A network digital virtual asset classification automatic management method and system

Abstract

The present invention relates to the field of data management, and in particular to a method and system for automated classification management of network digital virtual assets. An automated classification management system for network digital virtual assets comprises: a network digital virtual asset data acquisition module to be classified, a time domain analysis module, a frequency domain analysis module, a distribution feature extraction module, a discrete feature extraction module and a network digital virtual asset classification module. The present invention can preliminarily classify network digital virtual assets through a built-in random forest model, and determine the feature contribution rate of feature items according to the results of the preliminary classification. The feature contribution rate of different dimensional features of the network digital virtual asset data to be classified can be analyzed through a multi-layer perceptron, and then the network digital virtual asset feature vector is enhanced through a first feature weight vector and a second feature weight vector, which can improve the classification accuracy of network digital virtual assets.

Description

Technical Field

[0001] The present invention relates to the field of data management, and in particular to a method and system for automatic classification management of network digital virtual assets. Background Art

[0002] With the rapid development of the Internet and digital economy, online digital virtual assets (such as digital artworks, virtual game items, etc.) play an increasingly important role in modern society. These digital assets are of various types and forms, which brings huge challenges to management and classification. Traditional manual classification methods are not only time-consuming and labor-intensive, but also prone to errors, and cannot meet the needs of efficient and accurate management.

[0003] Currently, some classification methods based on machine learning and artificial intelligence have been applied to the field of digital asset management. However, the classification accuracy of these methods still needs to be improved when facing complex and changeable digital assets. The lack of accuracy not only affects management efficiency, but may also lead to misclassification of assets, which in turn leads to a series of problems, such as limited asset liquidity, transaction errors, and reduced user trust. Summary of the invention

[0004] The present invention realizes multi-dimensional analysis of the network digital virtual asset data to be classified by performing time domain analysis, frequency domain analysis, distribution feature extraction and discrete feature extraction on the network digital virtual asset data to be classified, and then classifies and identifies the network digital virtual asset data through a network digital virtual asset classification model to realize automatic classification of the network digital virtual asset; in the network digital virtual asset classification model, the network digital virtual asset can be preliminarily classified through a built-in random forest model, and the feature contribution rate of the feature item is determined according to the result of the preliminary classification, and the feature contribution rate of the different dimensional features of the network digital virtual asset data to be classified can be analyzed through a multi-layer perceptron, and then the network digital virtual asset feature vector is feature-enhanced through a first feature weight vector and a second feature weight vector, which can improve the classification accuracy of the subsequent first classification layer, and finally based on the first classification probability vector and the second classification probability vector, a comprehensive analysis is performed to improve the accuracy of the automatic classification of the network digital virtual asset.

[0005] A network digital virtual asset classification and automated management method, comprising:

[0006] Obtaining network digital virtual asset data to be classified;

[0007] Perform time domain analysis on the network digital virtual asset data to be classified and construct the time domain feature vector of the network digital virtual asset;

[0008] The network digital virtual asset data to be classified is converted into the frequency domain through Fourier transform, and the frequency domain analysis is performed to construct the frequency domain feature vector of the network digital virtual asset;

[0009] Extracting distribution features of the network digital virtual asset data to be classified to construct a network digital virtual asset distribution feature vector;

[0010] Extracting discrete features of the network digital virtual asset data to be classified to construct a discrete feature vector of the network digital virtual asset;

[0011] The network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector are spliced ​​to construct a network digital virtual asset feature vector, and then the network digital virtual asset feature vector is sent to the network digital virtual asset classification model for processing, and a classification label is output;

[0012] The network digital virtual asset classification model includes a random forest processing layer, a multi-layer perceptron layer, a feature enhancement layer, a first classification layer and a second classification layer, wherein the random forest processing layer is used to send the network digital virtual asset feature vector into the built-in random forest model for processing to output a first classification probability vector and a first feature weight vector; the multi-layer perceptron layer is used to analyze the weights corresponding to the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to construct a second feature weight vector; the feature enhancement layer is used to enhance the network digital virtual asset feature vector based on the first feature weight vector and the second feature weight vector to construct an enhanced feature vector of the network digital virtual asset; the first classification layer is established based on the BP neural network, and is used to process the enhanced feature vector of the network digital virtual asset and output a second classification probability vector; the second classification layer is used to output a classification label based on the first classification probability vector and the second classification probability vector.

