User identification methods, systems, devices, media and products

By acquiring log data and calculating target scores, and adjusting the scores using a directed weighted graph and causal relationship matrix, peer-to-peer content distribution network users are identified. This solves the problem of insufficient user identification accuracy in existing technologies and achieves higher identification accuracy and reliability.

CN123123265APending Publication Date: 2026-11-17CHINA MOBILE GRP FUJIAN CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611173931.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-04
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, peer-to-peer content delivery network user identification methods rely on authentication, authorization, and billing traffic analysis, which leads to a temporary increase in bandwidth and insufficient accuracy in user identification, resulting in false positives and false negatives.

Method used

By acquiring log data, extracting user interaction data with peer-to-peer content delivery network services, calculating target scores based on behavioral characteristics, and adjusting scores by combining directed weighted graphs and causal relationship matrices, the system identifies users who actually use peer-to-peer content delivery networks.

Benefits of technology

This improves the accuracy of user identification, reduces the probability of normal users being misidentified as peer-to-peer content distribution network users, and ensures the reliability of the judgment process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123265A_ABST
    Figure CN123123265A_ABST
Patent Text Reader

Abstract

This disclosure relates to the technical field of network data processing, specifically to a user identification method, system, device, medium, and product, including: acquiring log data; extracting interaction data between users and peer-to-peer content delivery network (P2P) services from the log data; identifying candidate users suspected of using the P2P service based on traffic usage data and interaction data; extracting behavioral characteristics of the candidate users from the log data and calculating a target score based on the behavioral characteristics; and determining the candidate user as an actual user of the P2P service if the target score meets the score requirements. This disclosure acquires information related to P2P user behavior from multiple dimensions, achieving joint determination of traffic and behavioral dimensions, and thereby measuring the likelihood that a user possesses P2P user behavior characteristics, improving the accuracy of user identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of network data processing technology, and more specifically, to a user identification method, system, device, medium, and product. Background Technology

[0002] Peer-to-Peer Content Delivery Network (PCDN) utilizes idle resources of carrier network users to build low-cost content delivery network services, with applications covering video-on-demand, live streaming, and large file downloads. PCDN is based on peer-to-peer (P2P) technology and combines it with traditional Content Delivery Networks (CDNs) to distribute content delivery tasks to multiple user devices, thereby sharing bandwidth and resources and improving content delivery efficiency.

[0003] In related technologies, bandwidth data is usually obtained and analyzed by relying on Authentication Authorization and Accounting (AAA) traffic. However, normal business applications may also cause a temporary increase in bandwidth, resulting in a large number of false positives and false negatives in the method of user identification through AAA traffic, leading to insufficient accuracy in user identification. Summary of the Invention

[0004] This disclosure is made in view of the above-mentioned problems. This disclosure provides a user identification method, system, device, medium, and product.

[0005] In a first aspect, embodiments of this disclosure provide a user identification method, comprising: Acquire log data; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data. Extract interaction data between users and the peer-to-peer content delivery network service from the log data. Based on traffic usage data and interaction data, identify candidate users who are suspected of using the peer-to-peer content delivery network. Extract behavioral characteristics of the candidate users from the log data, and calculate a target score based on the behavioral characteristics; wherein the target score is used to indicate the confidence level of the candidate user's actual use of the peer-to-peer content delivery network. If the target score meets the scoring requirements, determine the candidate user as a user who is actually using the peer-to-peer content delivery network.

[0006] In one embodiment of the first aspect, calculating a target score based on behavioral features includes: Determine the initial score for each behavioral feature; the initial score indicates the degree to which the behavioral feature deviates from the baseline of normal user behavior. Adjust the initial scores based on log data and behavioral features to obtain the adjusted scores for each behavioral feature. Perform a weighted summation of the adjusted scores for each behavioral feature to obtain the target score.

[0007] In one embodiment of the first aspect, the initial score is adjusted based on log data and behavioral characteristics to obtain adjusted scores for each behavioral characteristic, including: A directed weighted graph is constructed based on log data. In this graph, nodes are network node identifiers, directed edges indicate traffic transmission relationships between nodes, and edge weights indicate the proportion of traffic transmission between nodes. Network node identifiers include at least one of the following: user network address, autonomous system number. A causal relationship matrix is ​​constructed based on behavioral features. Rows and columns of the causal relationship matrix correspond to behavioral features, and element values ​​indicate whether a causal relationship exists between the behavioral feature in the row and column. Path scores for anomaly propagation paths are calculated based on the directed weighted graph, and the feature influence strength between behavioral features is determined based on the causal relationship matrix. Feature influence strength is represented by element values ​​in the causal relationship matrix. Initial scores are adjusted based on path scores and feature influence strength to obtain adjusted scores for each behavioral feature.

[0008] In one embodiment of the first aspect, calculating the path score of the anomaly propagation path based on a directed weighted graph includes: Nodes exhibiting abnormal flow behavior in the directed weighted graph are identified as anomalous nodes. A flood propagation algorithm is used to trace the propagation path of the anomalous nodes along the edges of the directed weighted graph. The anomalousness of nodes along the propagation path, the weights between adjacent nodes, and the path depth are determined. A path score is calculated based on the anomalousness, the weights between adjacent nodes, and the path depth.

[0009] In one embodiment of the first aspect, constructing a causal relationship matrix based on behavioral features includes: Extract time-series data for any two behavioral features within a time window. Perform a Granger causality test on the time-series data and obtain the test results; the test results indicate whether historical data of one behavioral feature has predictive power for current data of another behavioral feature. Determine whether a causal relationship exists between the behavioral features based on the test results, and construct a causal relationship matrix based on the obtained determination results.

[0010] In one embodiment of the first aspect, the initial score is adjusted based on the path score and the feature influence strength to obtain the adjusted score for each behavioral feature, including: For any given behavioral feature, determine the mean of the path scores associated with that feature. Based on the causal relationship matrix, determine the influence strength of other behavioral features on that feature, and calculate the product between the influence strength and the initial scores of other behavioral features. Perform a weighted summation on the mean and the product to obtain the processed result. Finally, sum the initial score and the processed result for that behavioral feature to obtain the adjusted score.

[0011] In one embodiment of the first aspect, the adjusted scores corresponding to each behavioral feature are weighted and summed to obtain the target score, including: The weighting coefficients are dynamically determined based on historical data corresponding to each behavioral feature. The adjusted scores for each behavioral feature and their corresponding weighting coefficients are then summed in a weighted manner to obtain the target score.

[0012] In one embodiment of the first aspect, the historical data includes the historical false alarm rate of the behavioral feature and the proportion of the initial score of the behavioral feature in the path score; The weighting coefficients are dynamically determined based on historical data corresponding to each behavioral feature, including: If the historical false alarm rate exceeds the false positive threshold, the weight coefficient of that behavioral feature is reduced. If the proportion exceeds the proportion threshold, the weight coefficient of that behavioral feature is increased.

