An encrypted traffic analysis method and electronic device based on intra-flow burst interaction graph

By using a method based on intra-flow burst interaction graphs, the problems of inaccurate intra-flow burst threshold division and high training overhead of multiple models are solved, enabling efficient identification and fine-grained analysis of unknown application traffic.

CN117896144BActive Publication Date: 2026-11-17NAT UNIV OF DEFENSE TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410064464.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-16
Publication Date
2026-11-17
Estimated Expiration
2044-01-16

AI Technical Summary

Technical Problem

Existing technologies for encrypted traffic analysis suffer from problems such as inaccurate burst threshold division within a flow, high overhead of training multiple models, and inability to distinguish between unknown applications.

Method used

We employ a method based on intra-flow burst interaction graphs, which involves network flow clustering, constructing flow interaction graphs, and training with a two-layer bidirectional recurrent neural network and a graph neural network to achieve fine-grained analysis of intra-flow bursts.

Benefits of technology

It achieves adaptive burst threshold division within the flow under different environments, can identify unknown application traffic with high accuracy, and supports multi-application traffic classification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117896144B_ABST
    Figure CN117896144B_ABST
Patent Text Reader

Abstract

This invention provides an encrypted traffic analysis method and electronic device based on an intra-flow burst interaction graph. The method includes: clustering network flows, extracting the long sequence of network flow packets from each network flow, and further extracting intra-flow bursts from each network flow; constructing a traffic interaction graph of intra-flow bursts, treating each intra-flow burst as a node, sequentially connecting intra-flow bursts within the same network flow to represent their spatial context, and sequentially connecting intra-flow bursts generated at the same time to represent their temporal context; generating node embeddings for each intra-flow burst using a two-layer bidirectional recurrent neural network; training using a two-layer GNN, and finally using a fully connected layer for traffic identification. It can autonomously adapt to intra-flow burst thresholds under different environments. It can learn the differences between different categories, thus enabling the analysis of traffic from unknown applications. It can complete traffic identification with high accuracy. The model supports node classification and can be used for more fine-grained analysis tasks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of encrypted traffic analysis technology, specifically relating to an encrypted traffic analysis method and electronic device based on intra-flow burst interaction graphs. Background Technology

[0002] FOAP is a machine learning-based model for identifying fine-grained user actions. It transforms user action identification in network streams into a sequence labeling problem under a conditional random field (CRF). The modeling method for CRF is as follows... Figure 1 As shown, Figure 1 As shown, FOAP treats a network flow as a sequence of bursts within several flows (m a ,m b ,m c Each intra-flow burst consists of multiple data packets, and its corresponding user action is (b1, b2, b3, b4). Intra-flow bursts within the same network flow have spatial context relationships, while intra-flow bursts in different network flows but at the same time have temporal context relationships.

[0003] Ultimately, based on the two relationships mentioned above, the burst sequences within a flow can form a Conditional Random Field (CRF), a probabilistic graphical model widely used in sequence labeling. The FOAP model, on the other hand, treats the traffic feature sequence as the sequence to be labeled and the user behavior recognition result as the output sequence. In this case, the traffic behavior recognition problem is transformed into a part-of-speech tagging problem, which can be solved directly by applying the CRF model.

[0004] The FOAP model has the following drawbacks:

[0005] Intra-flow bursts are classified based on thresholds. The threshold parameters need to be adjusted under different traffic scenarios, which can easily lead to inaccurate classification.

[0006] FOAP's CRF model can only analyze traffic from one type of app. For mixed traffic from multiple apps, multiple models need to be trained, resulting in significant training overhead.

[0007] When identifying unknown application traffic, FOAP, due to its stacked multiple single-application classifiers, requires multiple identification processes and ultimately can only label unknown traffic as "unknown." It cannot distinguish between different unknown applications. Summary of the Invention

[0008] This invention provides an encrypted traffic analysis method and electronic device based on intra-flow burst interaction graphs to solve at least one of the above-mentioned problems in the prior art.

