Standard machine readable capability evaluation method

Abstract

The invention discloses a standard machine readable capability evaluation method, which comprises the following steps of: respectively converting the same instruction into structured data and unstructured data, and transmitting the structured data and the unstructured data to a machine input end; obtaining byte information according to the structured data, and obtaining feature information according to the unstructured data; the byte information and the feature information are clustered based on action ranges, and a first transaction unit set and a second transaction unit set are obtained; selecting a typical label in the same transaction unit and adding an index; obtaining a response instruction at the machine execution end, analyzing the response instruction to the index for difference calculation, and obtaining a deviation coefficient; and analyzing the weight distribution of the deviation coefficient, constructing an evaluation model through a neural network based on the weight distribution, and evaluating the readable capability of the machine. The method not only can effectively evaluate the processing capability of the machine for different types of data, but also has good interpretability, and can adapt to various complex machine readable capability evaluation requirements.

Description

Technical Field

[0001] The present invention relates to the field of machine-readable standards, and in particular to a method for evaluating the machine-readable capability of standards. Background Art

[0002] In today's era of rapid digital development, the demand for data processing and utilization in various fields is growing. With the widespread application of machines in various industries, machine readability has become an important indicator to measure machine performance.

[0003] Traditional standard evaluation methods often focus on evaluating a single performance indicator of a machine, such as processing speed, storage capacity, etc., but lack comprehensive and effective evaluation methods for the machine's readability when processing different types of data. In practical applications, machines need to process multiple forms of data, such as structured data and unstructured data. These data come from a wide range of sources and in various forms, which poses a huge challenge to the machine's readability.

[0004] For structured data, traditional methods may process it through simple data format conversion and parsing, but for complex structured data, this approach may not be able to fully mine the key information in the data. For unstructured data, such as text files, image files, audio files, and video files, traditional technologies usually have difficulty in effectively extracting feature information and accurately analyzing and processing it.

[0005] In summary, existing technologies are insufficient in evaluating machine readability, and a new standard machine readability evaluation method is needed to meet the growing needs of digitalization. Summary of the invention

[0006] The object of the present invention is to provide a standard machine readability evaluation method.

[0007] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0008] A first aspect of the present invention provides a method for evaluating standard machine readability, comprising the following steps:

[0009] S100 converts the same instruction into structured data and unstructured data respectively and transmits them to the machine input end;

[0010] S200 reads data from the structured data in a buffered byte stream manner, performs an alienation operation on the read byte data to obtain byte information; performs feature conversion on the unstructured data to obtain feature information;

[0011] S300 clusters the byte information and the feature information based on the scope, aggregates the information in the same cluster into the same transaction unit, and obtains a first transaction unit set and a second transaction unit set;

[0012] S400 selects a typical tag in the same transaction unit and adds an index to the typical tag;

[0013] S500 obtains a response instruction at the machine execution end, parses the response instruction to the index, performs difference calculation with the first transaction unit set and the second transaction unit set, and obtains a deviation coefficient;

[0014] S600 analyzes the weight distribution of the deviation coefficient, constructs an evaluation model through a neural network based on the weight distribution, and evaluates the machine readability.

[0015] As a further method, the method of converting the same instruction into structured data and unstructured data respectively includes:

[0016] Use graph neural networks to learn the node features of instructions and identify key information points in instructions;

[0017] By assigning a unique key to each key information point through the key-value pair method, structured data is obtained;

[0018] Convert key information points into other forms for storage, including but not limited to using text files, image files, audio files and video files to store instructions and obtain unstructured data.

[0019] As a further method, the method of reading data from the structured data in a buffered byte stream manner and performing an alienation operation on the read bytes includes:

[0020] Determine the storage location of the structured data. If it is a file, obtain the file path; if it is a network data source, determine the network address and port;

[0021] Select the input stream based on the data storage type of the structured data. If it is a file, create a file input stream based on the file path; if it is a network data source, create a network input stream based on the network address;

[0022] Wrap the input stream in a buffered input stream and use the read() method of the buffered byte stream to read the byte data byte by byte;

[0023] Perform an XOR operation on each bit of the byte data with its adjacent bits.