[0013] In one embodiment of the first aspect, based on traffic usage data and interaction data, candidate users suspected of using a peer-to-peer content delivery network are identified, including: If, based on traffic usage data, it is determined that a user's uplink traffic usage is greater than their downlink traffic usage, the user is assigned a first label. If, based on interaction data, it is determined that there is interaction between the user and the peer-to-peer content delivery network service, the user is assigned a second label. Users with both the first and second labels are identified as candidate users.

[0014] In one embodiment of the first aspect, if it is determined, based on interaction data, that there is an interaction between a user and a peer-to-peer content delivery network service, a second label is applied to the user, including: Extract the domain name information requested by the user from the interaction data. If the domain name information matches the preset peer-to-peer content delivery network service domain name database, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is marked with a second tag.

[0015] In one embodiment of the first aspect, if it is determined, based on interaction data, that there is an interaction between a user and a peer-to-peer content delivery network service, a second label is applied to the user, including: Extract the DNS resolution address corresponding to the user's requested domain name and the target address actually accessed by the user from the interaction data. If the DNS resolution address and the target address match successfully, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is then marked with a second tag.

[0016] In one embodiment of the first aspect, the behavioral characteristics include at least one of the following: traffic scale characteristics reflecting the user's bandwidth usage, port characteristics reflecting the user's access port behavior, protocol characteristics reflecting the user's protocol type, preset interaction characteristics reflecting the user and peer-to-peer content delivery network services, activity characteristics reflecting the user's traffic time distribution pattern, address association characteristics reflecting the consistency between the user's actual access address and the domain name resolution result, and traffic flow characteristics reflecting whether the user's traffic flows across domains.

[0017] In one embodiment of the first aspect, after acquiring the log data, the method further includes: Application layer data from the deep packet inspection system is stored as application layer logs. Traffic usage data from the authentication, authorization, and accounting system is stored as traffic metering logs. Network layer traffic transmission data from the network flow analysis system and network path data from the border gateway protocol system are stored as traffic transmission logs.

[0018] In a second aspect, embodiments of this disclosure provide a user identification system, including: The system includes a log acquisition module for acquiring log data, which includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data; a data extraction module for extracting interaction data between users and the peer-to-peer content delivery network service from the log data; a user identification module for identifying candidate users suspected of using the peer-to-peer content delivery network based on traffic usage data and interaction data; a calculation module for extracting behavioral characteristics of candidate users from the log data and calculating a target score based on the behavioral characteristics, whereby the target score indicates the confidence level of a candidate user's actual use of the peer-to-peer content delivery network; and a user determination module for determining that a candidate user is a user actually using the peer-to-peer content delivery network if the target score meets the scoring requirements.

[0019] In a third aspect, embodiments of this disclosure provide an electronic device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps in the user identification method described above are performed.

[0020] In a fourth aspect, embodiments of this disclosure provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the user identification method described above.

[0021] In a fifth aspect, embodiments of this disclosure provide a computer program product stored in a storage medium, the program product being executed by at least one processor to implement the steps in the user identification method described above.

[0022] As will be described in detail below, a user identification method, system, device, medium, and product according to embodiments of this disclosure are disclosed. In embodiments of this disclosure, log data is first acquired; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data. Interaction data between the user and the peer-to-peer (PCDN) content delivery network service is extracted from the log data. Then, based on the traffic usage data and interaction data, candidate users suspected of using the PCDN are identified. Behavioral characteristics of the candidate users are extracted from the log data, and a target score is calculated based on the behavioral characteristics; wherein the target score indicates the confidence level of the candidate user's actual use of the PCDN. Finally, if the target score meets the scoring requirements, the candidate user is determined to be a user actually using the PCDN. Through the above processing method, this disclosure obtains information related to PCDN user behavior from multiple dimensions through log data, and identifies candidate users through traffic usage data and interaction data, achieving joint determination of traffic and behavioral dimensions, effectively reducing the probability of misidentifying normal users as PCDN users. Furthermore, the system measures the likelihood of a user possessing PCDN user behavior characteristics by calculating a target score. Finally, when the target score meets the preset score requirements, the user is determined to be a user who actually uses PCDN services, thereby ensuring the reliability of the determination process and further improving the accuracy of user identification.

[0023] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0024] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0025] Figure 1 This is a flowchart of a user identification method according to an embodiment of the present disclosure.

[0026] Figure 2 This is a flowchart of a user identification process according to an embodiment of the present disclosure.

[0027] Figure 3 This is a schematic diagram of a user identification system according to an embodiment of the present disclosure.

[0028] Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present disclosure.

[0029] Figure 5 This is a schematic diagram of a computer program product according to an embodiment of the present disclosure. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0031] To facilitate understanding of this embodiment, a user identification method disclosed in this disclosure will first be described in detail, see [link to relevant documentation]. Figure 1 The diagram shown is a flowchart of a user identification method provided in an embodiment of this disclosure. This method can be applied to, but is not limited to, operator platforms. The method includes steps S101-S105, wherein: Step S101: Obtain log data; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data.

[0032] In this embodiment of the disclosure, log data can be obtained from multiple data sources, including at least one of the following: Deep Packet Inspection System (DPI), AAA, Network Flow Analysis System (Netflow), and Border Gateway Protocol (BGP).

[0033] DPI is used to provide application layer data such as user domain name requests to determine whether a user is accessing domains related to PCDN services; AAA is used to provide user uplink and downlink traffic data, i.e., traffic usage data, to determine whether the user's uplink and downlink traffic distribution shows the typical PCDN pattern of uplink being significantly greater than downlink; Netflow is used to provide network layer traffic transmission data, including but not limited to fields such as source Internet Protocol (IP), destination IP, source port, destination port, protocol type, and traffic size, to determine which network addresses the user actually accessed and which transmission protocols were used; BGP is used to provide network path data, recording the sequence of autonomous systems that data packets pass through from the source to the destination, to determine whether user traffic crosses the boundaries of the province or network.

[0034] Step S102: Extract the interaction data between the user and the peer-to-peer content delivery network service from the log data.

[0035] Here, interactive data refers to evidence reflecting communication between the user terminal and the PCDN server. PCDN service is based on P2P technology, where user devices act as both content consumers and content providers, offering services to other users. Therefore, users of the PCDN service engage in specific interactive behaviors with the PCDN server, including but not limited to: initiating domain name resolution requests to the PCDN server, obtaining resource addresses returned by the PCDN server, and initiating data transmission requests to specific server addresses of the PCDN server.

[0036] The interactive data includes at least one of the following: the domain name information requested by the user, the resolution return address corresponding to the domain name requested by the user, the target address actually accessed by the user, the port information used by the user to access the target address, and the traffic transmission records generated by the user accessing the target address.

[0037] Specifically, the domain name request record issued by the user can be extracted from the application layer data. This domain name request record contains the complete domain name requested by the user. By comparing the requested domain name with the domain name characteristics of the PCDN service provider, it can be determined whether the user intends to access the PCDN service.

