Artificial intelligence digital science and technology platform

Abstract

The invention discloses an artificial intelligence digital science and technology platform, which belongs to the technical field of data processing, and is characterized in that a data acquisition module acquires real-time network flow data and historical network flow data; the first preprocessing module preprocesses historical network traffic data, the network detection module detects real-time network traffic data by using a trained deep learning-based intrusion detection model to obtain abnormal network traffic data, and the second preprocessing module preprocesses the abnormal network traffic data to obtain target data. The feature extraction module encodes target data and performs feature extraction on the encoded target data to obtain complete semantic features, the information detection module detects the complete semantic features by using an activation function to obtain the probability that the complete semantic features are real information, and the complete semantic features corresponding to the probability greater than a preset threshold are reserved. According to the invention, the accuracy of real information detection and the efficiency of real information extraction can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to an artificial intelligence digital technology platform. Background Art

[0002] Artificial intelligence technology has been gradually applied to various fields. Artificial intelligence technology refers to the intelligent behavior exhibited by systems manufactured by humans. Among them, artificial intelligence includes technologies such as machine learning, deep learning, and natural language processing, and is widely used in fields such as medical diagnosis, autonomous driving vehicles, financial services, and smart homes. A digital technology platform refers to an online platform that uses digital technology to integrate various resources and provide comprehensive services for users. It connects users and service providers through Internet technology and has characteristics such as a multi-sided market, data-driven, openness, and flexibility. Digital technology platforms can be divided into e-commerce platforms, social media platforms, sharing economy platforms, content distribution platforms, and online education platforms. They usually connect multiple user groups, such as consumers, producers, and advertisers, and optimize services and enhance the user experience through data analysis.

[0003] Currently, with the rapid development of the Internet, we have also entered the big data era. The amount of network data collected from various media platforms or devices has increased explosively. Due to the virtual nature of the Internet, a large amount of false information has also flowed into the network, with an increasing spread speed and a wider spread range. As a result, a large amount of false information has also been mixed into various media platforms or devices, challenging the authenticity and reliability of the data. The current extraction and classification methods are difficult to meet the detection requirements for false information in network data to obtain real information.

[0004] Therefore, how to provide an effective technical solution to meet the detection requirements for false information in network data to obtain real information has become an urgent problem to be solved in the existing technology. Summary of the Invention

[0005] The purpose of the present invention is to provide an artificial intelligence digital technology platform to solve the problem in the existing technology that it is difficult to meet the detection requirements for false information in network data to obtain real information.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an artificial intelligence digital technology platform, including a data acquisition module, a first preprocessing module, a network detection module, a second preprocessing module, a feature extraction module, and an information detection module; The data acquisition module is used to acquire the real-time network traffic data and historical network traffic data of the media device. The real-time network traffic data includes real-time media data and user data. The historical network traffic data is uploaded to the first preprocessing module, and the real-time network traffic data is uploaded to the network detection module; The first preprocessing module is used to perform feature preprocessing on the historical network traffic data to obtain network feature data, and upload the network feature data to the network detection module; The network detection module is used to input the network feature data into an intrusion detection model based on deep learning for training, and use the trained intrusion detection model based on deep learning to detect the real-time network traffic data to obtain abnormal network traffic data, trigger an alarm based on the abnormal network traffic data, and upload the abnormal network traffic data to the second preprocessing module; The second preprocessing module is used to perform data preprocessing on the abnormal network traffic data to obtain target data, and upload the target data to the feature extraction module; The feature extraction module is used to encode the target data to obtain the encoded target data, extract features from the encoded target data to obtain complete semantic features, and upload the complete semantic features to the information detection module; The information detection module is used to detect the complete semantic features using an activation function to obtain the probability that the complete semantic features are real information, and retain the complete semantic features corresponding to a probability greater than a preset threshold to obtain real information.