[0024] As a further method, the method for feature selection of the unstructured data includes: for text files, using a bag-of-words model to extract TF-IDF feature vectors as feature information; for image files, using a convolutional neural network to extract local feature descriptors and global features of the image as feature information; for audio files, using short-time Fourier transform to extract time series features and frequency domain features from the audio files as feature information; for video files, using a convolutional neural network to extract key frame features as feature vectors.

[0025] As a further method, the method of clustering the byte information and the feature information based on scopes, and aggregating the information in the same cluster into the same transaction unit includes:

[0026] Obtain log records, and determine the scope of byte information and feature information within the machine based on the log records;

[0027] The byte information and feature information are clustered based on the scope using the Gaussian mixture model, and each data point is assigned to the cluster with the highest probability of Gaussian distribution. The formula for determining the cluster to which each data point belongs is:

[0028]

[0029] Where k is the index of the scope, C k is the kth scope, j is C k The index of the Gaussian distribution, d is the dimension of the byte information, ∑ k,j Represented as in scope C k The covariance matrix of the jth Gaussian distribution, x i is the i-th data point, μ k,j For scope C k The mean vector of the j-th Gaussian distribution, T represents the transposition operation, Expressed as ∑ k,j The inverse of the covariance matrix, π k,j For scope C k The mixing coefficient of the jth Gaussian distribution, P(C k ) indicates that the data point belongs to scope C k The prior probability of , N is the total number of data points;

[0030] Aggregate data points in byte information of the same cluster into the same transaction unit, and obtain a byte information transaction unit set as a first transaction unit set; aggregate data points in feature information of the same cluster into the same transaction unit, and obtain a feature information transaction unit set as a second transaction unit set.

[0031] As a further method, the method of selecting typical features within the same transaction unit includes:

[0032] The information entropy increase of data points in the same transaction unit is calculated, and the data points with information entropy increase in the top 30% are selected as typical labels. The expression for calculating the information entropy increase of data points is:

[0033]

[0034] Where H(D) is the entropy of the transaction unit dataset D, A is the feature standard, H(D|A) is the conditional entropy of the transaction unit dataset D under the feature standard A, k is the number of different categories in the transaction unit dataset D, C i is the i-th category in the unit dataset D, |C i | belongs to category C i The number of data points, n is the number of different values ​​of feature standard A, D j The characteristic standard A is expressed as a j The corresponding data set is a j is the jth value of feature standard A, |D j | is a subset D j The number of samples in , |D| is the number of data points in the transaction unit dataset D.

[0035] As a further method, the method of parsing the response instruction to the index and performing difference calculation between the first transaction unit set and the second transaction unit set includes:

[0036] Extract feature information related to typical tags in the response instruction, and convert the extracted relevant feature information into a form that can be matched with the typical tag index through a dictionary mapping method;

[0037] At the same index, the difference between the response instruction and the first transaction unit set is calculated as the first deviation, and the difference between the response instruction and the second transaction unit set is calculated as the second deviation;

[0038] If there are typical labels at relevant positions between the first transaction unit set and the second transaction unit set, the difference between the first transaction unit set and the second transaction unit set at the same index is calculated as the third deviation;

[0039] The first deviation, the second deviation, and the third deviation are taken as deviation coefficients.

[0040] As a further method, the method of analyzing the weight distribution of the deviation coefficient and constructing an evaluation model through a neural network based on the weight distribution includes:

[0041] Assigning weights based on the historical processing frequency of structured data and unstructured data by the machine, assigning a weight to the first deviation that is in the same proportion as the historical processing frequency of structured data, and assigning a weight to the second deviation that is in the same proportion as the historical processing frequency of unstructured data;

[0042] A penalty term is constructed based on the third deviation, where the larger the value of the third deviation, the larger the value of the penalty term. The expression of the penalty term is:

[0043]

[0044] Among them, A is the amplitude of the penalty term, B is the control parameter of the growth rate of the exponential part, d3 is the third deviation, α is the power influence parameter of d3 in the exponential part, C and D are the coefficients for adjusting the contribution degree and frequency characteristics of the sine function part to the penalty term, β is the power influence parameter of d3 in the sine function part, E is the power function amplitude coefficient, and γ is the power influence parameter of d3 in the power function part;