[0038] Then, the actual target address and port information accessed by the user are extracted from the network layer traffic transmission data. The domain name resolved by the Domain Name System (DNS) corresponds to one or more IP addresses, while the actual target address accessed by the user may match the DNS resolution result or it may differ. By correlating and verifying the actual target address accessed by the user with the address returned by the DNS resolution, it can be confirmed whether the user has truly accessed the service address corresponding to the requested domain name.

[0039] In the above implementation, the user's business access intent and the user's actual communication behavior are reflected from the two dimensions of application layer and network layer, respectively. Thus, while ensuring the integrity of the interaction data, it is possible to effectively confirm whether the user has had a substantial interaction with the PCDN server.

[0040] Step S103: Based on traffic usage data and interaction data, identify potential users who may be using peer-to-peer content delivery networks.

[0041] This disclosure can utilize a pre-defined PCDN traffic analysis model to comprehensively analyze the interaction between the user terminal and the PCDN server, as well as the uplink and downlink traffic values ​​of the user terminal, based on traffic usage data and interaction data. This allows for the selection of users who may be using PCDN services as suspected PCDN users, i.e., candidate users suspected of using peer-to-peer content delivery networks.

[0042] In terms of traffic characteristics, since PCDN service requires users to use uplink bandwidth to provide content distribution to other users, the uplink traffic usage of PCDN users is usually significantly higher than the downlink traffic usage. In terms of interactive behavior, PCDN users need to communicate with the PCDN server, including but not limited to requesting domain names related to PCDN service, obtaining resource addresses, and initiating data transmission to the server address of the PCDN server.

[0043] To address this, a user is only identified as a candidate user when their traffic usage data shows an upward trend exceeding a downward trend, and the interaction data indicates genuine interaction between the user and the PCDN server. This dual-condition joint judgment method can eliminate a large number of normal users with only a single abnormal feature during the candidate user identification stage, improving the accuracy of suspected user identification and providing more reliable candidate samples for subsequent accurate identification.

[0044] Step S104: Extract behavioral features of candidate users from log data and calculate target scores based on the behavioral features; wherein, the target score is used to indicate the confidence level of candidate users in actually using peer-to-peer content delivery networks.

[0045] Here, behavioral characteristics may include, but are not limited to, at least one of the following: traffic scale characteristics reflecting the user's bandwidth usage, port characteristics reflecting the user's access port behavior, protocol characteristics reflecting the user's protocol type, preset interaction characteristics reflecting the user and peer-to-peer content delivery network services, activity characteristics reflecting the user's traffic time distribution pattern, address association characteristics reflecting the consistency between the user's actual access address and the domain name resolution result, and traffic flow characteristics reflecting whether the user's traffic flows across domains.

[0046] Specifically, traffic scale characteristics include at least one of the following: total uplink traffic, total downlink traffic, uplink-to-downlink traffic ratio, etc., which can be obtained from the RADIUS logs of the AAA system.

[0047] Port characteristics include at least one of the following: the distribution of source port numbers used by users to initiate communication, the distribution of target port numbers accessed by users, etc. Port characteristics can be obtained from Netflow logs.

[0048] Protocol characteristics include at least one of the following: the distribution ratio of various transport layer protocols (e.g., Transmission Control Protocol (TCP), User Datagram Protocol (UDP)) and application layer protocols (e.g., P2P protocols) in user traffic.

[0049] The preset interaction features include at least one of the following: the number and frequency of user requests for PCDN-related domain names, the number of resource addresses returned by the PCDN server, etc. The interaction features can be obtained from the DNS logs of the DPI system.

[0050] Activity characteristics include at least one of the following: distribution of user traffic during active periods, duration of continuous online activity, traffic fluctuation patterns, etc. Activity characteristics can be obtained from network layer traffic transmission data and traffic usage data to reflect the distribution pattern of user traffic in the time dimension.

[0051] Address association features include at least one of the following: the DNS resolution return address corresponding to the domain name requested by the user, the target address actually accessed by the user, and the degree of matching between the DNS resolution return address and the actual address. Address association features can be obtained from DNS logs and Netflow logs.

[0052] Traffic flow characteristics include at least one of the following: the autonomous system to which the destination IP of the user traffic belongs, the geographical distribution of the destination IP, and the proportion of cross-domain traffic to the total traffic. Traffic flow characteristics can be obtained from BGP routing information.

[0053] In this embodiment of the disclosure, each behavioral feature can be compared with normal user behavior to obtain an initial score indicating the degree to which the behavioral feature deviates from the normal range. The target score is then calculated by combining the initial scores corresponding to each behavioral feature, thereby integrating the individual information of each behavioral feature and the correlation between features.

[0054] The target score directly reflects the confidence level that the candidate user belongs to PCDN. The higher the target score, the greater the match between the candidate user's behavior pattern and that of PCDN users, and the greater the likelihood that the user actually uses PCDN services; the lower the target score, the closer the candidate user's behavior pattern is to that of normal users, and the less likely the user is to actually use PCDN services.

[0055] In the above embodiments, due to the various differences in network behavior between PCDN users and ordinary users, this disclosure assesses whether there is an abnormal increase in uplink traffic through traffic scale characteristics at the traffic dimension, whether the accessed ports are concentrated in the port range commonly used by PCDN services at the transport layer dimension through port characteristics, whether the protocols used are mainly P2P-type protocols at the transport layer dimension through protocol characteristics, whether there are frequent requests for domain names related to PCDN services at the application layer dimension through interaction characteristics, whether there is continuous and fluctuating activity throughout the day through activity characteristics at the time dimension, whether the actual accessed address is inconsistent with the DNS resolution result through address association characteristics at the address association dimension, and whether there is a large amount of traffic flowing to cross-domain destination addresses at the routing dimension through traffic flow characteristics. This comprehensively characterizes the behavioral patterns of PCDN users from multiple dimensions, helping to ensure the accuracy of assessing the actual use of peer-to-peer content delivery networks by candidate users.

[0056] Step S105: If the target score meets the score requirements, determine the candidate users as users who actually use the peer-to-peer content delivery network.

[0057] Here, the scoring requirement can be set as a scoring threshold or a dynamically adjusted judgment condition. The scoring threshold can be set to 0.85 when 1 is the full score, and to 85 when 100 is the full score, or it can be adjusted adaptively as needed.

[0058] For example, in scenarios requiring high accuracy, a higher score threshold can be set to ensure that candidate users identified as PCDN users have a very high level of confidence, thereby minimizing the risk of false positives. In scenarios requiring high recall, a lower score threshold can be set to ensure that as many PCDN users as possible are identified.