[0009] According to a first aspect of the present invention, one or more embodiments of this application provide an encrypted traffic analysis method based on an intra-flow burst interaction graph, comprising the following steps:

[0010] Step 1: Cluster the network flows, extract the long sequence of network flow packets in each network flow, and continue to extract intra-flow bursts for each network flow;

[0011] Step 2: Construct a flow interaction graph of intra-flow bursts, treating each intra-flow burst as a node. Intra-flow bursts within the same network flow are connected sequentially to represent their spatial context relationship. Intra-flow bursts generated at the same time are also connected sequentially to represent their temporal context relationship.

[0012] Step 3: Use a two-layer bidirectional recurrent neural network to generate node embeddings for each burst within the flow;

[0013] Step 4: Use a two-layer GNN for training. For node classification, classify the features of each node directly to predict the fine-grained fingerprint corresponding to each burst in the flow. For graph classification, perform mean pooling on the output node representations to generate the overall traffic fingerprint. Finally, use a fully connected layer for traffic identification.

[0014] The network flow clustering steps include: selecting an unvisited data point from one of the data packets within the network flow; if the neighborhood density of this data point exceeds the neighborhood radius between data packets, then designating it as the core point of a new cluster; adding all unvisited data points within the neighborhood radius between the core point and the data packets to the cluster, and recursively expanding the cluster; if the neighborhood density of the core point is insufficient to form a cluster, then marking it as a noise point; repeating the above steps until all data points have been visited.

[0015] The construction steps of the intra-flow burst interaction graph include: combining the spatial and temporal context relationships between network flows to describe the information between network flows in a fine-grained manner; and using intra-flow bursts as nodes, establishing a flow interaction graph within each time window, with the size of the time window determined by the type of downstream task.

[0016] The bidirectional recurrent neural network includes a feature extraction network and a data embedding network.

[0017] The feature extraction network uses RNN to extract feature information from the input network stream packet long sequence, including converting the network stream packet long sequence into an embedding sequence;

[0018] The data embedding network uses a bidirectional GRU to model and predict the feature information in the long sequence of the input network packet. This includes inputting the embedding sequence into the bidirectional GRU in sequence and finally using the hidden layer of the bidirectional GRU as an intra-stream burst embedding.

[0019] In this method, intra-flow burst embedding is used to calculate the similarity between two network flows, and unknown traffic is distinguished based on the calculated similarity.

[0020] The training process using a two-layer GNN includes:

[0021] First, input a traffic interaction graph consisting of nodes and edges. This graph represents the interaction and traffic propagation between different nodes.

[0022] Use the first GNN layer to capture local information about the nodes;

[0023] A new node representation vector is obtained by weighted summing the node representation vector after capturing local information with the original node representation vector.

[0024] Use a second GNN layer to capture the distance between nodes;

[0025] The node representation vector and the distance representation vector are merged into a global representation vector, which contains the features of the node itself and the distance information between nodes.

[0026] The node classification process includes:

[0027] During the sampling phase, some nodes and their neighboring nodes are selected from the graph for sampling to obtain the local neighborhood graph of the nodes;

[0028] In the aggregation phase, for each sampled node, the set of its neighboring nodes is aggregated into a set, and a random node is selected. The set of neighboring nodes is used as input to obtain the global neighborhood graph of the node.

[0029] During the update phase, the neural network parameters are updated using the global neighborhood graph and label information to minimize the loss function. The above steps are repeated until the preset number of iterations is reached or the stopping condition is met.

[0030] Wherein, if the network flow packet length sequence is v and the generated embedding sequence is u, the process of intra-flow burst embedding of the network flow packet length sequence through RNN is described as follows:

[0031] f(v)=RNN(v)=u.

[0032] The similarity is represented as follows:

[0033]

[0034] The similarity between all network flows belonging to the same category is represented as follows:

[0035]

[0036] The relationship between objective function modeling and overall similarity is expressed as follows:

[0037]

[0038] Where #(v1,v2) represents the number of times the traffic pair (v1,v2) appears in a certain category, I represents the category, and v1,v2 is a long sequence of network flow packets.