[0045] Based on weight distribution, the evaluation model is constructed through a convolutional neural network, including: input layer, convolution layer, logic layer, pooling layer, and output layer. Specifically:

[0046] The first layer is the input layer, which is used to receive input data, including the first deviation, the second deviation and the penalty term;

[0047] The second layer is the convolutional layer, which learns local patterns in the data by convolution and feature extraction of the input data;

[0048] The third layer is the logic layer, which is used to process the output of the convolutional layer and learn the high-level features of the input data;

[0049] The fourth layer is the pooling layer, which uses average pooling to extract key information from the features of the input layer;

[0050] The fifth layer is the output layer, which consists of a single neuron and is used to represent continuous evaluation scores. A linear activation function is selected.

[0051] In a second aspect, an embodiment of the present application further provides an electronic device, comprising: a processor; and a memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor executes the method steps described in the first aspect.

[0052] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple applications, the electronic device executes the method steps described in the first aspect.

[0053] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0054] (1) The present invention can examine the machine's ability to process numerical values ​​and logical relationships by processing structured data, and can focus on the machine's ability to understand text semantics and extract features such as images and audio by processing unstructured data, and can reflect the machine's comprehensive performance when processing different types of data.

[0055] (2) The present invention comprehensively considers multiple factors of the machine in the process of processing data through steps such as clustering of byte information and feature information, selection of typical labels, difference calculation and weight allocation, which helps to more accurately evaluate the machine's readability and provide comprehensive guidance for optimizing machine performance.

[0056] (3) The present invention uses the deviation coefficient to perform standardized evaluation on the machine, which can not only reflect the readability of structured data and unstructured data, but also reflect the difference comparison between the two. By analyzing the weight distribution of the deviation coefficient, the weight of the deviation coefficient distribution is based on the historical processing frequency. This makes the evaluation result more objective and also provides users with a clear direction for improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The present invention is a flowchart of the steps of a standard machine readability evaluation method.

[0058] Figure 2 It is a schematic diagram of the structure of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] Reference Figure 1 As shown, the present invention provides a standard machine readability evaluation method, comprising:

[0061] S100 converts the same instruction into structured data and unstructured data respectively and transmits them to the machine input end;

[0062] It needs to be explained that different types of data have different advantages and uses when processing complex information and tasks. Converting instructions into structured and unstructured data allows machines to understand and process instructions from multiple perspectives. It is an important indicator of machine readability and reflects the comprehensive processing capabilities of the machine.

[0063] It should be understood that unstructured data needs to be determined based on the machine's recognizable capabilities. If the machine can recognize text data, then the unstructured data includes text files. Similarly, if the machine can recognize other types of data, then the unstructured data includes files of the corresponding types.

[0064] In the actual evaluation, for the intelligent data analysis platform, the instruction is "Count the project completion status of each department of the company in the past quarter, find the department with the highest completion efficiency, and make a detailed report." The graph neural network is used to identify the key information points as "past quarter", "company departments", "project completion status", "department with the highest completion efficiency", and "detailed report". The key-value pairs are assigned to obtain structured data: {"Time range": "past quarter", "data source": "project data of each department of the company", "analysis target": "department with the highest completion efficiency", "output form": "detailed report"}; converted to unstructured data: the text file storage instruction is "Count the project completion status of each department of the company in the past quarter, find the department with the highest completion efficiency, and make a detailed report.txt".

[0065] S200 reads data from the structured data in a buffered byte stream manner, performs an alienation operation on the read byte data to obtain byte information; performs feature conversion on the unstructured data to obtain feature information;

[0066] In the actual evaluation, structured data is stored as a local file with the file path " / data / structured_data.txt". The file path is obtained, and a file input stream is created based on the file path. The input stream is wrapped in a buffered input stream, and the byte data is read byte by byte. The read byte data is [110, 105, 116, 114, 117, 99, 116, 117, 114, 101, 100, 95, 100, 97, 116, 97, 46, 116, 120, 116]. The byte data is XORed to obtain the alienated byte data. For text files, the bag-of-words model is used to extract TF-IDF feature vectors. The words that appear in the text files are "statistics", "past quarter", "company", "various departments", "project completion status", "highest completion efficiency", "department", and "detailed report". The feature vectors [0.15, 0.2, 0.12, 0.18, 0.1, 0.16, 0.11, 0.18] are obtained after calculation.