[0007] In a possible design, performing feature preprocessing on the historical network traffic data to obtain network feature data includes: Extracting feature vectors from the historical network traffic data, where the feature vectors include length, time, and protocol type; Generating a feature matrix according to the feature vectors to obtain network feature data.

[0008] In a possible design, the intrusion detection model based on deep learning is constructed according to the CNN model. Among them, the intrusion detection model based on deep learning at least includes a convolutional layer, a pooling layer, and a fully connected layer; using the trained intrusion detection model based on deep learning to detect the real-time network traffic data to obtain abnormal network traffic data includes: Extracting local features of the real-time network traffic data through the convolutional layer; Reducing the dimension of the local features through the pooling layer to obtain reduced-dimensional features; Performing non-linear combination on the reduced-dimensional features through the fully connected layer to obtain abnormal network traffic data.

[0009] In a possible design, data preprocessing is performed on abnormal network traffic data to obtain target data, including: Calculate the average value of abnormal network traffic data, and supplement the missing values ​​of abnormal network traffic data according to the average value; Identify the erroneous data in the abnormal network traffic data, replace the erroneous data with the average value, and normalize the abnormal network traffic data after replacing the average value to obtain the target data.

[0010] In a possible design, the target data is encoded to obtain the encoded target data, and feature extraction is performed on the encoded target data to obtain complete semantic features, including: Encoding the target data to obtain encoded target data, and extracting features from the encoded target data to obtain first feature data; Extracting deep semantic features of the first feature data to obtain second feature data; The attention mechanism is used to perform weighted processing on each feature in the second feature data and the weight corresponding to each feature to obtain a complete semantic feature.

[0011] In a possible design, encoding the target data to obtain the encoded target data, and extracting features from the encoded target data to obtain first feature data include: Serialize the target data to obtain a text sequence of the target data; The text sequence of the target data is input into the Transformer model for feature extraction, and the first feature data is output.

[0012] In a possible design, the text sequence of the target data is input into the Transformer model for feature extraction, and the first feature data is output, including: Input the text sequence of the target data into the Transformer model; The Attention mask mechanism is used inside the Transformer model, and the text sequence of the target data is fully arranged based on the two-stream self-attention mechanism to obtain a fully arranged text sequence; Sampling optimization is performed on the fully arranged text sequence to obtain a sampled optimized text sequence, and feature extraction is performed on the sampled optimized text sequence to obtain first feature data.

[0013] In a possible design, extracting deep semantic features of the first feature data to obtain second feature data includes: Input the first feature data into the bidirectional GRU model, and output the forward hidden layer state and the reverse hidden layer state at the current moment; The forward hidden layer state and the backward hidden layer state at the current moment are superimposed and calculated to obtain the second feature data.

[0014] In a possible design, the calculation expression for superimposing and calculating the forward hidden layer state and the backward hidden layer state is: ; In the above formula, represents the second feature data; represents the forward hidden layer state output after the first feature data is input into the bidirectional GRU model at the current moment t; represents the backward hidden layer state output after the first feature data is input into the bidirectional GRU model at the current moment t; represents the first feature data; represents the forward hidden layer state at the previous moment t - 1; represents the backward hidden layer state at the previous moment t - 1; represents the bias parameter corresponding to the current moment t.

[0015] Furthermore, when the current moment t is the first moment, and are both set to all-zero vectors as the states at the previous moment t - 1 of the first moment. After calculating the forward hidden layer state and the backward hidden layer state at the current moment t, the forward hidden layer state at the current moment t is used as the forward hidden layer state at the next moment t + 1, and the backward hidden layer state at the current moment t is used as the backward hidden layer state at the next moment t + 1.

[0016] In a possible design, the calculation expression for detecting the complete semantic feature using the activation function is: ; In the above formula, represents the probability that the complete semantic feature is real information; represents the weight coefficient matrix; represents the complete semantic feature; represents the bias term.

