Device network access behavior detection method and system based on traffic fingerprint distillation

Through the equipment network access behavior detection method based on traffic fingerprint distillation, the problem of difficulty in identifying the network access behavior of unauthorized mobile devices in the existing technology is solved, and the accurate identification and model classification of mobile device network access behavior is realized, and the detection accuracy and coverage are improved. scope.

CN117857201BActive Publication Date: 2024-11-26INSTITUTE OF INFORMATION ENGINEERING CHINESE ACADEMY OF SCIENCES
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410078748.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-19
Publication Date
2024-11-26
Estimated Expiration
2044-01-19

AI Technical Summary

Technical Problem

Existing network traffic management technologies are difficult to effectively identify and monitor unauthorized mobile device access behavior, especially in the environment of built-in equipment policy. Traditional endpoint detection methods are costly and difficult to deploy, and privacy regulations limit intermittent network access. detection.

Method used

Using the equipment network behavior detection method based on traffic fingerprint distillation, the equipment network network traffic fingerprint distillation algorithm and mobile device fingerprint construction algorithm are designed to build the network behavior traffic fingerprint of each type of mobile device, and the fingerprint is optimized to improve recognition accuracy and reduce errors. Report rate, real-time detection of mobile device network access behavior and classification device models.

Benefits of technology

It realizes accurate identification of mobile device network access behavior and classification of device models, improves detection accuracy and coverage, reduces false alarm rate, and is suitable for network monitoring in complex network environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117857201B_ABST
    Figure CN117857201B_ABST
Patent Text Reader

Abstract

The present invention discloses a device network access behavior detection method and system based on traffic fingerprint distillation, the method comprising: collecting traffic data generated when different mobile devices join an internal local area network, segmenting the network access traffic based on a single network access event within a certain period of time; extracting domain name features in each network access event based on segmented network access traffic, and obtaining the initial fingerprint of each type of mobile device by constructing an initial domain name forest for each type of mobile device; distilling and optimizing the initial fingerprint of each type of mobile device; auditing the network traffic passing through based on the optimized mobile device fingerprint, and obtaining the device network access behavior detection result. The present invention can realize the identification of mobile device network access behavior and the classification of specific device models.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of network traffic management, relates to network traffic behavior detection technology, and specifically relates to a device network access behavior detection method and system based on traffic fingerprint distillation. Background Art

[0002] The goal of network traffic management techniques is to extract as much useful information as possible from authorized network domains. Within the context of enterprise networks, especially those subject to strict confidentiality agreements, a high degree of network monitoring is essential. In these complex network configurations, unauthorized devices are strictly prohibited from accessing the network due to inherent security risks, including the potential for privacy violations. Unauthorized personal devices, including mobile phones and personal computers, can infiltrate internal networks using wireless hotspots provided by authorized devices and routers. This access method remains covert and difficult to detect for network administrators.

[0003] Traditionally, network operators have adopted active approaches to identify network access behaviors, leveraging software that can be easily controlled and deployed on authorized devices. These approaches have been extensively studied in both industry and academia. A notable example is the implementation of endpoint detection and response (EDR) systems, known for their ability to identify access behaviors as soon as a device equipped with monitoring software is connected to the network. Although such client / server-based monitoring systems excel in accurately detecting network access, the underestimation of the costs associated with developing and maintaining such systems poses a practical challenge to the actual deployment of these approaches. On the one hand, poor universality is a problem due to the complexity of developing monitoring software on various platforms and operating systems (OS). The associated maintenance efforts, including debugging and version updates, further increase the cost of these monitoring solutions. Due to the wide variety of devices, it is difficult to deploy monitoring software on all potential access devices, especially when adopting a bring your own device (BYOD) policy. On the other hand, software deployment may be restricted due to privacy regulations, hindering the detection of mobile devices with intermittent network access.