[0013] As a preferred aspect, time domain analysis is performed on the network digital virtual asset data to be classified to construct a time domain feature vector of the network digital virtual asset, which specifically includes the following steps:

[0014] Performing mean calculation, variance calculation and standard deviation calculation on the network digital virtual asset data to be classified, so as to construct the mean value of the network digital virtual asset data, the variance of the network digital virtual asset data and the standard deviation of the network digital virtual asset data;

[0015] Calculating the autocorrelation function of the network digital virtual asset data to be classified to construct an autocorrelation vector of the network digital virtual asset data;

[0016] The network digital virtual asset data mean, network digital virtual asset data variance, network digital virtual asset data standard deviation and network digital virtual asset data autocorrelation vector are spliced ​​to construct the network digital virtual asset time domain feature vector.

[0017] As a preferred aspect, the network digital virtual asset data to be classified is converted into the frequency domain through Fourier transform, and the frequency domain analysis is performed to construct the network digital virtual asset frequency domain feature vector, which specifically includes the following contents:

[0018] The network digital virtual asset data to be classified is converted into frequency domain through Fourier transform to construct a frequency domain sequence of the network digital virtual asset to be classified;

[0019] Calculate the power spectrum density of the frequency domain sequence of the network digital virtual assets to be classified to construct the power spectrum density of the network digital virtual assets;

[0020] Extract frequency features of the frequency domain sequence of the network digital virtual assets to be classified to construct a frequency feature vector of the network digital virtual assets;

[0021] The network digital virtual asset power spectrum density and the network digital virtual asset frequency feature vector are spliced ​​to construct the network digital virtual asset frequency domain feature vector.

[0022] As a preferred aspect, distribution feature extraction is performed on the network digital virtual asset data to be classified to construct a network digital virtual asset distribution feature vector, which specifically includes the following contents:

[0023] Conducting data ratio analysis on the network digital virtual asset data to be classified, so as to construct network digital virtual asset ratio data;

[0024] Performing entropy calculation on the network digital virtual asset data to be classified to construct network digital virtual asset entropy data;

[0025] The network digital virtual asset proportion data and the network digital virtual asset entropy data are combined into a network digital virtual asset distribution feature vector.

[0026] As a preferred aspect, discrete feature extraction is performed on the network digital virtual asset data to be classified to construct a discrete feature vector of the network digital virtual asset, which specifically includes the following steps:

[0027] Counting the number of bit transitions in the network digital virtual asset data to be classified to construct a network digital virtual asset bit transition feature vector;

[0028] Counting the occurrence frequency of the standard bit pattern in the network digital virtual asset data to be classified to construct a network digital virtual asset bit pattern feature vector;

[0029] The network digital virtual asset bit transformation feature vector and the network digital virtual asset bit pattern feature vector are concatenated to construct a network digital virtual asset discrete feature vector.

[0030] As a preferred aspect, the network digital virtual asset feature vector is sent to the network digital virtual asset classification model for processing, and the classification label is output, which specifically includes the following steps:

[0031] In the random forest processing layer, the network digital virtual asset feature vector is sent to the built-in random forest model for processing to output a first classification probability vector, and the classification label corresponding to the maximum probability value in the first classification probability vector is selected as the initial classification label, and then the first feature weight vector corresponding to the initial classification label is obtained from the feature weight library according to the initial classification label. The feature weight library stores the classification labels and their corresponding first feature weight vectors;

[0032] In the multi-layer perceptron layer, the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the network digital virtual asset feature vector are respectively sent to the corresponding multi-layer perceptron for processing, and then the outputs of all the multi-layer perceptrons are spliced ​​to construct a second feature weight vector;

[0033] In the feature enhancement layer, each data in the second feature weight vector is multiplied with the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the corresponding network digital virtual asset feature vector to achieve the first feature enhancement, and then the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the network digital virtual asset feature vector that has undergone the first feature enhancement are spliced ​​and then dot-producted with the first feature weight vector to achieve the second feature enhancement, and the network digital virtual asset enhanced feature vector is output;

[0034] Send the network digital virtual asset enhanced feature vector to the first classification layer for processing, and output the second classification probability vector;

[0035] In the second classification layer, the first classification probability vector and the second classification probability vector are weightedly added to construct a third classification probability vector, and the classification label corresponding to the maximum probability value in the third classification probability vector is selected for output.