[0017] The beneficial effects of the present invention are as follows: The present invention provides an artificial intelligence digital technology platform, including a data acquisition module, a first preprocessing module, a network detection module, a second preprocessing module, a feature extraction module, and an information detection module. The data acquisition module acquires real-time network traffic data and historical network traffic data of a media device. The first preprocessing module performs feature preprocessing on the historical network traffic data to obtain network feature data. The network detection module uses a trained intrusion detection model based on deep learning to detect the real-time network traffic data to obtain abnormal network traffic data. After the second preprocessing module preprocesses the abnormal network traffic data, target data is obtained. The feature extraction module encodes the target data to obtain the encoded target data, extracts features from the encoded target data to obtain complete semantic features, and the information detection module uses an activation function to detect the complete semantic features to obtain the probability that the complete semantic features are real information, and retains the complete semantic features corresponding to the probability greater than a preset threshold. The present invention first uses the network detection module to detect the real-time network traffic data, and uses the intrusion detection model based on deep learning for detection, which can effectively improve the response efficiency and accuracy of abnormal traffic data. The feature extraction module is used to process the abnormal network traffic data. The information detection module uses an activation function to detect the complete semantic features to obtain the probability of real information, and retains the complete semantic features corresponding to the probability of real information greater than the preset threshold to obtain real information, extracts real information from the abnormal network traffic information, improves the utilization rate of data, and effectively improves the accuracy of real information detection and the efficiency of extracting real information, saving network resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A module diagram of an artificial intelligence digital technology platform provided for Embodiment 1; Figure 2 A flowchart of an implementation method of an artificial intelligence digital technology platform provided for Embodiment 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the present invention in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the drawings is only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. It should be noted here that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation to the present invention.

[0020] Embodiment 1: As Figure 1As shown in the figure, this embodiment provides an artificial intelligence digital technology platform, including a data acquisition module, a first preprocessing module, a network detection module, a second preprocessing module, a feature extraction module, and an information detection module; The data acquisition module is used to acquire the real-time network traffic data and historical network traffic data of the media device, upload the historical network traffic data to the first preprocessing module, and upload the real-time network traffic data to the network detection module; The first preprocessing module is used to perform feature preprocessing on the historical network traffic data to obtain network feature data, and upload the network feature data to the network detection module; The network detection module is used to input the network feature data into an intrusion detection model based on deep learning for training, use the trained intrusion detection model based on deep learning to detect the real-time network traffic data, obtain abnormal network traffic data, trigger an alarm based on the abnormal network traffic data, and upload the abnormal network traffic data to the second preprocessing module; The second preprocessing module is used to perform data preprocessing on the abnormal network traffic data to obtain target data, and upload the target data to the feature extraction module; The feature extraction module is used to encode the target data to obtain the encoded target data, extract features from the encoded target data to obtain complete semantic features, and upload the complete semantic features to the information detection module; The information detection module is used to detect the complete semantic features using an activation function to obtain the probability that the complete semantic features are real information, and retain the complete semantic features corresponding to the probability greater than the preset threshold to obtain real information.

[0021] Furthermore, the real-time network traffic data includes user data and real-time media data. The user data includes data such as the current user's nickname, avatar, and account. The real-time media data includes various articles and posts etc. information published by the user in real time. The historical network traffic data includes historical user data and historical media data. The historical user data includes data such as the user's historical nickname, avatar, and account. The historical media data includes various articles and posts etc. information published by the user historically. According to the probability of real information of the user's historical social media data, the user can be marked, and in information detection, the real-time network traffic data of the marked user is detected with emphasis.

[0022] In a possible design, performing feature preprocessing on the historical network traffic data to obtain network feature data includes: Extracting the feature vectors in the historical network traffic data, where the feature vectors include length, time, and protocol type; Generating a feature matrix according to the feature vectors to obtain network feature data.