[0004] With the development of network traffic analysis, inferring access behavior from network traffic provides a way to bypass the constraints imposed by endpoint-based approaches. Traffic-based analyzers enhance accessibility to network management operators, especially in BYOD policy environments where these operators do not have direct control over devices. Deployed at the gateway of authorized networks, these analyzers can effectively detect network access, improve pervasiveness and coverage, and identify device models. Summary of the invention

[0005] The purpose of the present invention is to provide a device network access behavior detection method and system based on traffic fingerprint distillation, focusing on utilizing the designed device network access traffic fingerprint distillation algorithm, mobile device fingerprint construction algorithm and real-time device network access behavior detection algorithm to respectively construct the network access behavior traffic fingerprint of each type of mobile device, and the fingerprint of each type of mobile device constructed by distillation and optimization is used to audit the traffic passing through the network gateway, and detect the network traffic generated by the mobile device network access behavior in real time, so as to realize the identification of the mobile device network access behavior and the classification of the specific model of the device.

[0006] The present invention first collects the pure flow data generated when different models of mobile devices access the network, and divides the network access flow into units of single network access events within a certain period of time; uses a mobile device fingerprint construction algorithm to extract domain name features in each network access event, and constructs an initial domain name forest for each type of mobile device to initialize the fingerprint of each type of mobile device; uses a device network access flow fingerprint distillation algorithm to improve the model-related domain name tree in the fingerprint of each type of mobile device, and at the same time reduces the domain name tree of background flow in the fingerprint of each type of mobile device, thereby optimizing the fingerprint matching accuracy and false alarm rate; then, uses a real-time device network access behavior detection algorithm, in conjunction with the optimized device fingerprint, to audit the network flow passing through, identify the network access behavior of the mobile device and classify the specific model of the device.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A method for detecting device network access behavior based on traffic fingerprint distillation, the method comprising:

[0009] Collect the traffic data generated when different mobile devices join the internal LAN, and divide the incoming traffic into units of single network access events within a certain period of time;

[0010] Based on the segmented network traffic, domain name features in each network access event are extracted, and an initial fingerprint of each type of mobile device is obtained by constructing an initial domain name forest for each type of mobile device;

[0011] Distill and optimize the initial fingerprint of each type of mobile device;

[0012] Based on the optimized mobile device fingerprint, the network traffic passing through the audit is audited to obtain the device network access behavior detection results.

[0013] Furthermore, the initial fingerprint of each type of mobile device is obtained by constructing an initial domain name forest for each type of mobile device, including:

[0014] Counting the domain names of all network access events of each type of mobile device according to the domain name features, and generating a domain name feature pair set including the domain name features and the domain name counting results;

[0015] For each set of domain name feature pairs of mobile devices, a corresponding domain name tree is constructed based on the domain name feature pairs with the same top-level domain name i. Where y is the mobile device category, represents a set of nodes, It is a set of edges constructed based on two nodes with a first-order bias relationship. Represents the attribute set of a node;

[0016] Generate an initial domain forest for each type of mobile device And the initial domain name forest dnForest y As the initial fingerprint of mobile device category y; where m represents the number of top-level domain names corresponding to mobile devices of device category y.

[0017] Furthermore, the attributes of the node include: domain name characteristics, domain name count and domain name depth level.

[0018] Furthermore, the distillation and optimization of the initial fingerprint of each type of mobile device includes:

[0019] Calculate each domain name tree The complexity of the domain tree W y,i ;

[0020] According to the complexity of the domain name tree W y,i Adjust the domain name counting results to get the domain name tree Leaf Node The adjusted weight

[0021] According to the initial adjusted weights The IDF and TF values ​​of the initial adjusted weights Perform secondary adjustments to obtain optimized mobile device fingerprints.

[0022] Furthermore, the calculation of each domain name tree The complexity of the domain tree W y,i ,include:

[0023] Get the domain name tree All decision path sequences included

[0024] By multiplying the decision path sequence The count quotient of the two adjacent nodes on the same node is used to obtain the decision path sequence. Contribution

[0025] By accumulating the contribution of each decision path sequence Get the domain name tree The complexity of the domain tree W y,i .

[0026] Furthermore, the weight after the initial adjustment IDF value Where Y is the set of device categories, and device category y * Belongs to Y, Domain Forest The set of nodes in a is the judgment condition.

[0027] Furthermore, the weight after the initial adjustment TF value V y Represents the node set in the domain name forest corresponding to type y mobile devices.