[0059] When a candidate user's target score exceeds the score requirement, it indicates that the confidence level of the user belonging to PCDN has reached the determination level, and therefore the user is identified as a user who actually uses PCDN services; when a candidate user's target score does not exceed the score requirement, it indicates that the confidence level of the user belonging to PCDN is insufficient, and therefore the user is not identified as a PCDN user.

[0060] Through the aforementioned determination mechanism, this disclosure ensures that a positive determination is made only when the confidence level reflected by the target score reaches a sufficiently high level, thereby avoiding misjudgment when the evidence is insufficient and further improving the reliability of identification.

[0061] In the above embodiments, this disclosure obtains information related to PCDN user behavior from multiple dimensions through log data, and identifies candidate users through traffic usage data and interaction data, achieving joint determination of traffic and behavior dimensions, effectively reducing the probability of misclassifying normal users as PCDN users. Furthermore, it measures the likelihood of a user possessing PCDN user behavior characteristics through a calculated target score. Finally, when the target score meets a preset score requirement, the user is determined to be a user actually using PCDN services, thereby ensuring the reliability of the determination process and further improving the accuracy of user identification.

[0062] In an alternative implementation, after determining that the candidate users are actually using the peer-to-peer content delivery network, this disclosure may also output a PCDN user report.

[0063] The report structure of the PCDN user report may include at least one of the following: user IP, uplink traffic, downlink traffic, domain name matching rate, path score, protocol spoofing degree, target score, and judgment result.

[0064] For example, 192.168.1.100, 1024, 210, 92.3, 0.88, 0.45, 0.91, PCDN users; 10.20.30.40, 450, 380, 15.2, 0.31, 0.12, 0.42, non-PCDN users; 172.16.2.88, 890,150, 87.6, 0.79, 0.67, 0.89, PCDN users.

[0065] This disclosure also enhances the visualization of PCDN user reports using heatmaps and decision dashboards. The heatmap displays the distribution of feature scores (e.g., horizontal axis: 8 features, vertical axis: user IP range). The decision dashboard displays a real-time radar chart of the top 10 high-risk users and their key features. Further details are omitted here.

[0066] In an optional implementation, the above steps for calculating the target score based on behavioral features include: determining an initial score for each behavioral feature; wherein the initial score indicates the degree to which the behavioral feature deviates from a normal user behavior baseline; adjusting the initial scores based on log data and behavioral features to obtain adjusted scores for each behavioral feature; and performing a weighted summation of the adjusted scores corresponding to each behavioral feature to obtain the target score.

[0067] Here, the benchmark for normal user behavior is obtained through statistical analysis of historical behavioral data from a large number of ordinary users in the network. Specifically, behavioral characteristic data of ordinary users who are not identified as PCDN users during normal network use can be collected, and statistical distribution parameters of various behavioral characteristics in the normal user group can be calculated, including but not limited to at least one of the following: mean, median, and standard deviation.

[0068] After obtaining the baseline of normal user behavior, this disclosure compares the value of each behavioral feature of a candidate user with the corresponding baseline. The greater the deviation from the baseline, the more abnormal the behavioral feature, and the higher the initial score.

[0069] The initial score can be calculated using the following formula: in, : No. Item features (e.g., flow size features), Features In the Normalized value (range 0-1) within each detection period. Time decay weight ( (Recent data has a higher weight).

[0070] For example, if traffic activity characteristics The past 3 period values ​​are [0.7, 0.9, 0.8]. .

[0071] At the network level, a user's abnormal traffic behavior is usually not an isolated event, but rather related to the abnormal behavior of other users in the network. Furthermore, at the feature level, there may be causal relationships between various behavioral features. Therefore, this disclosure adjusts the initial score based on log data and behavioral features.

[0072] Through the above adjustments, the initial scores of each behavioral feature were revised to adjusted scores. The adjusted scores not only reflect the anomalousness of the feature itself, but also incorporate the feature's involvement in the propagation of network anomalies, as well as the causal relationships between features, making the score of each feature more comprehensively and accurately reflect its contribution to PCDN user identification.

[0073] Furthermore, after obtaining the adjusted scores of each behavioral feature, this disclosure performs a weighted summation of the adjusted scores corresponding to each behavioral feature to obtain the target score, thereby comprehensively and accurately reflecting the confidence level that the candidate user belongs to the PCDN user.

[0074] In the above embodiments, through the above scoring mechanism, this disclosure can integrate the various behavioral characteristics of candidate users into a unified target score. This target score comprehensively reflects the candidate user's degree of abnormality in multiple dimensions, the degree of participation in abnormal network propagation, the causal consistency between features, and the comprehensive contribution of each feature, thereby improving the accuracy and reliability of user identification.

[0075] In an optional implementation, the above steps adjust the initial scores based on log data and behavioral characteristics to obtain adjusted scores for each behavioral characteristic, including: constructing a directed weighted graph based on log data; wherein the nodes of the directed weighted graph are network node identifiers, the directed edges of the directed weighted graph are used to indicate the traffic transmission relationship between nodes, and the edge weights of the directed weighted graph are used to indicate the proportion of traffic transmission between nodes; the network node identifiers include at least one of the following: user network address, autonomous system number. Constructing a causal relationship matrix based on behavioral characteristics; wherein the rows and columns of the causal relationship matrix correspond to behavioral characteristics, and the element values ​​in the causal relationship matrix are used to indicate whether there is a causal relationship between the behavioral characteristic corresponding to the row and the behavioral characteristic corresponding to the column. Calculating the path score of the abnormal propagation path based on the directed weighted graph, and determining the feature influence strength between behavioral characteristics based on the causal relationship matrix; wherein the feature influence strength is represented by the element values ​​in the causal relationship matrix. Adjusting the initial scores based on the path scores and feature influence strength to obtain adjusted scores for each behavioral characteristic.

[0076] This disclosure enables the construction of directed weighted graphs based on log data (e.g., Netflow log data), which is a graph data structure used to represent the traffic transmission relationships between nodes in a network.

[0077] Specifically, the directed weighted graph is, In this context, node V represents the user's network address or Autonomous System (AS) number (the source / destination IP and BGP AS path can be extracted from Netflow logs). Directed edges E indicate the traffic transmission relationship between nodes: if node A transmits traffic to node B, then an edge exists... The edge weight W is used to indicate the proportion of traffic transmission between nodes. .

[0078] In practical applications, the node type can be selected according to the required granularity of analysis. When fine-grained analysis of traffic behavior at the individual user level is required, the user's network address can be used as the node; when analyzing traffic flow at the region level is required, the autonomous system number can be used as the node.

[0079] Furthermore, the causal relationship matrix is ​​a data structure used to represent whether there is a causal relationship between various behavioral characteristics.

[0080] The element values ​​in the causal relationship matrix are used to indicate whether there is a causal relationship between the behavioral feature corresponding to the row and the behavioral feature corresponding to the column. For example, if the element value in the i-th row and j-th column of the matrix is ​​1 (or a non-zero value), it means that the j-th behavioral feature is the cause of the i-th behavioral feature, that is, feature j has a causal influence on feature i; if the element value is 0, it means that there is no causal relationship between the two.