[0067] S300 clusters the byte information and the feature information based on the scope, aggregates the information in the same cluster into the same transaction unit, and obtains a first transaction unit set and a second transaction unit set;

[0068] It should be explained that the purpose is to effectively classify and integrate byte information and feature information in order to better understand and process data. Aggregating information in the same cluster into the same transaction unit means treating data points that are divided into the same group during the clustering process as a transaction unit. This can integrate information with similar characteristics and functions, facilitating subsequent analysis and processing. The transaction units in these sets correspond to different scopes and data characteristics, providing a clear data structure and analysis basis for subsequent steps.

[0069] It should be understood that scope refers to the specific functional range or impact area of ​​data information within the machine, and different data information may play a role in different scopes. For example, in a logistics management system, byte information and feature information may form different scopes based on different aspects such as cargo type, transportation route, time requirements, etc.

[0070] In the actual evaluation, analysis of log records revealed that byte information plays a role in the data analysis module, and feature information of unstructured data plays a role in the visualization module and report generation module, thereby determining three scopes. For byte information and feature information, Gaussian mixture models are used for clustering under different scopes. By comparing the probabilities of data points under different Gaussian distributions, each data point is assigned to the cluster where the Gaussian distribution with the highest probability is located. Data points in byte information of the same cluster are aggregated into the same transaction unit, and a byte information transaction unit set is obtained as a first transaction unit set. Data points in feature information of the same cluster are aggregated into the same transaction unit, and a feature information transaction unit set is obtained as a second transaction unit set.

[0071] S400 selects a typical tag in the same transaction unit and adds an index to the typical tag;

[0072] It should be explained that the purpose of selecting typical tags within the same transaction unit is to extract representative features from numerous data points in order to better understand and analyze the data. Typical tags can represent the main features or trends within the transaction unit, which helps to compare and compare data; the purpose of adding indexes to typical tags is to facilitate the identification, management and query of tags. The index can be used as a unique identifier for the tag, so that specific tags can be quickly located and accessed in subsequent processing steps.

[0073] In the actual evaluation, the information entropy increase of data points in the same transaction unit is calculated, and the data points with information entropy increase in the top 30% are selected as typical labels, and indexes are added to the typical labels.

[0074] S500 obtains a response instruction at the machine execution end, parses the response instruction to the index, performs difference calculation with the first transaction unit set and the second transaction unit set, and obtains a deviation coefficient;

[0075] It should be explained that parsing the reply instruction to the index means matching the key information in the reply instruction with the index in the previously established transaction unit set, and performing difference calculation with the first transaction unit set and the second transaction unit set, in order to determine the degree of difference between the reply instruction and the existing transaction units, wherein the deviation coefficient includes a first deviation of the difference between the reply instruction and the first transaction unit set at the same index, and a second deviation of the difference between the reply instruction and the second transaction unit set; if there are typical labels at relevant positions between the first transaction unit set and the second transaction unit set, the difference between the first transaction unit set and the second transaction unit set is calculated at the same index as the third deviation;

[0076] It should be understood that the deviation coefficient reflects the degree of deviation between the response instruction and the existing transaction unit, and also reflects the machine's processing and readability of structured and unstructured data. Taking the first deviation as an example, if it is small, it means that the response instruction is relatively close to the existing transaction unit, and the machine's readability is good; on the contrary, if it is large, it means that the response instruction is quite different from the existing transaction unit, and the machine may have certain difficulties in processing the response instruction, and the readability is relatively weak. However, the weight of the deviation coefficient needs to be determined by the machine's usage frequency;