[0036] As a preferred aspect, the training of the built-in random forest model specifically includes the following steps:

[0037] Acquire a number of network digital virtual asset classification training samples, label the network digital virtual asset classification training samples by classification labels, and form a network digital virtual asset classification training set with all the labeled network digital virtual asset classification training samples, and then send the network digital virtual asset classification training set to the parameter-initialized built-in random forest model for training, calculate the corresponding first accuracy rate, and determine whether the first accuracy rate is higher than the first preset threshold value. If the first accuracy rate is higher than the first preset threshold value, output the trained built-in random forest model; otherwise, continue to train the built-in random forest model by using the network digital virtual asset classification training set;

[0038] The construction of the feature weight library includes the following contents: obtaining a trained built-in random forest model, performing the following operations for each classification label, selecting network digital virtual asset classification training samples corresponding to the classification label and sending them one by one into the trained built-in random forest model for processing, determining the feature contribution rate of different feature items in each network digital virtual asset classification training sample to the decision path corresponding to the classification label, and then adding and averaging the feature contribution rates of different feature items in all network digital virtual asset classification training samples corresponding to the classification label to construct the first feature weight vector corresponding to the classification label; and forming a feature weight library with all classification labels and their corresponding first feature weight vectors.

[0039] As a preferred aspect, the training of the network digital virtual asset classification model specifically includes the following steps:

[0040] The network digital virtual asset classification training set is sent to the parameter-initialized network digital virtual asset classification model for training. The random forest processing layer in the parameter-initialized network digital virtual asset classification model uses the trained built-in random forest model to calculate the corresponding second accuracy rate, and judge whether the second accuracy rate is higher than the second preset threshold. If the second accuracy rate is higher than the second preset threshold, the trained network digital virtual asset classification model is output; otherwise, the network digital virtual asset classification model continues to be trained through the network digital virtual asset classification training set; and the second preset threshold is higher than the first preset threshold.

[0041] A network digital virtual asset classification automatic management system, comprising:

[0042] A module for acquiring network digital virtual asset data to be classified, used for acquiring network digital virtual asset data to be classified;

[0043] A time domain analysis module is used to perform time domain analysis on the network digital virtual asset data to be classified and construct a time domain feature vector of the network digital virtual asset;

[0044] The frequency domain analysis module is used to convert the network digital virtual asset data to be classified into the frequency domain through Fourier transform, and perform frequency domain analysis to construct the frequency domain feature vector of the network digital virtual asset;

[0045] A distribution feature extraction module is used to extract distribution features of the network digital virtual asset data to be classified, so as to construct a network digital virtual asset distribution feature vector;

[0046] A discrete feature extraction module is used to extract discrete features from the network digital virtual asset data to be classified, so as to construct a discrete feature vector of the network digital virtual asset;

[0047] The network digital virtual asset classification module is used to splice the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to construct a network digital virtual asset feature vector, and then send the network digital virtual asset feature vector into the network digital virtual asset classification model for processing and output a classification label.

[0048] The present invention has the following advantages:

[0049] The present invention realizes multi-dimensional analysis of the network digital virtual asset data to be classified by performing time domain analysis, frequency domain analysis, distribution feature extraction and discrete feature extraction on the network digital virtual asset data to be classified, and then classifies and identifies the network digital virtual asset data through a network digital virtual asset classification model to realize automatic classification of the network digital virtual asset; in the network digital virtual asset classification model, the network digital virtual asset can be preliminarily classified through a built-in random forest model, and the feature contribution rate of the feature item is determined according to the result of the preliminary classification, and the feature contribution rate of the different dimensional features of the network digital virtual asset data to be classified can be analyzed through a multi-layer perceptron, and then the network digital virtual asset feature vector is feature-enhanced through a first feature weight vector and a second feature weight vector, which can improve the classification accuracy of the subsequent first classification layer, and finally based on the first classification probability vector and the second classification probability vector, a comprehensive analysis is performed to improve the accuracy of the automatic classification of the network digital virtual asset. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a schematic diagram of the structure of the network digital virtual asset classification automatic management system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0052] Embodiment 1, a method for automatic classification management of network digital virtual assets, comprising:

[0053] Obtaining the network digital virtual asset data to be classified. It should be noted that the network digital virtual asset data to be classified includes various forms of data that exist in the digital environment and have value, such as virtual currency, equipment and skins in games. The actual storage form of the network digital virtual asset data to be classified in the network is generally a series of binary characters. In the process of executing the management of network digital virtual assets, the classification of network digital virtual assets is extremely important;

[0054] Performing time domain analysis on the network digital virtual asset data to be classified, and constructing a network digital virtual asset time domain feature vector, wherein the network digital virtual asset time domain feature vector specifically includes change information on the time series corresponding to the network digital virtual asset data to be classified;

[0055] The network digital virtual asset data to be classified is converted into the frequency domain through Fourier transform, and the frequency domain analysis is performed to construct the network digital virtual asset frequency domain feature vector. The data in the network digital virtual asset frequency domain feature vector represents the performance of the network digital virtual asset data to be classified in the frequency domain.