[0028] Furthermore, the network traffic passing through the mobile device fingerprint audit after optimization is obtained to obtain the device network access behavior detection result, including:

[0029] Taking the DNS request as the detection start mark, obtain the network traffic T seconds after the DNS request;

[0030] Extract the domain name feature set D′ corresponding to the network traffic T seconds after the DNS request;

[0031] Based on the optimized mobile device fingerprint, the domain name feature set D′ is matched with confidence to obtain a confidence score; wherein the optimized mobile device fingerprint is deployed at the network gateway between the intranet and the Internet;

[0032] If the confidence score of any fingerprint is higher than the threshold, it is identified as a mobile device accessing the network, and the specific model of the mobile device accessing the network is identified by retrieving the matching fingerprint;

[0033] If the confidence scores of all fingerprints are lower than the threshold, it is determined as non-device network access behavior.

[0034] Furthermore, the confidence score of the mobile device of type y is Among them, V y represents the node set in the domain name forest corresponding to the mobile device of type y, F y Represents the domain name forest dnForest corresponding to class y mobile devices y , Indicates the domain name forest dnForest y Domain name tree in The corresponding domain name set, Is a collection of domain names A node in Represents a node The optimized fingerprint, d k Represents the kth domain name feature in the domain name feature set D′.

[0035] A device network access behavior detection system based on traffic fingerprint distillation, the system comprising:

[0036] The pre-processing module is used to collect the traffic data generated when different mobile devices join the internal LAN, and divide the incoming traffic into units of single network access events within a certain period of time;

[0037] A device fingerprint initialization module, used to extract domain name features in each network access event based on the segmented network access traffic, and obtain an initial fingerprint for each type of mobile device by constructing an initial domain name forest for each type of mobile device;

[0038] Traffic fingerprint distillation optimization module, used to distill and optimize the initial fingerprint of each type of mobile device;

[0039] The real-time device network access behavior detection and classification module is used to audit the network traffic passing through based on the optimized mobile device fingerprint to obtain the device network access behavior detection results.

[0040] Compared with the prior art, the present invention can use the traffic generated by the known mobile device network access behavior to construct the corresponding network access behavior traffic fingerprint, and use the device network access traffic fingerprint after distillation optimization to audit the traffic passing through the deployment node, so as to detect and discover the traffic generated by the mobile device network access behavior, and classify the specific models of the mobile device:

[0041] 1. The present invention proposes a method for constructing a network traffic fingerprint of a mobile device's network access behavior. This method can be used to construct the network traffic data generated when a mobile device accesses a local area network into a domain name forest based on a domain name tree set, thereby constructing an initial device network access traffic fingerprint.

[0042] 2. The present invention proposes a distillation optimization algorithm for the mobile device access traffic fingerprint. This method can increase the weight of the domain name tree that is strongly related to the corresponding category in the initial device access traffic fingerprint, and reduce the weight of the domain name tree generated by the background traffic that is not related to the corresponding category, thereby distilling and optimizing the purity of the fingerprint, improving the accuracy of fingerprint matching, and reducing the false alarm rate of background traffic.

[0043] 3. The present invention proposes a method for real-time detection and classification of device network access behavior traffic. The method audits the network traffic in a lightweight and real-time manner based on the device network access traffic fingerprint after distillation optimization, detects and discovers the traffic generated by the mobile device network access behavior, and classifies the specific models of the mobile devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of the proposed device network access behavior detection method and system based on traffic fingerprint distillation.

[0045] Figure 2 It is an example diagram of the proposed method for real-time detection and classification of device network access behavior traffic. DETAILED DESCRIPTION

[0046] The present invention is further described in detail below through specific embodiments and drawings.

[0047] The device network access behavior detection method and system based on traffic fingerprint distillation provided by the present invention are suitable for real-time detection and classification of mobile device network access behavior and specific device models from traffic. The schematic diagram of the present invention is shown in FIG. Figure 1 , as shown in 2. Figure 1 This is a flowchart of the proposed device network access behavior detection method and system based on traffic fingerprint distillation. Figure 2 It is an example diagram of the proposed method for real-time detection and classification of device network access behavior traffic.