[0023] In a possible design, the deep learning-based intrusion detection model is constructed according to a CNN model, where the deep learning-based intrusion detection model includes at least a convolutional layer, a pooling layer, and a fully connected layer; using the trained deep learning-based intrusion detection model to detect real-time network traffic data to obtain abnormal network traffic data, including: Extract local features of real-time network traffic data through the convolutional layer; Reduce the dimension of the local features through the pooling layer to obtain dimension-reduced features; Perform non-linear combination on the dimension-reduced features through the fully connected layer to obtain abnormal network traffic data.

[0024] In a possible design, perform data preprocessing on the abnormal network traffic data to obtain target data, including: Calculate the average value of the abnormal network traffic data, and supplement the missing values of the abnormal network traffic data according to the average value; Identify the error data in the abnormal network traffic data, replace the error data with the average value, and perform normalization processing on the abnormal network traffic data after replacing the average value to obtain target data.

[0025] In a possible design, encode the target data to obtain the encoded target data, and extract features from the encoded target data to obtain complete semantic features, including: Encode the target data to obtain the encoded target data, and extract features from the encoded target data to obtain the first feature data; Extract the deep semantic features of the first feature data to obtain the second feature data; Use the attention mechanism to perform weighted processing on each feature in the second feature data and the weights corresponding to each feature to obtain complete semantic features.

[0026] The attention mechanism is a technology that mimics the human visual and cognitive systems, allowing the neural network to focus on relevant parts when processing input data. The goal of using the attention mechanism is to select more critical information for the current task goal from numerous information, thereby improving the performance and generalization ability of the model.

[0027] In a possible design, encode the target data to obtain the encoded target data, and extract features from the encoded target data to obtain the first feature data, including: Serialize the target data to obtain the text sequence of the target data; Input the text sequence of the target data into the Transformer model for feature extraction, and output the first feature data.

[0028] The Transformer model is a neural network model based on the self-attention mechanism. By introducing the self-attention mechanism, it breaks the limitations of traditional RNNs in processing long sequence data, enabling the model to perform parallel computing and significantly improving the training and inference efficiency.

[0029] In a possible design, the text sequence of the target data is input into the Transformer model for feature extraction, and the first feature data is output, including: Input the text sequence of the target data into the Transformer model; Use the Attention mask mechanism inside the Transformer model and perform a full permutation of the text sequence of the target data based on the two-stream self-attention mechanism to obtain the fully permuted text sequence; Specifically, use the Attention mask mechanism (attention mask mechanism) to perform a full permutation of the text sequence of the target data, and use the two-stream self-attention mechanism to eliminate the influence of the initial position after permutation on the model training effect. The basic idea of the Attention mask mechanism is that when processing input data, the model can "focus" on certain parts of the input sequence instead of treating all parts equally, enabling the model to more effectively extract relevant information from the data and improve the processing performance. The two-stream self-attention mechanism is an attention mechanism used in natural language processing tasks. It is an extension based on the self-attention mechanism and processes different types of information by introducing two independent attention streams.

[0030] Perform sampling optimization on the fully permuted text sequence to obtain the sampled and optimized text sequence, and perform feature extraction on the sampled and optimized text sequence to obtain the first feature data.

[0031] In a possible design, extract the deep semantic features of the first feature data to obtain the second feature data, including: Input the first feature data into the bidirectional GRU model, and output the forward hidden state and backward hidden state at the current moment; Specifically, the bidirectional GRU model is a deep learning model composed of two independent GRU networks, which model the input sequence from two directions, forward and backward. The forward GRU model sequentially takes each word from the first feature data from front to back in time order and embeds the word into the forward GRU model; the backward GRU model sequentially takes each word from the first feature data from back to front in time order and embeds the word into the backward GRU model.

[0032] The forward hidden layer state and backward hidden layer state at the current moment are superimposed and calculated to obtain the second feature data.

[0033] Specifically, the outputs of the forward GRU and backward GRU are concatenated into a vector, and the second feature data is obtained through a fully connected layer.