[0077] In the actual evaluation, the response instruction is "the completion status of the projects of various departments of the company in the past quarter has been counted. The department with the highest completion efficiency is the R&D department. The detailed report has been prepared." The relevant feature information is extracted, which is "R&D department", "highest completion efficiency" and "detailed report". The feature information is converted into a form that can be matched with a typical label index through dictionary mapping. At the same index, the first deviation of the difference between the response instruction and the first transaction unit set is calculated, and the second deviation of the difference between the response instruction and the second transaction unit set is calculated. If there are typical labels at relevant positions between the first transaction unit set and the second transaction unit set, the difference between the first transaction unit set and the second transaction unit set is calculated at the same index as the third deviation;

[0078] S600 analyzes the weight distribution of the deviation coefficient, constructs an evaluation model through a neural network based on the weight distribution, and evaluates the machine readability.

[0079] In the actual evaluation, the intelligent data analysis platform has a historical processing frequency of 70% for structured data and a historical processing frequency of 30% for unstructured data. It assigns 70% weight to the first deviation and 30% weight to the second deviation. Based on the third deviation, it calculates the penalty term and builds an evaluation model through a convolutional neural network: the input layer receives the first deviation, the second deviation and the penalty term. The first deviation is [0.2, 0.1, 0.3], the second deviation is [0.15, 0.22, 0.18], and the penalty term is 0.35. A 3x3 convolution kernel is used to perform a convolution operation on the input data to obtain a new feature map. A fully connected layer is used to process the output of the convolution layer to further extract high-level features of the data. The average pooling method is used to extract key information from the features of the input layer and reduce the dimension of the data. The output layer is a single neuron and a linear activation function is used to output continuous evaluation scores.

[0080] In this embodiment, the output evaluation score is 0.8, indicating the machine's evaluation result of the readability of the instruction;

[0081] In another specific embodiment, a batch of 5 machines of the same model need to be evaluated for readability in a standardized manner, numbered M1 to M5, specifically:

[0082] M1 machine scoring: the first deviation is [0.1, 0.08, 0.12], the second deviation is [0.09, 0.1, 0.08], and the third deviation is 0.04. According to the weight distribution, the weight of structured data is 60%, and the weight of unstructured data is 40%. After building the evaluation model, the evaluation score is 0.82;

[0083] M2 machine score: the first deviation is [0.09, 0.07, 0.11], the second deviation is [0.08, 0.09, 0.07], the third deviation is 0.03, and the evaluation score is 0.85;

[0084] M3 machine score: the first deviation is [0.11, 0.09, 0.13], the second deviation is [0.1, 0.11, 0.09], the third deviation is 0.04, and the evaluation score is 0.80;

[0085] M4 machine score: the first deviation is [0.08, 0.07, 0.1], the second deviation is [0.07, 0.08, 0.06], the third deviation is 0.02, and the evaluation score is 0.88;

[0086] M5 machine scoring: the first deviation is [0.12, 0.1, 0.14], the second deviation is [0.11, 0.12, 0.1], the third deviation is 0.05, and the evaluation score is 0.78.

[0087] In this embodiment, the method of converting the same instruction into structured data and unstructured data respectively includes:

[0088] Use graph neural networks to learn the node features of instructions and identify key information points in instructions;

[0089] By assigning a unique key to each key information point through the key-value pair method, structured data is obtained;

[0090] Convert key information points into other forms for storage, including but not limited to using text files, image files, audio files and video files to store instructions and obtain unstructured data.

[0091] In this embodiment, the method of reading data from the structured data in a buffered byte stream manner and performing an alienation operation on the read bytes includes:

[0092] Determine the storage location of the structured data. If it is a file, obtain the file path; if it is a network data source, determine the network address and port;

[0093] Select the input stream based on the data storage type of the structured data. If it is a file, create a file input stream based on the file path; if it is a network data source, create a network input stream based on the network address;

[0094] Wrap the input stream in a buffered input stream and use the read() method of the buffered byte stream to read the byte data byte by byte;

[0095] Perform an XOR operation on each bit of the byte data with its adjacent bits.

[0096] In this embodiment, the method for feature selection of the unstructured data includes: for text files, using a bag-of-words model to extract TF-IDF feature vectors as feature information; for image files, using a convolutional neural network to extract local feature descriptors and global features of the image as feature information; for audio files, using short-time Fourier transform to extract time series features and frequency domain features from the audio files as feature information; for video files, using a convolutional neural network to extract key frame features as feature vectors.