[0048] The present invention constructs a traffic fingerprint of the network access behavior of a known mobile device, optimizes the initial fingerprint through a distillation optimization algorithm, and uses the optimized fingerprint to audit the traffic passing through the network gateway, thereby detecting and discovering the network access behavior of the mobile device, and classifying the specific model information of the mobile device. The present invention first collects the traffic data generated when the mobile device accesses the local area network, and then extracts the domain name feature set in each pre-processed event; then uses the designed mobile device fingerprint construction algorithm to construct the initial device network access traffic fingerprint; then uses the designed device network access traffic fingerprint distillation algorithm to optimize the initial device fingerprint to redistribute the weight of the domain name tree in the fingerprint; then uses the designed real-time device network access behavior detection algorithm to audit the network traffic passing through in real time to detect and discover the network access behavior of the mobile device, and classify the specific model information of the device.

[0049] A method and system for detecting device network access behavior based on traffic fingerprint distillation of the present invention comprises the following steps:

[0050] 1) Mobile device fingerprint construction algorithm: First, collect the traffic data generated when different models of mobile devices join the internal LAN, and divide the network traffic into units of single network access events within a certain period of time. For different models (categories) of mobile devices, collect the network traffic data generated within a certain period of time when they access the internal LAN through a wired network or a wireless network, and divide the traffic data into units of single network access events to obtain a set of network access traffic events for each model (category). For each network access traffic event in the set of network access traffic events, extract domain name features from the DNS traffic, TLS / SSL traffic and HTTP traffic to form a domain name feature set for each network access traffic event. For each domain name feature set, aggregate the domain name features with the same top-level domain name, and construct domain name trees respectively. The domain name tree set forms a domain name forest for the network access behavior of each type of mobile device, which is constructed as the initial network access behavior fingerprint.

[0051] 2) Device access traffic fingerprint distillation algorithm: For the initial access behavior fingerprint of each type of mobile device, first, calculate the complexity of each domain name tree, recalculate the weight of the nodes in each domain name tree according to the complexity of each domain name tree, increase the weight of the domain name tree related to the device model with high complexity, and reduce the weight of the domain name tree related to the background traffic with low complexity; for the access behavior fingerprint optimized in the first step, then calculate the intra-class similarity of the domain name features in each domain name tree in the domain name feature set of this category, and at the same time calculate its inter-class similarity in the domain name feature set of all categories, further improve the domain name features with high intra-class similarity and low inter-class similarity, and reduce the domain name features with low intra-class similarity and high inter-class similarity.

[0052] 3) Real-time device network access behavior detection algorithm: The optimized device network access traffic fingerprint is deployed at the network gateway between the intranet and the Internet. The DNS request is used as the detection start mark. The network traffic for a period of time is audited and matched with the optimized fingerprint. If the matching confidence is higher than the threshold, it is identified as a mobile device network access behavior, and the matching fingerprint is retrieved to identify the specific model of the network access device. If all fingerprints are lower than the threshold, it is determined as non-device network access behavior.

[0053] In a preferred embodiment of the present invention, the device network access traffic collection technology described in the mobile device fingerprint construction algorithm is as follows: for each model (category) of mobile device, it is connected to the internal LAN through a wired network or a wireless network, and the traffic data generated K seconds after the mobile device accesses the network is sniffed and captured as a single device network access event.

[0054] In a preferred embodiment of the present invention, the domain name feature extraction technology described in the mobile device fingerprint construction algorithm is as follows: for each device network access event, domain name information is extracted from the DNS traffic, TLS / SSL traffic and HTTP traffic respectively, and the domain names of all network access events of each type of device are counted to form a domain name feature pair set.

[0055] The following is a formal representation of a set of domain name feature pairs:

[0056]

[0057] Among them, D y is the set of domain-feature pairs of device category y, is the extracted domain name information (domain name features), is the count corresponding to the domain name information, N y is the number of domain names extracted for device category y.

[0058] In a preferred embodiment of the present invention, the domain name tree construction technology described in the mobile device fingerprint construction algorithm: for each set of domain name feature pairs of each type of mobile device, the same domain name tree is constructed for the domain name feature pairs with the same top-level domain name.