[0034] In a possible design, the calculation expression for superimposing and calculating the forward hidden layer state and backward hidden layer state at the current moment is: ; In the formula, represents the second feature data; represents the forward hidden layer state output after the first feature data is input into the bidirectional GRU model at the current moment t; represents the backward hidden layer state output after the first feature data is input into the bidirectional GRU model at the current moment t; represents the first feature data; represents the forward hidden layer state at the previous moment t - 1; represents the backward hidden layer state at the previous moment t - 1; represents the bias parameter corresponding to the current moment t.

[0035] Furthermore, when the current moment t is the first moment, and are both set to all-zero vectors as the states at the previous moment t - 1 of the first moment. After calculating the forward hidden layer state and backward hidden layer state at the current moment t, the forward hidden layer state at the current moment t is used as the forward hidden layer state at the next moment t + 1, and the backward hidden layer state at the current moment t is used as the backward hidden layer state at the next moment t + 1.

[0036] In a possible design, the calculation expression for detecting the complete semantic feature using an activation function is: ; In the formula, represents the probability that the complete semantic feature is real information; represents the weight coefficient matrix; represents the complete semantic feature; represents the bias term.

[0037] In a possible design, the parameters of the intrusion detection model based on deep learning are continuously optimized through the backpropagation algorithm to minimize the loss function and improve the classification accuracy of the model.

[0038] In summary, this embodiment discloses an artificial intelligence digital technology platform, including a data acquisition module, a first preprocessing module, a network detection module, a second preprocessing module, a feature extraction module, and an information detection module. The data acquisition module acquires real-time network traffic data and historical network traffic data of a media device. The first preprocessing module performs feature preprocessing on the historical network traffic data to obtain network feature data. The network detection module uses a trained deep learning-based intrusion detection model to detect the real-time network traffic data to obtain abnormal network traffic data. After the second preprocessing module preprocesses the abnormal network traffic data, target data is obtained. The feature extraction module encodes the target data to obtain the encoded target data, extracts features from the encoded target data to obtain complete semantic features, and the information detection module uses an activation function to detect the complete semantic features to obtain the probability that the complete semantic features are real information, and retains the complete semantic features corresponding to the probability greater than a preset threshold. In this embodiment, first, the network detection module is used to detect the real-time network traffic data, and a deep learning-based intrusion detection model is used for detection to improve the detection efficiency and accuracy of abnormal traffic data. The activation function is used to detect the complete semantic features to obtain the real information probability, and the complete semantic features corresponding to the real information probability greater than the preset threshold are retained to obtain real information, extracting real information from the abnormal network traffic information, improving the utilization rate of data, and effectively improving the accuracy of real information detection and the efficiency of extracting real information, saving network resources. The parameters of the deep learning-based intrusion detection model are continuously optimized through the backpropagation algorithm to minimize the loss function and improve the classification accuracy of the model.

[0039] Embodiment 2: This embodiment provides a method for implementing an artificial intelligence digital technology platform, including the following steps: S1. Acquire real-time network traffic data and historical network traffic data of a media device; S2. Perform feature preprocessing on the historical network traffic data to obtain network feature data; S3. Input the network feature data into a deep learning-based intrusion detection model for training, use the trained deep learning-based intrusion detection model to detect the real-time network traffic data to obtain abnormal network traffic data, and trigger an alarm based on the abnormal network traffic data; S4. Perform data preprocessing on the abnormal network traffic data to obtain target data; S5. Encode the target data to obtain the encoded target data, extract features from the encoded target data to obtain complete semantic features; S6. Use an activation function to detect the complete semantic features, obtain the probability that the complete semantic features are real information, and retain the complete semantic features corresponding to a probability greater than a preset threshold to obtain real information.