[0056] Extracting distribution features of the network digital virtual asset data to be classified to construct a network digital virtual asset distribution feature vector, wherein the network digital virtual asset distribution feature vector represents the distribution of corresponding data in the network digital virtual asset data to be classified;

[0057] Extracting discrete features of the network digital virtual asset data to be classified to construct a discrete feature vector of the network digital virtual asset, wherein the discrete feature vector of the network digital virtual asset can reflect the structure of the network digital virtual asset data to be classified;

[0058] The network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector are spliced ​​to construct the network digital virtual asset feature vector, and then the network digital virtual asset feature vector is sent to the network digital virtual asset classification model for processing, and the classification label is output. The classification label can be virtual goods, virtual real estate, etc., and is set by the developer according to the needs in the training;

[0059] The network digital virtual asset classification model includes a random forest processing layer, a multi-layer perceptron layer, a feature enhancement layer, a first classification layer and a second classification layer, wherein the random forest processing layer is used to send the network digital virtual asset feature vector to the built-in random forest model for processing to output a first classification probability vector and a first feature weight vector, the first classification probability vector stores the probability value that the network digital virtual asset data to be classified belongs to different classification labels, and the first feature weight vector reflects the contribution of different feature items in the network digital virtual asset feature vector to the classification result under the initial classification condition, and is used for subsequent feature enhancement; the multi-layer perceptron layer is used to analyze the weights corresponding to the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to construct the second feature vector. The feature weight vector, the second feature weight vector reflects the feature contribution rate of the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to the classification result, which is used for subsequent feature enhancement. For example, virtual goods are composed of a series of transaction record data, and their corresponding time domain features will have a strong improvement effect on the classification result; the feature enhancement layer is used to enhance the network digital virtual asset feature vector based on the first feature weight vector and the second feature weight vector to construct a network digital virtual asset enhanced feature vector; the first classification layer is established based on the BP neural network, which is used to process the network digital virtual asset enhanced feature vector and output the second classification probability vector; the second classification layer is used to output the classification label based on the first classification probability vector and the second classification probability vector;

[0060] The present application realizes multi-dimensional analysis of the network digital virtual asset data to be classified by performing time domain analysis, frequency domain analysis, distribution feature extraction and discrete feature extraction on the network digital virtual asset data to be classified, and then classifies and identifies the network digital virtual assets through the network digital virtual asset classification model to realize the automatic classification of the network digital virtual assets; in the network digital virtual asset classification model, the network digital virtual assets can be preliminarily classified through the built-in random forest model, and the feature contribution rate of the feature item can be determined according to the result of the preliminary classification, and the feature contribution rate of the different dimensional features of the network digital virtual asset data to be classified can be analyzed through the multi-layer perceptron, and then the network digital virtual asset feature vector is feature-enhanced through the first feature weight vector and the second feature weight vector, which can improve the classification accuracy of the subsequent first classification layer, and finally based on the comprehensive analysis of the first classification probability vector and the second classification probability vector, the accuracy of the current automatic classification of network digital virtual assets can be improved.

[0061] Performing time domain analysis on the network digital virtual asset data to be classified and constructing the network digital virtual asset time domain feature vector specifically includes the following steps:

[0062] Perform mean calculation, variance calculation and standard deviation calculation on the network digital virtual asset data to be classified, so as to construct the mean value of the network digital virtual asset data, the variance of the network digital virtual asset data and the standard deviation of the network digital virtual asset data. It should be noted that the mean calculation, the variance calculation and the standard deviation calculation can all be calculated using built-in functions;

[0063] Performing autocorrelation function calculation on the network digital virtual asset data to be classified to construct an autocorrelation vector of the network digital virtual asset data, wherein the autocorrelation vector of the network digital virtual asset data includes information such as energy value and cycle length corresponding to the network digital virtual asset data to be classified;

[0064] The network digital virtual asset data mean, network digital virtual asset data variance, network digital virtual asset data standard deviation and network digital virtual asset data autocorrelation vector are spliced ​​to construct the network digital virtual asset time domain feature vector.