[0097] In this embodiment, the method of clustering the byte information and the feature information based on the scopes and aggregating the information in the same cluster into the same transaction unit includes:

[0098] Obtain log records, and determine the scope of byte information and feature information within the machine based on the log records;

[0099] The byte information and feature information are clustered based on the scope using the Gaussian mixture model, and each data point is assigned to the cluster with the highest probability of Gaussian distribution. The formula for determining the cluster to which each data point belongs is:

[0100]

[0101] Where k is the index of the scope, C k is the kth scope, j is C k The index of the Gaussian distribution, d is the dimension of the byte information, ∑ k,j Represented as in scope C k The covariance matrix of the jth Gaussian distribution, x i is the i-th data point, μ k,j For scope C k The mean vector of the j-th Gaussian distribution, T represents the transposition operation, Expressed as ∑ k,j The inverse of the covariance matrix, π k,j For scope C k The mixing coefficient of the jth Gaussian distribution, P(C k ) indicates that the data point belongs to scope C k The prior probability of , N is the total number of data points;

[0102] Aggregate data points in byte information of the same cluster into the same transaction unit, and obtain a byte information transaction unit set as a first transaction unit set; aggregate data points in feature information of the same cluster into the same transaction unit, and obtain a feature information transaction unit set as a second transaction unit set.

[0103] In this embodiment, the method for selecting typical features in the same transaction unit includes:

[0104] The information entropy increase of data points in the same transaction unit is calculated, and the data points with information entropy increase in the top 30% are selected as typical labels. The expression for calculating the information entropy increase of data points is:

[0105]

[0106] Where H(D) is the entropy of the transaction unit dataset D, A is the feature standard, H(D|A) is the conditional entropy of the transaction unit dataset D under the feature standard A, k is the number of different categories in the transaction unit dataset D, C i is the i-th category in the unit dataset D, |C i | belongs to category C i The number of data points, n is the number of different values ​​of feature standard A, D j The characteristic standard A is expressed as aj The corresponding data set is a j is the jth value of feature standard A, |D j | is a subset D j The number of samples in , |D| is the number of data points in the transaction unit dataset D.

[0107] In this embodiment, the method of parsing the response instruction to the index and performing difference calculation between the first transaction unit set and the second transaction unit set includes:

[0108] Extract feature information related to typical tags in the response instruction, and convert the extracted relevant feature information into a form that can be matched with the typical tag index through a dictionary mapping method;

[0109] At the same index, the difference between the response instruction and the first transaction unit set is calculated as the first deviation, and the difference between the response instruction and the second transaction unit set is calculated as the second deviation;

[0110] If there are typical labels at relevant positions between the first transaction unit set and the second transaction unit set, the difference between the first transaction unit set and the second transaction unit set at the same index is calculated as the third deviation;

[0111] The first deviation, the second deviation, and the third deviation are taken as deviation coefficients.

[0112] In this embodiment, the method of analyzing the weight distribution of the deviation coefficient and constructing an evaluation model through a neural network based on the weight distribution includes:

[0113] Assigning weights based on the historical processing frequency of structured data and unstructured data by the machine, assigning a weight to the first deviation that is in the same proportion as the historical processing frequency of structured data, and assigning a weight to the second deviation that is in the same proportion as the historical processing frequency of unstructured data;

[0114] A penalty term is constructed based on the third deviation, where the larger the value of the third deviation, the larger the value of the penalty term. The expression of the penalty term is:

[0115]

[0116] Among them, A is the amplitude of the penalty term, B is the control parameter of the growth rate of the exponential part, d3 is the third deviation, α is the power influence parameter of d3 in the exponential part, C and D are the coefficients for adjusting the contribution degree and frequency characteristics of the sine function part to the penalty term, β is the power influence parameter of d3 in the sine function part, E is the power function amplitude coefficient, and γ is the power influence parameter of d3 in the power function part;

[0117] Based on weight distribution, the evaluation model is constructed through a convolutional neural network, including: input layer, convolution layer, logic layer, pooling layer, and output layer. Specifically:

[0118] The first layer is the input layer, which is used to receive input data, including the first deviation, the second deviation and the penalty term;

[0119] The second layer is the convolutional layer, which learns local patterns in the data by convolution and feature extraction of the input data;

[0120] The third layer is the logic layer, which is used to process the output of the convolutional layer and learn the high-level features of the input data;

[0121] The fourth layer is the pooling layer, which uses average pooling to extract key information from the features of the input layer;

[0122] The fifth layer is the output layer, which consists of a single neuron and is used to represent continuous evaluation scores. A linear activation function is selected.