[0039] According to another aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the encrypted traffic analysis method based on intra-flow burst interaction graph as described in any of the above claims.

[0040] The beneficial effects of this invention are that it provides an encrypted traffic analysis method and electronic device based on intra-flow burst interaction graphs, proposes a clustering-based intra-flow burst partitioning method, and can autonomously adapt to intra-flow burst thresholds under different environments. A recurrent neural network model based on contrastive loss is designed, which can learn the differences between different categories, thereby enabling the analysis of unknown application traffic. An RNN+GNN network model is designed, which can complete traffic identification with high accuracy. Simultaneously, this model supports node classification for more fine-grained analysis tasks. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the prior art in the background technology of an encrypted traffic analysis method and electronic device based on an intra-flow burst interaction graph, which is an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of the overall architecture of an encrypted traffic analysis method and electronic device based on an intra-flow burst interaction graph according to an embodiment of the present invention.

[0043] Figure 3 This is a schematic diagram illustrating an encrypted traffic analysis method based on an intra-flow burst interaction graph according to an embodiment of the present invention, and the construction of an intra-flow burst traffic interaction graph for an electronic device.

[0044] Figure 4 This is a schematic diagram of a method for encrypted traffic analysis based on intra-flow burst interaction graphs and a network flow embedding process of a bidirectional recurrent neural network (GRU) for electronic devices, according to an embodiment of the present invention.

[0045] Figure 5 This is a schematic diagram of an encrypted traffic analysis method based on intra-flow burst interaction graphs and a GNN training architecture for an electronic device, according to an embodiment of the present invention.

[0046] Figure 6This is a schematic diagram illustrating an encrypted traffic analysis method based on intra-flow burst interaction graphs and a ranking analysis of the single-class application behavior recognition effect of electronic devices, according to an embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0048] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in one or more embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0049] like Figure 1 As shown, FOAP treats a network flow as a sequence of bursts within several flows (m a ,m b ,m c Each burst within a stream consists of multiple data packets, and their corresponding user actions are (b1, b2, b3, b4). For example... Figure 1 As shown, intra-flow bursts within the same network flow exhibit spatial contextual relationships, while intra-flow bursts from different network flows occurring at the same time exhibit temporal contextual relationships. Ultimately, based on these two relationships, the sequence of intra-flow bursts can constitute a Conditional Random Field (CRF), a probabilistic graphical model widely used in sequence labeling. The FOAP model, on the other hand, treats the traffic feature sequence as the sequence to be labeled and the user behavior recognition result as the output sequence. In this case, the traffic behavior recognition problem is transformed into a part-of-speech tagging problem.

[0050] like Figure 2 As shown, one or more embodiments of this application illustrate an encrypted traffic analysis method based on an intra-flow burst interaction graph, which includes the following steps:

[0051] Step 1: Cluster the network flows, extract the long sequence of network flow packets in each network flow, and continue to extract intra-flow bursts for each network flow; the subsequent model will identify each cluster with intra-flow bursts as the basic unit. In order to deal with the problem of inaccurate segmentation based on threshold, this embodiment uses a clustering method based on DB-scan in this process, which can perform clustering without specifying the number of clusters.

[0052] Step 2: Construct a flow interaction graph of intra-flow bursts, treating each intra-flow burst as a node. Intra-flow bursts within the same network flow are sequentially connected to represent their spatial context relationship. Intra-flow bursts generated at the same time are also sequentially connected to represent their temporal context relationship. Treating intra-flow bursts as nodes allows for effective characterization of intra-flow feature relationships through edges within the network flow, and also characterizes inter-flow relationships through edges between network flows. Furthermore, refining the recognition granularity to the network flow level enables more sophisticated recognition tasks.