[0065] The network digital virtual asset data to be classified is converted into the frequency domain through Fourier transform, and the frequency domain analysis is performed to construct the frequency domain feature vector of the network digital virtual asset, which specifically includes the following contents:

[0066] The network digital virtual asset data to be classified is converted into frequency domain through Fourier transform to construct a frequency domain sequence of the network digital virtual asset to be classified;

[0067] The power spectrum density is calculated for the frequency domain sequence of the network digital virtual assets to be classified, so as to construct the power spectrum density of the network digital virtual assets. The power spectrum density of the network digital virtual assets reflects the spectrum characteristics corresponding to the frequency domain sequence of the network digital virtual assets to be classified.

[0068] Extract frequency features of the frequency domain sequence of the network digital virtual asset to be classified to construct a frequency feature vector of the network digital virtual asset, wherein the frequency feature vector of the network digital virtual asset includes various frequency components and corresponding amplitudes;

[0069] The network digital virtual asset power spectrum density and the network digital virtual asset frequency feature vector are spliced ​​to construct the network digital virtual asset frequency domain feature vector.

[0070] The distribution features of the network digital virtual asset data to be classified are extracted to construct the network digital virtual asset distribution feature vector, which specifically includes the following contents:

[0071] Performing data ratio analysis on the network digital virtual asset data to be classified, specifically calculating the ratio of 0 and 1 in the network digital virtual asset data to be classified, so as to construct network digital virtual asset ratio data;

[0072] Entropy calculation is performed on the network digital virtual asset data to be classified, specifically the proportion of 0 and 1 in the network digital virtual asset data to be classified, and the information entropy corresponding to 0 and 1 is calculated by the entropy calculation formula to construct the network digital virtual asset entropy data;

[0073] The network digital virtual asset proportion data and the network digital virtual asset entropy data are combined into a network digital virtual asset distribution feature vector.

[0074] The discrete feature extraction is performed on the network digital virtual asset data to be classified to construct a discrete feature vector of the network digital virtual asset, which specifically includes the following steps:

[0075] Counting the number of bit transitions in the network digital virtual asset data to be classified, that is, the number of transitions from 0 to 1 and from 1 to 0, to construct a network digital virtual asset bit transition feature vector;

[0076] Count the frequency of occurrence of standard bit patterns in the network digital virtual asset data to be classified, such as the frequency of occurrence of "001". The standard bit pattern is determined in advance by the developer to construct a network digital virtual asset bit pattern feature vector;

[0077] The network digital virtual asset bit transformation feature vector and the network digital virtual asset bit pattern feature vector are concatenated to construct a network digital virtual asset discrete feature vector.

[0078] The network digital virtual asset feature vector is sent to the network digital virtual asset classification model for processing, and the classification label is output, which specifically includes the following steps:

[0079] In the random forest processing layer, the network digital virtual asset feature vector is sent to the built-in random forest model for processing to output a first classification probability vector, which stores the probability values ​​of the network digital virtual asset feature vector belonging to different leaf nodes in the built-in random forest model. The classification label corresponding to the maximum probability value in the first classification probability vector is selected as the initial classification label, and then the first feature weight vector corresponding to the initial classification label is obtained from the feature weight library according to the initial classification label. The feature weight library stores the classification label and its corresponding first feature weight vector;

[0080] In the multi-layer perceptron layer, the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the network digital virtual asset feature vector are respectively sent to the corresponding multi-layer perceptron for processing, and then the outputs of all the multi-layer perceptrons are spliced ​​to construct a second feature weight vector, and different multi-layer perceptrons are set in the form of the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector;

[0081] In the feature enhancement layer, each data in the second feature weight vector is multiplied with the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the corresponding network digital virtual asset feature vector to achieve the first feature enhancement, and then the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the network digital virtual asset feature vector that has undergone the first feature enhancement are spliced ​​and then dot-producted with the first feature weight vector to achieve the second feature enhancement, and the network digital virtual asset enhanced feature vector is output;

[0082] Sending the network digital virtual asset enhanced feature vector to the first classification layer for processing, outputting a second classification probability vector, wherein the specific setting of the first classification layer is set based on the scale of the network digital virtual asset enhanced feature vector and the existing technology;

[0083] In the second classification layer, the first classification probability vector and the second classification probability vector are weightedly added, that is, the first classification probability vector and the second classification probability vector are multiplied by the corresponding weights respectively and then added. The weights corresponding to the first classification probability vector and the second classification probability vector are determined in advance by the developer to construct the third classification probability vector, and the classification label corresponding to the maximum probability value in the third classification probability vector is selected for output.