[0081] After constructing the directed weighted graph, the path score of the anomaly propagation path is calculated based on the directed weighted graph. The path score reflects the strength of the anomaly's spread along a specific propagation path in the network. The higher the path score, the more severe the anomaly propagation in the network, and the higher the correlation between the nodes located on that path and the anomaly propagation.

[0082] After calculating the path score of the abnormal propagation path and constructing the causal relationship matrix, the initial score is adjusted based on the path score and the feature influence strength to obtain the adjusted score of each behavioral feature.

[0083] The strength of feature influence is represented by the element values ​​in the causal relationship matrix. When an element in the causal relationship matrix is ​​non-zero, that element value is directly used as the strength of feature influence between corresponding behavioral features. For example, if the value of the element in the i-th row and j-th column of the causal relationship matrix is ​​Cij, then Cij represents the strength of feature influence of feature j on feature i.

[0084] For each behavioral feature, a path score is determined for the anomaly propagation path associated with that feature. Since anomaly propagation paths consist of nodes, and nodes correspond to network node identifiers such as user IP addresses, when a candidate user is located on one or more anomaly propagation paths, a higher path score indicates a higher degree of involvement of the candidate user's anomalous behavior in the network anomaly propagation; therefore, the initial score of its related behavioral features should be increased accordingly.

[0085] When other behavioral characteristics have a causal influence on this behavioral characteristic, an abnormality in the other behavioral characteristics will lead to an abnormality in this behavioral characteristic. Therefore, the initial score of the other behavioral characteristics should be multiplied by the influence coefficient and then added to the initial score of this behavioral characteristic.

[0086] In this embodiment, path score adjustment focuses on the structural position of candidate users within the network traffic transmission topology. By adjusting the path score, candidate users in anomaly propagation chains can be effectively identified. Feature influence strength adjustment focuses on the causal logical consistency between various behavioral features. By adjusting the feature influence strength, the score can be made consistent with the causal logic between features. Thus, by adjusting the initial score through path score and feature influence strength, the adjusted score reflects the candidate user's network topology position and the causal logical relationship between features, thereby improving the reliability of the scoring.

[0087] In an optional implementation, the above steps, based on a directed weighted graph, calculate the path score of the anomaly propagation path, including: identifying nodes with abnormal flow behavior in the directed weighted graph as anomalous nodes; tracing the anomaly propagation path from the anomalous nodes along the edges of the directed weighted graph using a flood propagation algorithm; determining the anomalousness of nodes in the anomaly propagation path, the weights between adjacent nodes, and the path depth; and calculating the path score based on the anomalousness, the weights between adjacent nodes, and the path depth.

[0088] Here, nodes exhibiting abnormal traffic behavior refer to nodes whose traffic transmission behavior does not conform to normal patterns. Specifically, when the uplink traffic usage of a user associated with a node is significantly greater than its downlink traffic usage (e.g., ... This means that the uplink traffic exceeds a preset multiple threshold for downlink traffic, and the user's traffic activity exceeds a preset activity threshold (i.e., When this happens, the node can be identified as having abnormal traffic behavior.

[0089] In practical applications, the rules for identifying abnormal nodes can be set according to the specific network environment and identification needs. For example, they can be determined solely based on the ratio of uplink traffic to downlink traffic, or they can be combined with other factors such as traffic activity and protocol type for a comprehensive determination. This disclosure will not elaborate further here.

[0090] After identifying the anomalous node, the flood propagation algorithm is used to trace the anomalous propagation path from the anomalous node along the edge directions of the directed weighted graph.

[0091] Among them, the flood propagation algorithm is an algorithm that simulates the propagation of information or influence along the edges of a graph structure. This disclosure uses the flood propagation algorithm to spread the anomalous influence outward hop by hop from the anomalous node along the edge direction of the directed weighted graph, thereby identifying all node sequences affected by the anomalous node.

[0092] Specifically, for each node marked as an anomalous node, the system traverses the directed weighted graph along the edges pointing from that node to its downstream nodes. During the traversal, the system propagates along edges whose weights exceed a preset weight threshold (e.g., 0.5).

[0093] If the edge weight exceeds a preset weight threshold, it indicates that the node will concentrate most of its outgoing traffic to downstream nodes, suggesting that the abnormal impact may propagate along this direction. For each edge that meets the propagation condition, the nodes it passes through are added to the path sequence in turn until it is impossible to continue traversing. The resulting node sequence is an abnormal propagation path.

[0094] After tracing the anomaly propagation path, the degree of anomaly of the nodes in the propagation path, the weights between adjacent nodes, and the path depth are determined. The degree of anomaly of a node reflects the severity of the anomaly at each node in the path, the edge weights reflect the concentration of anomaly propagation, and the path depth reflects the propagation range.

[0095] Among them, the anomaly degree of a node is used to quantify the severity of the abnormal traffic behavior of that node, the weight between adjacent nodes is the weight value of the corresponding directed edge in the directed weighted graph, which represents the proportion of traffic transmitted from the upstream node to the downstream node to the total outflow of the upstream node, and the path depth refers to the number of nodes contained in the abnormal propagation path.

[0096] For example, the path depth of path A→B→C is 3, and the path depth of path A→B→C→D is 4. Path depth reflects the number of hops that an anomaly's impact takes to propagate along the network from its source. The greater the path depth, the wider the scope of the anomaly's impact and the longer the propagation chain.

[0097] After determining the degree of anomaly of nodes in the anomaly propagation path, the weights between adjacent nodes, and the path depth, the path score is calculated based on the degree of anomaly, the weights between adjacent nodes, and the path depth.

[0098] Path score The calculation formula is: ,in, The path The abnormal strength of each node. : No. To the Edge weights of nodes (from a directed graph). Attenuation factor (default value 0.2), controls the attenuation intensity of anomalies as they travel further. : Current path length (number of nodes).

[0099] For each path, the score is accumulated according to the formula (for example, the score for path A→B→D is...). ).

[0100] In practical applications, the formula for calculating path score can be adjusted and modified according to specific application scenarios and accuracy requirements. For example, different aggregation functions or different decay functions can be used, which will not be elaborated further in this disclosure.

[0101] In the above implementation, by constructing a directed weighted graph and identifying abnormal nodes, the abnormal source point can be located from a global perspective of the network topology. It can also accurately identify the propagation direction and propagation chain of the abnormal impact in the network, comprehensively reflect the propagation characteristics of PCDN traffic, and by introducing a mechanism that decays with increasing depth in the path score calculation, the path score can accurately identify PCDN users.

[0102] In an optional implementation, the above steps for constructing a causal relationship matrix based on behavioral features include: extracting time-series data of any two behavioral features within a time window; performing a Granger causality test on the time-series data to obtain the test result; wherein the test result indicates whether historical data of one behavioral feature has a predictive effect on the current data of another behavioral feature; determining whether a causal relationship exists between the behavioral features based on the test result, and constructing a causal relationship matrix based on the obtained determination result.