[0123] Figure 2 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 2 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0124] The processor, the network interface and the memory may be interconnected via an internal bus, which may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an ELISA (Extended Industry Standard Architecture) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc.

[0125] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0126] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a standard machine-readable capability evaluation device at the logical level. The processor executes the program stored in the memory and is specifically used to execute any of the above-mentioned standard machine-readable capability evaluation methods.

[0127] The present invention can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit, a network processor (Network Processor, NP), etc.; it can also be a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field-Programmab l eGate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0128] An embodiment of the present application also proposes a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple applications, any one of the aforementioned standard machine-readable capability evaluation methods is executed.

[0129] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A standard machine readability evaluation method, characterized in that: The following steps are involved: Convert the same instruction into structured data and unstructured data respectively, and transmit them to the machine input end; Reading data from the structured data in a buffered byte stream manner, performing an alienation operation on the read byte data to obtain byte information; Performing feature conversion on the unstructured data to obtain feature information; Clustering the byte information and the feature information based on the scope respectively, aggregating the information in the same cluster into the same transaction unit, and obtaining a first transaction unit set and a second transaction unit set; Selecting a typical tag in the same transaction unit, and adding an index to the typical tag; Acquire a response instruction at the machine execution end, parse the response instruction to the index, perform difference calculation with the first transaction unit set and the second transaction unit set, and obtain a deviation coefficient; The weight distribution of the deviation coefficient is analyzed, and an evaluation model is constructed through a neural network based on the weight distribution to evaluate the machine readability.

2. A standard machine readability evaluation method according to claim 1, characterized in that: The method for converting the same instruction into structured data and unstructured data respectively comprises: Use graph neural networks to learn the node features of instructions and identify key information points in instructions; By assigning a unique key to each key information point through the key-value pair method, structured data is obtained; Convert key information points into other forms for storage, including but not limited to using text files, image files, audio files and video files to store instructions and obtain unstructured data.

3. A standard machine readability evaluation method according to claim 1, characterized in that: The method of reading data from the structured data in a buffered byte stream manner and performing an alienation operation on the read bytes comprises: Determine the storage location of the structured data. If it is a file, obtain the file path; if it is a network data source, determine the network address and port; Select the input stream based on the data storage type of the structured data. If it is a file, create a file input stream based on the file path; if it is a network data source, create a network input stream based on the network address; Wrap the input stream in a buffered input stream and use the read() method of the buffered byte stream to read the byte data byte by byte; Perform an XOR operation on each bit of the byte data with its adjacent bits.

4. A standard machine readability evaluation method according to claim 1, characterized in that: The method for feature selection of the unstructured data includes: for text files, using a bag-of-words model to extract TF-IDF feature vectors as feature information; for image files, using a convolutional neural network to extract local feature descriptors and global features of the image as feature information; for audio files, using short-time Fourier transform to extract time series features and frequency domain features from the audio files as feature information; for video files, using a convolutional neural network to extract key frame features as feature vectors.