[0053] Step 3: Generate node embeddings for each intra-flow burst using a two-layer bidirectional recurrent neural network. For each intra-flow burst, the corresponding feature is the original packet-length sequence; therefore, node embeddings for each intra-flow burst need to be generated before training the graph neural network. Currently, recurrent neural networks are a mature embedding method for packet-length sequences, so this method selects this network as the main method for intra-flow burst embedding. During the flow embedding process, the distance between samples of different classes is introduced as part of the loss function, enabling the network to learn the differences between classes.

[0054] Step 4: Use a two-layer GNN for training. For node classification, classify the features of each node directly to predict the fine-grained fingerprint corresponding to each burst in the flow. For graph classification, perform mean pooling on the output node representations to generate the overall traffic fingerprint. Finally, use a fully connected layer for traffic identification.

[0055] In this embodiment, unless otherwise specified, the definition of an intra-flow burst is given first: given a threshold ε, an intra-flow burst is a subsequence of packets in a network flow, where the time interval between two packets is less than ε, but the time interval between two inflow queues is not less than ε. That is, an intra-flow burst can be regarded as traffic interaction between two silent periods.

[0056] In practice, the threshold ε varies depending on the scenario and network flow. Using a fixed threshold ε for burst segmentation within a flow is insufficient to accurately characterize this pattern. Therefore, this embodiment employs a DB-Scan-based clustering method, which clusters data packets based on packet density. This allows for the autonomous discovery of a variable threshold to segment bursts within a flow. Thus, this problem is abstracted as a clustering problem for one-dimensional sequence data, where the clustering between each sample point represents the actual time interval of the data packets.

[0057] DB-Scan-based clustering methods can automatically discover clusters of different shapes, sizes, and densities in data, and effectively handle noise and outliers. Its main idea is to determine cluster boundaries by calculating the density around each data point. High-density areas are considered the core of a cluster, while low-density areas are considered noise or boundaries between two clusters. The DB-Scan algorithm requires two parameters: radius ∈ and minimum number of points MinPts, used to define cluster boundaries. In this scenario, radius ∈ is the neighborhood radius between data packets, which can be set to half the minimum burst time span within the flow. Minimum number of points MinPts can be set to the minimum number of burst data packets within the flow. These two parameters do not change significantly with different scenarios like the threshold ε, therefore this method has good adaptability to different scenarios.

[0058] The specific process is as follows: Select an unvisited data point from one of the data packets within the network flow. If the neighborhood density of this data point exceeds the neighborhood radius between data packets, then designate it as the core point of a new cluster. Add all unvisited data points within the neighborhood radius between the core point and the data packets to this cluster, and continue to recursively expand the cluster. If the neighborhood density of the core point is insufficient to form a cluster, then mark it as a noise point. Repeat the above steps until all data points have been visited. The advantage of this method is that it can discover clusters of arbitrary shapes and can effectively handle noise and outliers.

[0059] After clustering, each network flow can be viewed as a combination of multiple intra-flow bursts. For example... Figure 3 As shown, the sequential connection of bursts within the same network flow represents their spatial context relationship. The sequential connection of bursts within the same flow at the same time represents their temporal context relationship. This mainly includes the following steps: To characterize inter-flow information with fine granularity, it is necessary to combine the spatial and temporal context features between network flows. This flow interaction graph actually describes the spatial context features horizontally and the temporal context features vertically. Simultaneously, using bursts within the flow as nodes, an objectively sized flow interaction graph is built within each smaller time window, enabling the model to analyze traffic relationships with finer granularity and better meet the accuracy requirements under small identification windows. The flow interaction graph of bursts within the flow can be constructed and updated based on real-time data, making the analysis and response to network conditions faster and more timely. This provides convenience for real-time monitoring, security emergency response, and network management. Furthermore, the flow interaction graph of bursts within the flow can visually display the complex relationships of network traffic, making the understanding of network phenomena more intuitive.