[0103] As mentioned above, this disclosure provides an exemplary description of the process of constructing a causal relationship matrix.

[0104] To determine whether a causal relationship exists between various behavioral features, this disclosure first extracts time-series data of any two behavioral features within a time window. The time-series data indicates the value sequence of the same behavioral feature at multiple consecutive time points, and the length of the time window can be set according to the actual application scenario.

[0105] After extracting time-series data of any two behavioral features, Granger causality tests are performed on the time-series data to obtain the test results.

[0106] Among them, the Granger causality test is a statistical hypothesis testing method used to determine whether one time series has a predictive effect on another time series.

[0107] For any feature pair Establish a vector autoregressive model: Where Y is the feature being predicted, X is the feature being predicted, and p is the lag order. Let Y be the historical value of the previous i periods. Let X be the historical value of the previous i periods. The regression coefficient of the lagged term of Y, The regression coefficient of the lagged term X, This represents the model residual noise.

[0108] Through the Granger causality test described above, a test result is obtained for each pair of behavioral features. This test result indicates whether the historical data of one behavioral feature has a predictive effect on the current data of the other behavioral feature.

[0109] After obtaining the results of the Granger causality test, we determine whether there is a causal relationship between the behavioral characteristics based on the test results, and construct a causal relationship matrix based on the obtained judgment results.

[0110] Specifically, for each pair of behavioral features, the F-statistic and corresponding p-value of the Granger causality test are used to make a judgment: if the p-value is less than the preset significance level (e.g., 0.05), the null hypothesis is rejected and feature j is determined to have a causal relationship with feature i; if the p-value is greater than or equal to the preset significance level, feature j is determined to have no causal relationship with feature i.

[0111] After determining the causal relationship of all feature pairs, the results are summarized to construct a causal relationship matrix. The value of the element M(i,j) in the i-th row and j-th column of the matrix follows these rules: When j ≠ i, if the Granger causality test determines that feature j is a cause of feature i (i.e., feature j has a causal relationship with feature i), then M(i,j) takes a non-zero value (e.g., 1); if no causal relationship is determined, then M(i,j) takes a zero value. For the case i = j, M(i,i) can be set to 0 or a null value, indicating that the feature has no causal relationship with itself.

[0112] Example of causal relationship:

[0113] Example of a causal relationship matrix:

[0114] In the above implementation, by constructing a causal relationship matrix through Granger causality verification, the causal relationship direction and causal chain between various features can be identified from the time series data of candidate users, thereby greatly improving the logical consistency between the score and the PCDN behavior pattern and helping to improve the accuracy of PCDN user identification.

[0115] In an optional implementation, the above steps adjust the initial scores based on path scores and feature influence strength to obtain adjusted scores for each behavioral feature, including: for any behavioral feature, determining the mean of the path scores associated with that behavioral feature; determining the feature influence strength of other behavioral features on that behavioral feature based on a causal relationship matrix, and calculating the product between the feature influence strength and the initial scores of other behavioral features; performing a weighted summation on the mean and the product to obtain a processing result; and summing the initial score of that behavioral feature and the processing result to obtain the adjusted score of that behavioral feature.

[0116] This disclosure provides a method for identifying a candidate user’s behavioral characteristic, which may be associated with one or more abnormal propagation paths. After obtaining the scores of all paths associated with the behavioral characteristic, the average of the path scores is calculated to obtain the mean of the path scores associated with the behavioral characteristic.

[0117] The mean path score reflects the average level of participation of this behavioral characteristic in the anomaly propagation path. The higher the mean, the greater the participation of the candidate user in the anomaly propagation chain, and the more likely they are to be a PCDN user.

[0118] As mentioned earlier, the rows and columns of the causal relationship matrix correspond to various behavioral features, and the element values ​​in the matrix indicate whether a causal relationship exists between the row features and the column features. For the currently processed behavioral feature i, the influence coefficients Cij of all other behavioral features j on this feature are obtained from the causal relationship matrix, which is the feature influence strength.

[0119] Then, the product of the influence strength of this feature and the initial scores of other behavioral features is calculated, and the mean of the path scores and the product are weighted and summed to obtain the processed result. The weight coefficients can be adaptively adjusted as needed. Finally, the initial score of this behavioral feature and the processed result are summed to obtain the adjusted score of this behavioral feature.

[0120] Among them, the adjusted score The formula is: ,in, : The initial score of feature i. : The mean of the path scores associated with feature i. The influence of feature j on i in the causal relationship matrix. : Regulation factors (e.g., ).

[0121] For example, if the uplink traffic volume has a BaseScore of 80, a mean PathScore of 1.29, and is causally affected by the destination port characteristics (BaseScore = 70)... ),but: .

[0122] In the above implementation, the adjusted score not only includes the abnormality information of the behavioral feature itself, but also integrates the structural participation information of the behavioral feature in the abnormal propagation of the network, as well as the causal information between the behavioral feature and other behavioral features, so as to more comprehensively and accurately reflect the contribution of the behavioral feature in PCDN user identification based on the adjusted score.

[0123] In an optional implementation, the above steps involve weighted summation of the adjusted scores corresponding to each behavioral feature to obtain the target score, including: dynamically determining weight coefficients based on historical data corresponding to each behavioral feature; and weighted summation of the adjusted scores and corresponding weight coefficients for each behavioral feature to obtain the target score.

[0124] Here, historical data may include, but is not limited to, at least one of the following: the historical false positive rate of behavioral features, and the proportion of the initial score of a behavioral feature in the path score. The false positive rate refers to the frequency with which a certain behavioral feature is incorrectly used to identify PCDN users in historical identification. If the false positive rate of a certain feature is high, it indicates that the feature is less reliable in distinguishing between PCDN users and ordinary users, and its importance in the current identification should be reduced accordingly.

[0125] In one optional implementation, if the historical false alarm rate exceeds a false alarm threshold, the weight coefficient of the behavioral feature is reduced (e.g., reduced by 10%). If the proportion exceeds a proportion threshold (e.g., 0.3), the weight coefficient of the behavioral feature is increased (e.g., increased by 15%).

[0126] After determining the weight coefficients of each behavioral feature, this disclosure calculates the target score by weighting and summing the adjusted scores and corresponding weight coefficients for each behavioral feature.

[0127] In this embodiment of the disclosure, if the misjudgment rate of a certain behavioral feature exceeds a preset misjudgment threshold (e.g., 30%), it indicates that the feature has low reliability in historical identification, and its weight coefficient is reduced by a certain percentage (e.g., reduced by 10%). If the misjudgment rate of a certain behavioral feature does not exceed the preset misjudgment threshold, it indicates that the reliability of the feature is within an acceptable range, and its weight coefficient can remain unchanged or be finely adjusted according to the level of the misjudgment rate.