[0084] The training of the built-in random forest model includes the following steps:

[0085] Obtain several network digital virtual asset classification training samples, which include network digital virtual asset feature vectors obtained by the restaurant staff based on the network digital virtual asset data in the historical records, label the network digital virtual asset classification training samples through classification labels, and form a network digital virtual asset classification training set with all labeled network digital virtual asset classification training samples, and then send the network digital virtual asset classification training set to the parameter-initialized built-in random forest model for training, calculate the corresponding first accuracy rate, and determine whether the first accuracy rate is higher than a first preset threshold value, which is determined by the developer. If the first accuracy rate is higher than the first preset threshold value, output the trained built-in random forest model; otherwise, continue to train the built-in random forest model through the network digital virtual asset classification training set;

[0086] The construction of the feature weight library includes the following contents: obtaining a trained built-in random forest model, performing the following operations for each classification label, selecting network digital virtual asset classification training samples corresponding to the classification label and sending them one by one into the trained built-in random forest model for processing, determining the feature contribution rate of different feature items in each network digital virtual asset classification training sample to the decision path corresponding to the classification label, and then adding and averaging the feature contribution rates of different feature items in all network digital virtual asset classification training samples corresponding to the classification label to construct the first feature weight vector corresponding to the classification label; and forming a feature weight library with all classification labels and their corresponding first feature weight vectors.

[0087] The training of the network digital virtual asset classification model specifically includes the following steps:

[0088] The network digital virtual asset classification training set is sent to the parameter-initialized network digital virtual asset classification model for training. The random forest processing layer in the parameter-initialized network digital virtual asset classification model adopts the trained built-in random forest model to calculate the corresponding second accuracy rate, and judge whether the second accuracy rate is higher than the second preset threshold. The second preset threshold is determined by the developer. If the second accuracy rate is higher than the second preset threshold, the trained network digital virtual asset classification model is output; otherwise, the network digital virtual asset classification model continues to be trained through the network digital virtual asset classification training set; and the second preset threshold is higher than the first preset threshold.

[0089] Embodiment 2, a network digital virtual asset classification automatic management system, such as Figure 1 As shown, including:

[0090] The module for acquiring network digital virtual asset data to be classified is used to acquire network digital virtual asset data to be classified. It should be noted that the network digital virtual asset data to be classified includes various forms of data that exist in the digital environment and have value, such as virtual currency, equipment and skin in the game. The actual storage form of the network digital virtual asset data to be classified in the network is generally a series of binary characters. In the process of executing the management of network digital virtual assets, the classification of network digital virtual assets is extremely important.

[0091] A time domain analysis module is used to perform time domain analysis on the network digital virtual asset data to be classified, and construct a network digital virtual asset time domain feature vector, wherein the network digital virtual asset time domain feature vector specifically includes change information on the time series corresponding to the network digital virtual asset data to be classified;

[0092] The frequency domain analysis module is used to convert the network digital virtual asset data to be classified into the frequency domain through Fourier transform, and perform frequency domain analysis to construct the network digital virtual asset frequency domain feature vector. The data in the network digital virtual asset frequency domain feature vector represents the performance of the network digital virtual asset data to be classified in the frequency domain.

[0093] A distribution feature extraction module is used to extract distribution features of the network digital virtual asset data to be classified, so as to construct a network digital virtual asset distribution feature vector, wherein the network digital virtual asset distribution feature vector represents the distribution of corresponding data in the network digital virtual asset data to be classified;

[0094] A discrete feature extraction module is used to extract discrete features from the network digital virtual asset data to be classified, so as to construct a discrete feature vector of the network digital virtual asset. The discrete feature vector of the network digital virtual asset can reflect the structure of the network digital virtual asset data to be classified.

[0095] The network digital virtual asset classification module is used to splice the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to construct a network digital virtual asset feature vector, and then send the network digital virtual asset feature vector to the network digital virtual asset classification model for processing, and output classification labels. The classification labels can be virtual commodities, virtual real estate, etc., which are set by developers during training according to needs.