[0059] The following is a formal representation of the domain name tree:

[0060]

[0061] in, is the i-th domain name tree in device category y. Each domain name tree is a directed acyclic graph with a node set Used to indicate that in D y Domain name feature pairs with the same top-level domain name in .

[0062] The following is a set of nodes The formal expression of:

[0063]

[0064] in, It is the domain name information with the same top-level domain name. is the count corresponding to the domain name, and is a set of nodes The two domain name information in . It is the edge set of two domain name information with a first-order bias relationship in the same domain name tree. Between two domain names with a first-order bias relationship, one domain name is a subdomain of the other domain name, such as store.m.apple.com and m.apple.com have a first-order bias relationship, but store.m.apple.com and apple.com, store.m.apple.com and app.m.apple.com do not have a first-order bias relationship. Used to represent the attributes of each node, including domain name information Domain Count Domain depth level

[0065] In a preferred embodiment of the present invention, the domain name forest construction technology described in the mobile device fingerprint construction algorithm is as follows: for each type of mobile device, a domain name forest is constructed respectively, and the constructed domain name forest is the initial device network access traffic fingerprint.

[0066] The following is a formal expression of the domain forest:

[0067]

[0068] Among them, dnForest y Domain forests for device category y. Each domain forest contains all domain trees generated in that category.

[0069] In a preferred embodiment of the present invention, the domain name tree complexity calculation technology described in the device network traffic fingerprint distillation algorithm is as follows: for each domain name tree, its complexity is calculated by measuring the number of its decision paths, depth and number of leaf nodes. Given a domain name tree with n leaf nodes from device type y It has a set of n decision paths Θ y,i .

[0070] The following is the set of decision paths Θ y,i The formal expression of:

[0071]

[0072]

[0073] in, is the set of decision paths Θ y,i A decision path sequence with a depth of L in is the node with depth l in the decision path sequence. Given a node Returns the child nodes of this node. If is a leaf node, then Return the domain name count of the node. Then, according to the decision path set Θ y,i , calculate the complexity of each domain name tree.

[0074] The following is a formal expression for calculating the complexity of the domain name tree based on a set of decision paths:

[0075]

[0076]

[0077] Among them, W y,i is the complexity of the domain name tree, which is calculated by accumulating the contribution of each decision path Get the contribution of the decision path It is obtained by multiplying the count quotients of two adjacent nodes on a decision path, and σ is a predefined noise tolerance hyperparameter.

[0078] In a preferred embodiment of the present invention, the domain name tree preliminary weight adjustment technology described in the device network traffic fingerprint distillation algorithm: using the complexity calculated by each domain name tree mentioned above, readjust the weight of each leaf node in the domain name tree, given a node The raw count The initial adjusted weights are

[0079] In a preferred embodiment of the present invention, the domain name tree secondary weight adjustment technology described in the device network traffic fingerprint distillation algorithm: first, for each leaf node count in each domain name tree Calculate its IDF value and TF value.

[0080] The following is a formal expression for calculating the IDF value:

[0081]

[0082] Where Y is the set of device categories, and device category y * Belongs to Y. Domain Forest The set of nodes in is the IDF value of the node.

[0083] The following is a formal expression of δ(·):

[0084]

[0085] Among them, a is the judgment condition. When the judgment condition is true, the function returns 1, otherwise it returns 0.

[0086] The following is a formal expression of TF value calculation:

[0087]

[0088] in, is the TF value of the leaf node, is the count of leaf nodes, |V y | is the number of nodes of device type y. Then, the weight of the leaf node is updated again according to the obtained IDF value and TF value.

[0089] The following is a formal expression of quadratic weight optimization:

[0090]

[0091] in, is the weight after secondary optimization, The node obtains the TF value through calculation. The node obtains the IDF value through calculation. is the weight of the node after initial optimization.

[0092] In a preferred embodiment of the present invention, the real-time device network access behavior traffic fingerprint matching technology described in the real-time device network access behavior detection algorithm: audit the traffic passing through the gateway, capture the traffic T seconds after the DNS request, extract the domain name features in the traffic segment, and use the fingerprint to perform confidence matching on the domain name feature set.