[0128] In the above implementation, the weight coefficients can be dynamically optimized based on the actual recognition performance of each feature. This dynamic optimization mechanism can learn from historical recognition results and continuously improve, adaptively adjusting the importance allocation of features without manual intervention. After completing the dynamic adjustment, the weight coefficients of all features are normalized to ensure the stability of the target score's numerical range. Moreover, through this dynamic weight fusion method, the final target score includes both the individual information of each feature and fully considers their relative importance in different scenarios, providing a more accurate and reliable quantitative basis for PCDN user recognition.

[0129] In an optional implementation, the above steps, based on traffic usage data and interaction data, identify potential candidate users of the peer-to-peer content delivery network, including: if, based on traffic usage data, it is determined that a user's uplink traffic usage is greater than their downlink traffic usage, assigning a first label to the user; if, based on interaction data, it is determined that there is interaction between the user and the peer-to-peer content delivery network service, assigning a second label to the user; and determining users with both the first and second labels as candidate users.

[0130] Here, traffic usage data includes at least the user's uplink and downlink traffic usage within a specific time period.

[0131] For ordinary users, since daily internet activity mainly involves receiving data, their downlink traffic is usually much greater than their uplink traffic. However, for PCDN users, because they utilize uplink bandwidth to provide content distribution services to other users, their uplink traffic usage increases significantly, usually exceeding their downlink traffic usage.

[0132] In this regard, this disclosure uses the condition that uplink traffic usage is greater than downlink traffic usage as the primary criterion for determining whether a user is a suspected PCDN user. When a user's uplink traffic usage is greater than their downlink traffic usage, it indicates that the user's traffic distribution pattern differs from that of ordinary users and exhibits the traffic characteristics of PCDN services; therefore, this user is initially flagged. If a user's uplink traffic usage is less than or equal to their downlink traffic usage, it indicates that the user's traffic distribution pattern is closer to that of ordinary users and does not possess the typical traffic characteristics of PCDN services; therefore, the possibility of them being a PCDN user can be preliminarily ruled out.

[0133] In addition, a ratio method can also be used to determine whether the ratio between uplink traffic usage and downlink traffic usage exceeds a preset ratio threshold, which will not be elaborated further here.

[0134] Based on this, if it is determined from the interaction data that there is an interaction between the user and the peer-to-peer content delivery network service, then the user is marked a second time.

[0135] The interactive data may include, but is not limited to, communication information between the user and the PCDN server. This primarily includes data from DNS log files and Netflow log files. Data in the DNS log files includes, but is not limited to, the user's requested domain name and the domain name resolution response address. Data in the Netflow log files includes, but is not limited to, the destination network address, the destination port number, and the traffic direction identifier.

[0136] When using PCDN services, PCDN users need to interact with the PCDN server, such as requesting content resources and obtaining resource addresses. These interactions are recorded in the log data. For example, the DNS requests issued by the user may contain domain names related to the PCDN service, or the target address actually accessed by the user may point to the PCDN server.

[0137] When the interaction data indicates that there are one or more interaction behaviors between the user and the PCDN server, it means that the user has evidence of using the PCDN service, and therefore the user is marked a second time.

[0138] After completing the first and second marking, users with the first and second markings are identified as candidate users.

[0139] In the above implementation, the dual-label joint determination method ensures that candidate users simultaneously meet the two conditions of abnormal traffic characteristics and interactive behavior, thereby improving the accuracy of the user identification process.

[0140] In an optional implementation, if the above steps determine that there is an interaction between the user and the peer-to-peer content delivery network service based on the interaction data, and then perform a second tagging on the user, the steps include: extracting the domain name information requested by the user from the interaction data. If the domain name information matches successfully with a preset peer-to-peer content delivery network service domain name database, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is then second-tagged.

[0141] Here, domain information may include, but is not limited to, user identifier, request time, and requested domain.

[0142] This disclosure allows you to obtain the domain name information requested by the user by parsing DNS logs and extracting the query domain name field from each DNS request record. After extracting the domain name information, it determines whether the domain name information matches a pre-defined peer-to-peer content delivery network service domain name database.

[0143] The preset PCDN service domain name library is a pre-built collection of domain names, which may include, but is not limited to, domain names or domain name patterns used by known PCDN service providers.

[0144] During domain name matching, the extracted user-requested domain name is compared with entries in the domain name database. If the user-requested domain name matches any entry in the domain name database, the domain name information is considered to have successfully matched the preset PCDN service domain name database.

[0145] After confirming that the domain name information requested by the user matches the preset PCDN service domain name database, it is determined that there is an interaction between the user and the PCDN service, and the user is marked a second time.

[0146] In an optional implementation, if the above steps determine, based on interaction data, that there is interaction between the user and the peer-to-peer content delivery network service, and then perform a second tagging on the user, this includes: extracting the DNS resolution return address corresponding to the user's requested domain name and the target address actually accessed by the user from the interaction data. If the DNS resolution return address and the target address match successfully, it is determined that there is interaction between the user and the peer-to-peer content delivery network service, and the user is then second-tagged.

[0147] After acquiring the interaction data, this disclosure extracts the DNS resolution return address corresponding to the user's requested domain name and the target address actually accessed by the user from the interaction data. The DNS resolution return address refers to the network address corresponding to the domain name returned to the user by the Domain Name System server after receiving the user's domain name resolution request. The target address actually accessed by the user refers to the destination network address of the data packets actually sent by the user when establishing a communication connection with the network server.

[0148] After extracting the DNS resolution return address and the actual accessed target address, the DNS resolution return address of the same user within the same time window is compared with the actual accessed target address. If at least one of the actual accessed target addresses is the same as at least one of the DNS resolution return addresses, the DNS resolution return address and the target address are considered to be matched successfully; if there is no intersection between the two, the match is considered to be unsuccessful.

[0149] If the first match fails, the time alignment range is expanded to ±15 minutes to check for the existence of IP alias mapping (e.g., the final IP in the resolution chain). If the match fails again, it is marked as an association anomaly, triggering dynamic adjustment of feature weights, reducing the initial weight of the address association feature in the causal relationship matrix (e.g., from 0.3 to 0.1).

[0150] Example scenario: A user requests the domain video.pcdn.com → DNS response IP 1.2.3.4, but there is no record of accessing 1.2.3.4 in Netflow. Upon inspection, it is found that the actual accessing IP is 5.6.7.8 (edge ​​node), which is then determined to be PCDN traffic spoofing, and the protocol spoofing feature score is increased. The higher the protocol spoofing score, the more normal and legitimate the user's internet traffic characteristics resemble, and the more difficult it is to be identified by traditional deep packet inspection devices.

[0151] The successful match between the resolved address and the target address indicates that the user has obtained the IP address of the PCDN server through domain name resolution and has actually established a communication connection with that address. After confirming a successful match between the resolved address and the target address, this disclosure determines that there is interaction between the user and the PCDN service and then marks the user with a second tag.