[0060] Traditional recurrent neural networks (RNNs) and convolutional neural networks (CNNs) primarily focus on local and static features in sequential data. GRU, building upon RNNs, introduces a gating mechanism, enabling it to better handle gradients and time-varying information present in long sequences. The network stream embedding method based on the bidirectional recurrent neural network GRU mainly comprises two parts: a feature extraction network and a data embedding network. For example... Figure 4 As shown,

[0061] Feature extraction networks typically employ multiple convolutional and pooling layers to extract feature information from the input sequence. In the diagram, the network stream contains long sequences L1, L2, ..., L... n Transform into embedded sequences e1, e2, ..., e n Data embedding networks utilize GRU to model and predict information in the input sequence. The embedding sequence is sequentially input into a bidirectional GRU, and the hidden layers of the GRU are ultimately used as burst embeddings.

[0062] Compared to traditional recurrent neural networks, GRU-based network flow embedding methods can better handle gradients and time-varying information when processing sequential data, thereby improving model performance. Furthermore, GRU's gating mechanism effectively avoids problems such as vanishing and exploding gradients, making the model more robust when processing long sequences of data.

[0063] Among them, such as Figure 5 As shown, the training process using a two-layer GNN includes:

[0064] First, input a traffic interaction graph consisting of nodes and edges. This graph represents the interaction and traffic propagation between different nodes.

[0065] Use the first GNN layer to capture local information about the nodes;

[0066] A new node representation vector is obtained by weighted summing the node representation vector after capturing local information with the original node representation vector.

[0067] Use a second GNN layer to capture the distance between nodes;

[0068] The node representation vector and the distance representation vector are merged into a global representation vector, which contains the features of the node itself and the distance information between nodes.

[0069] The node classification process includes:

[0070] During the sampling phase, some nodes and their neighboring nodes are selected from the graph for sampling to obtain the local neighborhood graph of the nodes;

[0071] In the aggregation phase, for each sampled node, the set of its neighboring nodes is aggregated into a set, and a random node is selected. The set of neighboring nodes is used as input to obtain the global neighborhood graph of the node.

[0072] During the update phase, the neural network parameters are updated using the global neighborhood graph and label information to minimize the loss function. The above steps are repeated until the preset number of iterations is reached or the stopping condition is met.

[0073] Wherein, if the network flow packet length sequence is v and the generated embedding sequence is u, the process of intra-flow burst embedding of the network flow packet length sequence through RNN is described as follows:

[0074] f(v)=RNN(v)=u.

[0075] The similarity is represented as follows:

[0076]

[0077] The similarity between all network flows belonging to the same category is represented as follows:

[0078]

[0079] The relationship between objective function modeling and overall similarity is expressed as follows:

[0080]

[0081] Where #(v1,v2) represents the number of times the traffic pair (v1,v2) appears in a certain category, I represents the category, and v1,v2 is a long sequence of network flow packets.

[0082] According to another aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the encrypted traffic analysis method based on intra-flow burst interaction graph as described in any of the above claims.

[0083] like Figure 6 As shown in the table below, which compares the experimental data of this method with FOAP, the existing method of this invention has a similar recognition effect on encrypted traffic applications as FOAP, and supports multi-application traffic classification.

[0084]

[0085] We ranked the recognition performance of various application behaviors according to their index values ​​to analyze the overall performance of each individual application behavior category. The ranking analysis of individual application behavior recognition performance is as follows: Figure 6As shown, overall, the GNN-based inter-flow feature extraction model performs better in the "app" recognition type, while the FOAP model performs better in the "app" recognition type. App recognition is relatively good, but there is still a gap between behavior recognition and FOAP. For behavior recognition, 60% of apps have an accuracy rate greater than 80%, and almost all apps have an accuracy rate greater than 60%.