[0096] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A network digital virtual asset classification and automated management method, characterized in that: include: Obtaining network digital virtual asset data to be classified; Perform time domain analysis on the network digital virtual asset data to be classified and construct the time domain feature vector of the network digital virtual asset; The network digital virtual asset data to be classified is converted into the frequency domain through Fourier transform, and the frequency domain analysis is performed to construct the frequency domain feature vector of the network digital virtual asset; Extracting distribution features of the network digital virtual asset data to be classified to construct a network digital virtual asset distribution feature vector; Extracting discrete features of the network digital virtual asset data to be classified to construct a discrete feature vector of the network digital virtual asset; The network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector are spliced ​​to construct a network digital virtual asset feature vector, and then the network digital virtual asset feature vector is sent to the network digital virtual asset classification model for processing, and a classification label is output; The network digital virtual asset classification model includes a random forest processing layer, a multi-layer perceptron layer, a feature enhancement layer, a first classification layer and a second classification layer, wherein the random forest processing layer is used to send the network digital virtual asset feature vector into the built-in random forest model for processing to output a first classification probability vector and a first feature weight vector; the multi-layer perceptron layer is used to analyze the weights corresponding to the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to construct a second feature weight vector; the feature enhancement layer is used to enhance the network digital virtual asset feature vector based on the first feature weight vector and the second feature weight vector to construct a network digital virtual asset enhanced feature vector; the first classification layer is established based on a BP neural network, and is used to process the network digital virtual asset enhanced feature vector and output a second classification probability vector; the second classification layer is used to output a classification label based on the first classification probability vector and the second classification probability vector; Performing time domain analysis on the network digital virtual asset data to be classified and constructing the network digital virtual asset time domain feature vector specifically includes the following steps: Performing mean calculation, variance calculation and standard deviation calculation on the network digital virtual asset data to be classified, so as to construct the mean value of the network digital virtual asset data, the variance of the network digital virtual asset data and the standard deviation of the network digital virtual asset data; Calculating the autocorrelation function of the network digital virtual asset data to be classified to construct an autocorrelation vector of the network digital virtual asset data; The network digital virtual asset data mean, the network digital virtual asset data variance, the network digital virtual asset data standard deviation and the network digital virtual asset data autocorrelation vector are spliced ​​to construct the network digital virtual asset time domain feature vector; The network digital virtual asset data to be classified is converted into the frequency domain through Fourier transform, and the frequency domain analysis is performed to construct the frequency domain feature vector of the network digital virtual asset, which specifically includes the following contents: The network digital virtual asset data to be classified is converted into frequency domain through Fourier transform to construct a frequency domain sequence of the network digital virtual asset to be classified; Calculate the power spectrum density of the frequency domain sequence of the network digital virtual assets to be classified to construct the power spectrum density of the network digital virtual assets; Extract frequency features of the frequency domain sequence of the network digital virtual assets to be classified to construct a frequency feature vector of the network digital virtual assets; The network digital virtual asset power spectrum density and the network digital virtual asset frequency feature vector are concatenated to construct the network digital virtual asset frequency domain feature vector; The distribution features of the network digital virtual asset data to be classified are extracted to construct the network digital virtual asset distribution feature vector, which specifically includes the following contents: Conducting data ratio analysis on the network digital virtual asset data to be classified, so as to construct network digital virtual asset ratio data; Performing entropy calculation on the network digital virtual asset data to be classified to construct network digital virtual asset entropy data; Combining the network digital virtual asset proportion data and the network digital virtual asset entropy data into a network digital virtual asset distribution feature vector; The discrete feature extraction is performed on the network digital virtual asset data to be classified to construct a discrete feature vector of the network digital virtual asset, which specifically includes the following steps: Counting the number of bit transitions in the network digital virtual asset data to be classified to construct a network digital virtual asset bit transition feature vector; Counting the occurrence frequency of the standard bit pattern in the network digital virtual asset data to be classified to construct a network digital virtual asset bit pattern feature vector; The network digital virtual asset bit transformation feature vector and the network digital virtual asset bit pattern feature vector are concatenated to construct a network digital virtual asset discrete feature vector.