[0040] In summary, this embodiment discloses a method for implementing an artificial intelligence digital technology platform, including: obtaining real-time network traffic data and historical network traffic data of a media device; performing feature preprocessing on the historical network traffic data to obtain network feature data; inputting the network feature data into an intrusion detection model based on deep learning for training, using the trained intrusion detection model based on deep learning to detect the real-time network traffic data to obtain abnormal network traffic data, and triggering an alarm based on the abnormal network traffic data; performing data preprocessing on the abnormal network traffic data to obtain target data; encoding the target data to obtain the encoded target data, extracting features from the encoded target data to obtain complete semantic features; using an activation function to detect the complete semantic features, obtaining the probability that the complete semantic features are real information, and retaining the complete semantic features corresponding to a probability greater than a preset threshold to obtain real information. This embodiment can effectively improve the accuracy and efficiency of extracting real information by processing the target data and using an activation function for detection; at the same time, an intrusion detection model based on deep learning is used to detect the real-time network traffic data, improving the detection efficiency and accuracy of abnormal traffic data.

[0041] Embodiment 3: This embodiment provides a computer-readable storage medium, on which instructions are stored. When the instructions run on a computer, they are used to execute the method for implementing the artificial intelligence digital technology platform as described in Embodiment 2. Among them, the computer-readable storage medium refers to a carrier for storing data, which can but is not limited to including computer-readable storage media such as floppy disks, optical discs, hard disks, flash memories, USB flash drives, and / or memory sticks. The computer can be a general-purpose computer, a dedicated computer, a computer network, or other programmable devices.

[0042] For the working process, working details, and technical effects of the foregoing computer-readable storage medium provided in this embodiment, reference can be made to the method for implementing the artificial intelligence digital technology platform as described in Embodiment 2, which will not be elaborated here.

[0043] Embodiment 4: This embodiment provides a computer device, including a memory, a processor, and a transceiver that are communicatively connected in sequence. Among them, the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the method for implementing the artificial intelligence digital technology platform as described in Embodiment 2.

[0044] Specifically, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in first-out memory (FIFO), and / or first-in last-out memory (FILO), etc.; the processor is not limited to using a microprocessor of the STM32F105 series, a processor with an architecture such as ARM (Advanced RISC Machines), X86, or a processor integrated with an NPU (neural-network processing units); the transceiver may include, but is not limited to, a WiFi (Wireless Fidelity) wireless transceiver, a Bluetooth wireless transceiver, a General Packet Radio Service (GPRS) wireless transceiver, a ZigBee (a low-power local area network protocol based on the IEEE 802.15.4 standard) wireless transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver, etc.

[0045] For the working process, working details, and technical effects of the device provided in this embodiment, reference may be made to the implementation method of the artificial intelligence digital technology platform described in Embodiment 2, which will not be elaborated here.

[0046] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An artificial intelligence digital technology platform, characterized in that: It includes a data acquisition module, a first preprocessing module, a network detection module, a second preprocessing module, a feature extraction module and an information detection module; The data acquisition module is used to acquire real-time network traffic data and historical network traffic data of the media device, upload the historical network traffic data to the first preprocessing module, and upload the real-time network traffic data to the network detection module; The first preprocessing module is used to perform feature preprocessing on the historical network traffic data to obtain network feature data, and upload the network feature data to the network detection module; The network detection module is used to input the network feature data into the deep learning-based intrusion detection model for training, use the trained deep learning-based intrusion detection model to detect the real-time network traffic data, obtain abnormal network traffic data, trigger an alarm based on the abnormal network traffic data and upload the abnormal network traffic data to the second preprocessing module; The second preprocessing module is used to perform data preprocessing on the abnormal network traffic data to obtain target data, and upload the target data to the feature extraction module; The feature extraction module is used to encode the target data to obtain the encoded target data, perform feature extraction on the encoded target data to obtain complete semantic features, and upload the complete semantic features to the information detection module; The information detection module is used to detect the complete semantic features using an activation function to obtain the probability that the complete semantic features are true information, and retain the complete semantic features corresponding to the probability greater than a preset threshold to obtain true information.