5. A standard machine readability evaluation method according to claim 1, characterized in that: The method of clustering the byte information and the feature information based on the scope respectively, and aggregating the information in the same cluster into the same transaction unit includes: Obtain log records, and determine the scope of byte information and feature information within the machine based on the log records; The byte information and feature information are clustered based on the scope using the Gaussian mixture model, and each data point is assigned to the cluster with the highest probability of Gaussian distribution. The formula for determining the cluster to which each data point belongs is: Where k is the index of the scope, C k is the kth scope, j is C k The index of the Gaussian distribution, d is the dimension of the byte information, ∑ k,j Represented as in scope C k The covariance matrix of the jth Gaussian distribution, x i is the i-th data point, μ k,j For scope C k The mean vector of the j-th Gaussian distribution, T represents the transposition operation, Expressed as ∑ k,j The inverse of the covariance matrix, π k,j For scope C k The mixing coefficient of the jth Gaussian distribution, P(C k ) indicates that the data point belongs to scope C k The prior probability of , N is the total number of data points; Aggregate data points in byte information of the same cluster into the same transaction unit, and obtain a byte information transaction unit set as a first transaction unit set; aggregate data points in feature information of the same cluster into the same transaction unit, and obtain a feature information transaction unit set as a second transaction unit set.

6. A standard machine readability evaluation method according to claim 1, characterized in that: The method for selecting typical features within the same transaction unit comprises: The information entropy increase of data points in the same transaction unit is calculated, and the data points with information entropy increase in the top 30% are selected as typical labels. The expression for calculating the information entropy increase of data points is: Where H(D) is the entropy of the transaction unit dataset D, A is the feature standard, H(D|A) is the conditional entropy of the transaction unit dataset D under the feature standard A, k is the number of different categories in the transaction unit dataset D, C i is the i-th category in the unit dataset D, |C i | belongs to category C i The number of data points, n is the number of different values ​​of feature standard A, D j It is expressed as the characteristic standard A taking the value of a j The corresponding data set is a j is the jth value of feature standard A, |D j | is a subset D j The number of samples in , |D| is the number of data points in the transaction unit dataset D.

7. A standard machine readability evaluation method according to claim 1, characterized in that: The method of parsing the response instruction to the index and performing difference calculation between the first transaction unit set and the second transaction unit set includes: Extract feature information related to typical tags in the response instruction, and convert the extracted relevant feature information into a form that can be matched with the typical tag index through a dictionary mapping method; At the same index, the difference between the response instruction and the first transaction unit set is calculated as the first deviation, and the difference between the response instruction and the second transaction unit set is calculated as the second deviation; If there are typical labels at relevant positions between the first transaction unit set and the second transaction unit set, the difference between the first transaction unit set and the second transaction unit set at the same index is calculated as the third deviation; The first deviation, the second deviation, and the third deviation are taken as deviation coefficients.

8. A standard machine readability evaluation method according to claim 1, characterized in that: The method of analyzing the weight distribution of the deviation coefficient and constructing an evaluation model through a neural network based on the weight distribution includes: Assigning weights based on the historical processing frequency of structured data and unstructured data by the machine, assigning a weight to the first deviation that is in the same proportion as the historical processing frequency of structured data, and assigning a weight to the second deviation that is in the same proportion as the historical processing frequency of unstructured data; A penalty term is constructed based on the third deviation, where the larger the value of the third deviation, the larger the value of the penalty term. The expression of the penalty term is: Among them, A is the amplitude of the penalty term, B is the control parameter of the growth rate of the exponential part, d3 is the third deviation, α is the power influence parameter of d3 in the exponential part, C and D are the coefficients for adjusting the contribution degree and frequency characteristics of the sine function part to the penalty term, β is the power influence parameter of d3 in the sine function part, E is the power function amplitude coefficient, and γ is the power influence parameter of d3 in the power function part; Based on weight distribution, the evaluation model is constructed through a convolutional neural network, including: input layer, convolution layer, logic layer, pooling layer, and output layer. Specifically: The first layer is the input layer, which is used to receive input data, including the first deviation, the second deviation and the penalty term; The second layer is the convolutional layer, which learns local patterns in the data by convolution and feature extraction of the input data; The third layer is the logic layer, which is used to process the output of the convolutional layer and learn the high-level features of the input data; The fourth layer is the pooling layer, which uses average pooling to extract key information from the features of the input layer; The fifth layer is the output layer, which consists of a single neuron and is used to represent continuous evaluation scores. A linear activation function is selected.

9. An electronic device, comprising: processor; as well as A memory arranged to store computer executable instructions, which when executed cause the processor to perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, enables the electronic device to execute the method according to any one of claims 1 to 8.

Images