2. A network digital virtual asset classification and automated management method according to claim 1, characterized in that: The network digital virtual asset feature vector is sent to the network digital virtual asset classification model for processing, and the classification label is output, which specifically includes the following steps: In the random forest processing layer, the network digital virtual asset feature vector is sent to the built-in random forest model for processing to output a first classification probability vector, and the classification label corresponding to the maximum probability value in the first classification probability vector is selected as the initial classification label, and then the first feature weight vector corresponding to the initial classification label is obtained from the feature weight library according to the initial classification label. The feature weight library stores the classification labels and their corresponding first feature weight vectors; In the multi-layer perceptron layer, the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the network digital virtual asset feature vector are respectively sent to the corresponding multi-layer perceptron for processing, and then the outputs of all the multi-layer perceptrons are spliced ​​to construct a second feature weight vector; In the feature enhancement layer, each data in the second feature weight vector is multiplied with the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the corresponding network digital virtual asset feature vector to achieve the first feature enhancement, and then the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector in the network digital virtual asset feature vector that has undergone the first feature enhancement are spliced ​​and then dot-producted with the first feature weight vector to achieve the second feature enhancement, and the network digital virtual asset enhanced feature vector is output; Send the network digital virtual asset enhanced feature vector to the first classification layer for processing, and output the second classification probability vector; In the second classification layer, the first classification probability vector and the second classification probability vector are weightedly added to construct a third classification probability vector, and the classification label corresponding to the maximum probability value in the third classification probability vector is selected for output.

3. A network digital virtual asset classification and automated management method according to claim 2, characterized in that: The training of the built-in random forest model includes the following steps: Acquire a number of network digital virtual asset classification training samples, label the network digital virtual asset classification training samples by classification labels, and form a network digital virtual asset classification training set with all the labeled network digital virtual asset classification training samples, and then send the network digital virtual asset classification training set to the parameter-initialized built-in random forest model for training, calculate the corresponding first accuracy rate, and determine whether the first accuracy rate is higher than the first preset threshold value. If the first accuracy rate is higher than the first preset threshold value, output the trained built-in random forest model; otherwise, continue to train the built-in random forest model by using the network digital virtual asset classification training set; The construction of the feature weight library includes the following contents: obtaining a trained built-in random forest model, performing the following operations for each classification label, selecting network digital virtual asset classification training samples corresponding to the classification label and sending them one by one into the trained built-in random forest model for processing, determining the feature contribution rate of different feature items in each network digital virtual asset classification training sample to the decision path corresponding to the classification label, and then adding and averaging the feature contribution rates of different feature items in all network digital virtual asset classification training samples corresponding to the classification label to construct the first feature weight vector corresponding to the classification label; and forming a feature weight library with all classification labels and their corresponding first feature weight vectors.

4. A network digital virtual asset classification and automated management method according to claim 3, characterized in that: The training of the network digital virtual asset classification model specifically includes the following steps: The network digital virtual asset classification training set is sent to the parameter-initialized network digital virtual asset classification model for training. The random forest processing layer in the parameter-initialized network digital virtual asset classification model uses the trained built-in random forest model to calculate the corresponding second accuracy rate, and judge whether the second accuracy rate is higher than the second preset threshold. If the second accuracy rate is higher than the second preset threshold, the trained network digital virtual asset classification model is output; otherwise, the network digital virtual asset classification model continues to be trained through the network digital virtual asset classification training set; and the second preset threshold is higher than the first preset threshold.

5. A network digital virtual asset classification automatic management system, characterized in that: The system applies a network digital virtual asset classification automatic management method as described in any one of claims 1 to 4, including: A module for acquiring network digital virtual asset data to be classified, used for acquiring network digital virtual asset data to be classified; A time domain analysis module is used to perform time domain analysis on the network digital virtual asset data to be classified and construct a time domain feature vector of the network digital virtual asset; The frequency domain analysis module is used to convert the network digital virtual asset data to be classified into the frequency domain through Fourier transform, and perform frequency domain analysis to construct the frequency domain feature vector of the network digital virtual asset; A distribution feature extraction module is used to extract distribution features of the network digital virtual asset data to be classified to construct a network digital virtual asset distribution feature vector; A discrete feature extraction module is used to extract discrete features from the network digital virtual asset data to be classified, so as to construct a discrete feature vector of the network digital virtual asset; The network digital virtual asset classification module is used to splice the network digital virtual asset time domain feature vector, the network digital virtual asset frequency domain feature vector, the network digital virtual asset distribution feature vector and the network digital virtual asset discrete feature vector to construct a network digital virtual asset feature vector, and then send the network digital virtual asset feature vector into the network digital virtual asset classification model for processing and output a classification label.

Images