[0093] The following is a formal expression of confidence matching:

[0094]

[0095]

[0096] Where D′ is the domain name feature set of the extracted T-second traffic, F y The domain forest dnForest representing device category y y . Is a collection of domain names One of the nodes. The function T(·,·) is used to return the matching result. If dk is If the match is successful, the count is returned. Otherwise, it returns 0. Then, the matching confidence threshold τ is set. If the fingerprint matching confidence of a certain category is greater than the threshold τ, the Score y >τ, the fingerprint match is successful, and it is detected that the traffic of T seconds contains the network access behavior of the device of this category.

[0097] Example 1: Use the mobile device fingerprint construction algorithm and the device network traffic fingerprint distillation algorithm to construct and optimize the device network traffic fingerprint

[0098] Taking the NetCess2023 dataset as an example dataset, the NetCess2023 dataset contains the network access traffic data of 7 mobile device brands and 26 mobile device models, with a data volume of 10G. The traffic data has been divided into independent network access events according to the time of 5s, 10s and 15s after network access:

[0099] 1) First, for the 26 categories of network access event sets in the NetCess2023 dataset, the domain name features of each network access event are extracted respectively, and the domain name features are extracted from the DNS traffic, TLS / SSL traffic and HTTP traffic respectively to form a 26-category domain name feature set;

[0100] 2) For each of the 26 domain name feature sets obtained in 1), an initialized domain name forest is constructed;

[0101] 3) In a domain name forest, the same domain name tree is constructed for domain name features with the same top-level domain name, and 26 domain name forests of 26 categories are obtained, forming 26 initial device network access behavior traffic fingerprints;

[0102] 4) Calculate the complexity of each domain name tree in each domain name forest obtained in 3), and calculate its complexity by measuring the number of decision paths, depth and number of leaf nodes to obtain the complexity of each domain name tree;

[0103] 5) Using the complexity of each domain name tree calculated in 4), readjust the weight of each leaf node in the domain name tree;

[0104] 6) For the node weights readjusted in 5), it is necessary to optimize and adjust all fingerprints again, so that the fingerprint information related to the device category is more prominent, and the fingerprint information generated by the background traffic that is not related to the device category is weakened. For each leaf node in 5), calculate its TF value and IDF value, and then update the weights of all leaf nodes simultaneously by using the TF value and IDF value;

[0105] 7) Re-adjust and normalize the weights of the domain name tree and domain name forest that have undergone secondary optimization in 6) to obtain the device network access behavior traffic fingerprint that has been distilled and optimized.

[0106] Example 2: Use real-time device network access behavior detection algorithm to detect and discover device network access behavior in real time

[0107] Use the traffic fingerprints of the 26 categories of device access behaviors obtained in Example 1 to audit the traffic passing through the gateway.

[0108] 1) Monitor the network traffic data passing through the gateway, capture the traffic 5s, 10s, and 15s after the DNS request traffic, extract the domain name feature set of the captured traffic, match it with the traffic fingerprints of 26 categories of device access behavior, calculate the matching confidence, set the threshold τ to 0.6, and classify and alarm the traffic with a matching confidence higher than 0.6.

[0109] The results of the present invention and other methods (FS-Net, FlowPrint, ProGraph) under real-time traffic are compared, and the results of the detection rate DR (%) and the false alarm rate FAR (%) of the background traffic are compared.

[0110] Table 1. Comparison of the detection rate DR (%) of the device access behavior and the false alarm rate FAR (%) of the background traffic of this method with other methods in real-time traffic, with a detection time of 5s

[0111] DR (%) FAR(%) FS-Net 62.07 19.15 FlowPrint 61.70 14.29 ProGraph 63.95 2.90 Method of the present invention 69.07 0.08

[0112] Table 2. Comparison of the detection rate DR (%) of the device access behavior and the false alarm rate FAR (%) of the background traffic of this method with other methods in real-time traffic, with a detection time of 10s

[0113] DR (%) FAR(%) FS-Net 67.89 12.87 FlowPrint 79.59 8.94 ProGraph 76.63 3.56 Method of the present invention 83.53 0.15

[0114] Table 3. Comparison of the detection rate DR (%) of the device access behavior and the false alarm rate FAR (%) of the background traffic of this method with other methods on real-time traffic, with a detection time of 15s

[0115] DR (%) FAR(%) FS-Net 72.56 11.95 FlowPrint 93.12 6.95 ProGraph 85.20 3.69 Method of the present invention 96.83 0.25

[0116] According to the results in Table 1, Table 2 and Table 3, the superiority of the device network access behavior detection method and system based on traffic fingerprint distillation proposed by the method of the present invention is reflected.

