The machine learning-based gamified creation system according to claim 2, characterized in that, The machine learning creation module sends ICMP requests to the target address, measures the round-trip time of the ICMP requests, obtains the current network latency A, and compares the current network latency A with the maximum network latency Amax and the minimum network latency Amin. Amax ≥ 150ms and Amin ≤ 50ms. Based on the comparison results, the network latency is judged, and the creation process of the game character is optimized in real time based on the judgment results.

4. The machine learning-based gamified creation system according to claim 3, characterized in that, The machine learning creation module obtains the current real-time GPU utilization rate z, and calculates the device computing power margin β based on the current real-time GPU utilization rate z and the theoretical maximum value zmax. The device computing power margin β is compared with the preset device computing power margin β0. Based on the comparison result, the current device computing power margin is judged, and the real-time optimization process of the game character creation process is adjusted according to the judgment result.

5. The machine learning-based gamified creation system according to claim 4, characterized in that, The game creation output module obtains the game image from the machine learning creation module and outputs the game image. The game creation output module obtains the number of visits K of the network image through network retrieval, compares the number of visits K with the preset number of visits K0, where K0≥10000 times, and judges the category of the network image based on the comparison result.

6. The machine learning-based gamified creation system according to claim 5, characterized in that, The game creation output module inputs the labeled feature data and gamification feature data into the similarity comparison model, outputs the public similarity coefficient Q of the game character, and compares the public similarity coefficient Q of the game character with the preset similarity coefficient Q0. Q0≥70%, judges the degree of public similarity of the game character based on the comparison result, and corrects the output result of the game character based on the judgment result.

7. The machine learning-based gamified creation system according to claim 6, characterized in that, The game creation output module inputs the game character into the basic subject proportion model and outputs the basic subject proportion rate. The basic subject proportion rate includes the basic body proportion r1 and the decoration proportion r2. The ratio of r1 to r2 is set to 100%. The basic body proportion r1 is compared with the preset body proportion r0. r0 ≥ 65%. The actual situation of the game character is judged based on the comparison result. The public image recognition process is optimized based on the judgment result.

8. A method applied to a machine learning-based gamification creation system as described in any one of claims 1-7, characterized in that, include: Step S1: Acquire source data for creation through a group of sensor nodes; Step S2, the gamification data recognition module, identifies the source data of the creation based on the data recognition model to obtain gamified data and non-gamified data, and also adjusts the acquisition process of the source data of the creation based on the number of non-gamified data from the same source; Step S3 involves creating game characters based on gamification data, monitoring network latency, optimizing the creation process of game characters in real time based on network latency, optimizing the acquisition process of creation source data based on real-time optimization methods, and adjusting the real-time optimization process of game character creation based on device computing power margin. Step S4 involves outputting the game character, identifying the public character based on the number of times the game character is accessed, correcting the output result of the game character based on the public similarity coefficient of the game character, and optimizing the identification process of the public character based on the basic subject ratio of the game character. Step S5: Optimize the acquisition process of source data in real time based on the frequency of gamified data points.

Images