[0086] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0091] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An encrypted traffic analysis method based on intra-flow burst interaction graphs, characterized by: It includes the following steps: Step 1: Cluster the network flows, extract the long sequence of network flow packets in each network flow, and continue to extract intra-flow bursts for each network flow; Step 2: Construct a flow interaction graph of intra-flow bursts, treating each intra-flow burst as a node. Intra-flow bursts within the same network flow are connected sequentially to represent their spatial context relationship. Intra-flow bursts generated at the same time are also connected sequentially to represent their temporal context relationship. Step 3: Generate node embeddings for each burst within the flow using a two-layer bidirectional recurrent neural network; the bidirectional recurrent neural network includes a feature extraction network and a data embedding network; The feature extraction network uses RNN to extract feature information from the input network stream packet long sequence, including transforming the network stream packet long sequence into an embedding sequence; the data embedding network uses bidirectional GRU to model and predict the feature information in the input network stream packet long sequence, including inputting the embedding sequence into the bidirectional GRU in sequence, and finally using the hidden layer of the bidirectional GRU as an intra-stream burst embedding. Step 4: Use a two-layer GNN for training. For node classification, classify the features of each node directly to predict the fine-grained fingerprint corresponding to each burst in the flow. For graph classification, the representations of each node in the output are averaged and pooled to generate an overall traffic fingerprint. Finally, a fully connected layer is used for traffic identification.

2. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 1, characterized in that, The steps of network flow clustering include: selecting an unvisited data point from one of the packets in the network flow; if the neighborhood density of this data point exceeds the neighborhood radius between packets, then designating it as the core point of a new cluster; adding all unvisited data points within the neighborhood radius between packets of this core point to the cluster, and recursively expanding the cluster; if the neighborhood density of this core point is insufficient to form a cluster, then marking it as a noise point; repeating the above steps until all data points have been visited.

3. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 1, characterized in that, The construction steps of the intra-flow burst interaction graph include: combining the spatial and temporal context relationships between network flows to describe the information between network flows in a fine-grained manner; at the same time, taking intra-flow bursts as nodes, a flow interaction graph is established within each time window, and the size of the time window is determined by the type of downstream task.

4. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 1, characterized in that, Intra-flow burst embedding is used to calculate the similarity between two network flows, and unknown traffic is distinguished based on the calculated similarity.

5. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 1, characterized in that, The process of training a two-layer GNN includes: First, input a traffic interaction graph consisting of nodes and edges. This graph represents the interaction and traffic propagation between different nodes. Use the first GNN layer to capture local information about the nodes; A new node representation vector is obtained by weighted summing the node representation vector after capturing local information with the original node representation vector. Use a second GNN layer to capture the distance between nodes; The node representation vector and the distance representation vector are merged into a global representation vector, which contains the features of the node itself and the distance information between nodes.

6. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 5, characterized in that, The node classification process includes: During the sampling phase, some nodes and their neighboring nodes are selected from the graph for sampling to obtain the local neighborhood graph of the nodes; In the aggregation phase, for each sampled node, the set of its neighboring nodes is aggregated into a set, and a random node is selected. The set of neighboring nodes is used as input to obtain the global neighborhood graph of the node. During the update phase, the neural network parameters are updated using the global neighborhood graph and label information to minimize the loss function. The above steps are repeated until the preset number of iterations is reached or the stopping condition is met.

7. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 4, characterized in that, If the network flow packet length sequence is v and the generated embedding sequence is u, the process of intra-flow burst embedding of the network flow packet length sequence through RNN is described as follows: 。 8. The encrypted traffic analysis method based on intra-flow burst interaction graph as described in claim 7, characterized in that, Similarity is expressed as follows: ; Similarity among all network flows belonging to the same category It is expressed as follows: ; The relationship between objective function modeling and overall similarity is expressed as follows: ; in, For traffic pairs The number of times something appears in a certain category, where I represents the category. It is a pair of network flow packet long sequences.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the encrypted traffic analysis method based on intra-flow burst interaction graph as described in any one of claims 1 to 8.

Citation Information

Patent Citations

  • Homologous binary file retrieval method and apparatus

    CN109063055A

  • Fine-grained encrypted website fingerprint classification method and device based on graph attention pooling network

    CN114510615A