2. An artificial intelligence digital technology platform according to claim 1, characterized in that: Perform feature preprocessing on historical network traffic data to obtain network feature data, including: Extracting feature vectors from historical network traffic data, wherein the feature vectors include length, time, and protocol type; Generate a feature matrix based on the feature vector to obtain network feature data.

3. The artificial intelligence digital technology platform according to claim 1, characterized in that: The deep learning-based intrusion detection model is constructed according to the CNN model, wherein the deep learning-based intrusion detection model at least includes a convolutional layer, a pooling layer, and a fully connected layer; the real-time network traffic data is detected using the trained deep learning-based intrusion detection model to obtain abnormal network traffic data, including: Extract local features of real-time network traffic data through convolutional layers; The dimension of local features is reduced through the pooling layer to obtain reduced-dimensional features; The dimension reduction features are nonlinearly combined through the fully connected layer to obtain abnormal network traffic data.

4. The artificial intelligence digital technology platform according to claim 1, characterized in that: Perform data preprocessing on abnormal network traffic data to obtain target data, including: Calculate the average value of abnormal network traffic data, and supplement the missing values ​​of abnormal network traffic data according to the average value; Identify the erroneous data in the abnormal network traffic data, replace the erroneous data with the average value, and normalize the abnormal network traffic data after replacing the average value to obtain the target data.

5. The artificial intelligence digital technology platform according to claim 1, characterized in that: Encode the target data to obtain the encoded target data, extract features from the encoded target data, and obtain complete semantic features, including: Encoding the target data to obtain encoded target data, and extracting features from the encoded target data to obtain first feature data; Extracting deep semantic features of the first feature data to obtain second feature data; The attention mechanism is used to perform weighted processing on each feature in the second feature data and the weight corresponding to each feature to obtain a complete semantic feature.

6. An artificial intelligence digital technology platform according to claim 5, characterized in that: Encoding the target data to obtain the encoded target data, and extracting features from the encoded target data to obtain first feature data, including: Serialize the target data to obtain a text sequence of the target data; The text sequence of the target data is input into the Transformer model for feature extraction, and the first feature data is output.

7. An artificial intelligence digital technology platform according to claim 6, characterized in that: The text sequence of the target data is input into the Transformer model for feature extraction, and the first feature data is output, including: Input the text sequence of the target data into the Transformer model; The Attention mask mechanism is used inside the Transformer model, and the text sequence of the target data is fully arranged based on the two-stream self-attention mechanism to obtain a fully arranged text sequence; Sampling optimization is performed on the fully arranged text sequence to obtain a sampled optimized text sequence, and feature extraction is performed on the sampled optimized text sequence to obtain first feature data.

8. The artificial intelligence digital technology platform according to claim 5, characterized in that: Extracting deep semantic features of the first feature data to obtain second feature data includes: Input the first feature data into the bidirectional GRU model, and output the forward hidden layer state and the reverse hidden layer state at the current moment; The forward hidden layer state and the reverse hidden layer state at the current moment are superimposed and calculated to obtain the second feature data.

9. An artificial intelligence digital technology platform according to claim 8, characterized in that: The calculation expression for superimposing the current forward hidden layer state and the reverse hidden layer state is: ； In the formula, represents the second characteristic data; It represents the forward hidden layer state outputted after the first feature data is input into the bidirectional GRU model at the current time t; It represents the reverse hidden layer state outputted after the first feature data is input into the bidirectional GRU model at the current time t; represents the first characteristic data; Represents the forward hidden layer state at the previous moment t-1; Represents the reverse hidden layer state at the previous moment t-1; Represents the bias parameter corresponding to the current time t.

10. The artificial intelligence digital technology platform according to claim 1, characterized in that: The calculation expression for detecting complete semantic features using the activation function is: ； In the formula, Indicates the probability that the complete semantic feature is true information; represents the weight coefficient matrix; Represents complete semantic features; Represents the bias term.

Images