[0117] Based on the same inventive concept, another embodiment of the present invention provides a device network access behavior detection system based on traffic fingerprint distillation, which includes:

[0118] The pre-processing module is used to collect the traffic data generated when mobile devices of different models and categories access the local area network, and divide the network access traffic into units of single network access events within a certain period of time;

[0119] The device fingerprint initialization module is used to extract the domain name feature set of the network access event, construct the domain name feature set into a domain name tree and a domain name forest, and form the initial mobile device traffic fingerprint;

[0120] Traffic fingerprint distillation optimization module, which is used to increase the weight of the device-related domain name tree in the initial mobile device traffic fingerprint and reduce the weight of the background-related domain name tree, thereby optimizing the fingerprint matching accuracy and false positive rate;

[0121] The real-time device network access behavior detection and classification module is used to audit the network traffic in real time based on the device network access traffic fingerprint after distillation optimization, detect and discover the traffic generated by the mobile device network access behavior, and classify the specific models of the mobile devices.

[0122] The specific implementation process of each module refers to the above description of the method of the present invention.

[0123] Based on the same inventive concept, another embodiment of the present invention provides an electronic device (computer, server, smart phone, etc.), which includes a memory and a processor, the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program includes instructions for executing each step in the method of the present invention.

[0124] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium (such as ROM / RAM, disk, CD), which stores a computer program. When the computer program is executed by a computer, it implements the various steps of the method of the present invention.

[0125] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. A person skilled in the art may modify or make equivalent substitutions for the technical solutions of the present invention without departing from the spirit and scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. A device network access behavior detection method based on traffic fingerprint distillation, characterized in that: The method comprises: Collect the traffic data generated when different mobile devices join the internal LAN, and divide the incoming traffic into units of single network access events within a certain period of time; Based on the segmented network traffic, the domain name features in each network access event are extracted, and an initial domain name forest is constructed for each type of mobile device to obtain an initial fingerprint of each type of mobile device; wherein, the initial fingerprint of each type of mobile device is obtained by constructing an initial domain name forest for each type of mobile device, including: Counting the domain names of all network access events of each type of mobile device according to the domain name features, and generating a domain name feature pair set including the domain name features and the domain name counting results; For each set of domain name feature pairs of mobile devices, a corresponding domain name tree is constructed based on the domain name feature pairs with the same top-level domain name i. Where y is the mobile device category, represents a set of nodes, is a set of edges constructed based on two nodes with a first-order bias relationship, P i y Represents the attribute set of a node; Generate an initial domain forest for each type of mobile device And the initial domain name forest dnForest y as the initial fingerprint of mobile device category y; wherein m represents the number of top-level domain names corresponding to mobile devices of device category y; Distilling and optimizing the initial fingerprint of each type of mobile device; wherein the distilling and optimizing the initial fingerprint of each type of mobile device includes: Calculate each domain name tree The complexity of the domain tree W y,i ; Wherein, the calculation of each domain name tree The complexity of the domain tree W y,i ,include: Get the domain name tree All decision path sequences included By multiplying the decision path sequence The count quotient of the two adjacent nodes on the same node is used to obtain the decision path sequence. Contribution By accumulating the contribution of each decision path sequence Get the domain name tree The complexity of the domain tree W y,i ; According to the complexity of the domain name tree W y,i Adjust the domain name counting results to get the domain name tree Leaf Node The adjusted weight Among them, the initial adjusted weight IDF value Y is a set of device categories, and device category y * Belongs to Y, V y* It is the domain name forest dnForest y* The set of nodes in a is the judgment condition; According to the initial adjusted weights The IDF and TF values ​​of the initial adjusted weights Perform secondary adjustments to obtain optimized mobile device fingerprints; Based on the optimized mobile device fingerprint, the network traffic passing through the audit is audited to obtain the device network access behavior detection results.

2. The method according to claim 1, characterized in that The attributes of the node include: domain characteristics, domain count and domain depth level.

3. The method according to claim 1, characterized in that The initial adjusted weights TF value V y Represents the node set in the domain name forest corresponding to type y mobile devices.

4. The method according to claim 1, characterized in that: The network traffic passing through the optimized mobile device fingerprint audit is used to obtain the device network access behavior detection result, including: Taking the DNS request as the detection start mark, obtain the network traffic T seconds after the DNS request; Extract the domain name feature set D corresponding to the network traffic T seconds after the DNS request ′ ; Based on the optimized mobile device fingerprint, the domain name feature set D ′ Perform confidence matching to obtain a confidence score; wherein the optimized mobile device fingerprint is deployed at a network gateway between an intranet and the Internet; If the confidence score of any fingerprint is higher than the threshold, it is identified as a mobile device accessing the network, and the specific model of the mobile device accessing the network is identified by retrieving the matching fingerprint; If the confidence scores of all fingerprints are lower than the threshold, it is determined as non-device network access behavior.

5. The method according to claim 4, characterized in that The confidence score of the mobile device of category y Among them, V y represents the node set in the domain name forest corresponding to the mobile device of type y, F y Represents the domain name forest dnForest corresponding to class y mobile devices y , Indicates the domain name forest dnForest y Domain name tree in The corresponding domain name set, Is a collection of domain names A node in Represents a node The optimized fingerprint, d k Represents the domain name feature set D ′ The k-th domain name feature in .

6. A device network access behavior detection system based on traffic fingerprint distillation, characterized in that: The system comprises: The pre-processing module is used to collect the traffic data generated when different mobile devices join the internal LAN, and divide the incoming traffic into units of single network access events within a certain period of time; The device fingerprint initialization module is used to extract the domain name features in each network access event based on the segmented network access traffic, and obtain the initial fingerprint of each type of mobile device by constructing an initial domain name forest for each type of mobile device; wherein, the initial fingerprint of each type of mobile device is obtained by constructing an initial domain name forest for each type of mobile device, including: Counting the domain names of all network access events of each type of mobile device according to the domain name features, and generating a domain name feature pair set including the domain name features and the domain name counting results; For each set of domain name feature pairs of mobile devices, a corresponding domain name tree is constructed based on the domain name feature pairs with the same top-level domain name i. Where y is the mobile device category, represents a set of nodes, is a set of edges constructed based on two nodes with a first-order bias relationship, P i y Represents the attribute set of a node; Generate an initial domain forest for each type of mobile device And the initial domain name forest dnForest y as the initial fingerprint of mobile device category y; wherein m represents the number of top-level domain names corresponding to mobile devices of device category y; The traffic fingerprint distillation optimization module is used to distill and optimize the initial fingerprint of each type of mobile device; wherein the distillation and optimization of the initial fingerprint of each type of mobile device includes: Calculate each domain name tree The complexity of the domain tree W y,i ; Wherein, the calculation of each domain name tree The complexity of the domain tree W y,i ,include: Get the domain name tree All decision path sequences included By multiplying the decision path sequence The count quotient of the two adjacent nodes on the same node is used to obtain the decision path sequence. Contribution By accumulating the contribution of each decision path sequence Get the domain name tree The complexity of the domain tree W y,i ; According to the complexity of the domain name tree W y,i Adjust the domain name counting results to get the domain name tree Leaf Node The adjusted weight Among them, the initial adjusted weight IDF value Y is a set of device categories, and device category y * Belongs to Y, V y* It is the domain name forest dnForest y* The set of nodes in a is the judgment condition; According to the initial adjusted weights The IDF and TF values ​​of the initial adjusted weights Perform secondary adjustments to obtain optimized mobile device fingerprints; The real-time device network access behavior detection and classification module is used to audit the network traffic passing through based on the optimized mobile device fingerprint to obtain the device network access behavior detection results.