[0152] In this embodiment of the disclosure, both address matching and the aforementioned domain name matching are implementations based on interactive data for second tagging. Domain name matching, at the application layer level, is used to check whether the user has requested the resolution of PCDN-related domain names, verifying the user's access intent; address matching, at the network layer level, checks whether the user has requested the resolution of PCDN-related domain names, and also verifies whether the user has actually accessed the IP address corresponding to the domain name, verifying the user's actual access behavior.

[0153] In the above embodiments, domain name matching can cover scenarios accessed via domain name resolution, and address matching can cover scenarios accessed via direct network connection, ensuring that at least one matching method can capture user interaction behavior. Furthermore, domain name matching provides highly sensitive intent-level verification, while address matching provides highly accurate behavior-level verification, thereby improving the comprehensiveness and reliability of interaction behavior recognition.

[0154] In an optional implementation, after acquiring the log data, the method further includes: storing application layer data from the deep packet inspection system as application layer logs; storing traffic usage data from the authentication, authorization, and accounting system as traffic metering logs; and storing network layer traffic transmission data from the network flow analysis system and network path data from the border gateway protocol system as traffic transmission logs.

[0155] In this embodiment of the disclosure, the application layer data obtained from the DPI system is stored as application layer logs. The application layer data includes, but is not limited to: Domain Name System request records issued by users, host header information in Hypertext Transfer Protocol request records, and server name indication information in the Secure Transport Layer Protocol handshake process.

[0156] Application layer logs store key fields of each application layer record in a structured format, including but not limited to: timestamp, user IP address, requested domain name, and request type.

[0157] Traffic usage data includes, but is not limited to, total uplink traffic, total downlink traffic, and session duration information for a user session. This traffic usage data is stored as a traffic metering log, which saves key fields for each traffic metering record in a structured format, including, but not limited to: user IP address, session identifier, uplink traffic volume, downlink traffic volume, session start time, and session end time.

[0158] Network layer traffic transmission data includes, but is not limited to: source IP address, destination IP address, source port, destination port, transport layer protocol type, traffic size, and traffic duration.

[0159] Network path data includes, but is not limited to: the Autonomous System number to which the IP address belongs, the connection relationship between Autonomous Systems, and the sequence of AS paths through which data packets are transmitted.

[0160] Network layer traffic transmission data and network path data are jointly stored as a traffic transmission log. The traffic transmission log saves the key fields of each traffic transmission record in a structured format, including but not limited to: timestamp, source IP address, destination IP address, source port, destination port, protocol type, traffic size, autonomous system number of the destination IP, and whether the traffic crosses domains.

[0161] Overall, this disclosure stores DNS log data obtained from the DPI system as DNS log files; stores Netflow metropolitan area and Internet Data Center (IDC) identifiers obtained from parsing Netflow packets, AS address allocations obtained from BGP routes, and IDC metropolitan area addresses as Netflow log files; Netflow packets are obtained through the Netflow system; and stores Remote Authentication Dial-In User Service (RADIUS) traffic-related data obtained from the AAA system as RADIUS log files.

[0162] This disclosure allows for parsing DPI logs in the DPI system to obtain DNS log data, and parsing AAA logs in the AAA system to obtain RADIUS traffic-related data.

[0163] The following is an example of the DPI log parsing process: #Example of raw DPI log format# TIMESTAMP|SRC_IP|DST_IP|PROTO|DOMAIN def parse_dpi_log(raw_line): fields = raw_line.split("|") # Pipe character delimiter parsing if fields[3] == "DNS": # Only extract DNS protocol logs return { "timestamp": fields[0], "src_ip": fields[1], "domain": fields[4] # Directly extract the requested domain name } # Output stored as a DNS log file (JSON format) This disclosure employs a streaming processing framework to parse TB-level DPI logs in real time and filters non-DNS traffic using regular expressions.

[0164] The following is an example of the AAA log parsing process: # AAA Log Raw Format (Binary RADIUS Message) def parse_aaa_log(binary_data): packet = radius.Packet(packet=binary_data) return{ "user_ip": packet["Framed-IP-Address"], "bytes_up": packet["Acct-Input-Octets"] #Upstream traffic "bytes_down":packet["Acct-Output-Octets"] #Downlink traffic} This disclosure implements key field mapping by calling the Pyrad library to decode RADIUS protocol messages. For example, Acct-Input-Octets → uplink traffic, Acct-Output-Octets → downlink traffic.

[0165] This categorized storage keeps data organized, with different data types in their proper places, facilitating quick and accurate retrieval and access, and improving data processing efficiency. For example, when identifying a suspected PCDN user, domain name request information can be quickly retrieved from the DNS log file for comparison.

[0166] This disclosure enables rapid access to DNS logs via the following process: Log Location: Based on the target user's IP address, locate the corresponding DNS log file (file index format: dns_<user IP range>.log) in the distributed storage system. Hash Query: Use a Bloom filter to quickly filter out the set of domain names requested by the user in the past 24 hours. Domain Name Matching: Perform parallel regular expression matching between the filtered domain names and a preset PCDN domain name database (stored in an in-memory Redis database).

[0167] Example: User IP 192.168.1.100 requests a domain name. If the regular expression rule matches successfully, the first interactive behavior flag is triggered.

[0168] In the above implementation, the method of partitioning log data can fully leverage the advantages of each data source and comprehensively preserve the value of the data. For example, Netflow log files integrate multi-source information, providing in-depth information support for PCDN user identification from multiple dimensions such as network topology and traffic distribution. This helps to more accurately determine user behavior characteristics and avoid identification biases caused by mixed or missing data.

[0169] Specifically, at the network topology level: Based on the AS path data generated from BGP routing, the Autonomous System (AS number) to which the destination IP of the traffic belongs is marked in the Netflow log. Application scenario: When user traffic is detected to be continuously flowing to non-carrier-cooperative AS (for example, AS413DST4 is a commonly used AS by PCDN service providers), its abnormal weight is automatically increased.

[0170] Traffic distribution dimension: Parse the src_ip, dst_ip, and bytes fields in the Netflow packet, and combine them with the stored metropolitan area data center (IDC) identifier. Application scenario: User 10.20.30.40 continuously uploads traffic to the IDC address range 58.32.0.0 / 16 (e.g., uplink / downlink ratio > 5:1), triggering the first flag.

[0171] The following is combined Figure 2 The user identification process described above is as follows: S201: Retrieve log data.

[0172] The log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data.

[0173] S202: Extract interaction data between users and peer-to-peer content delivery network services from log data.

[0174] S203: If, based on traffic usage data, it is determined that a user's uplink traffic usage is greater than their downlink traffic usage, then the user is marked as a first user.

[0175] S204: If, based on interaction data, it is determined that there is an interaction between a user and a peer-to-peer content delivery network service, the user is marked a second time.

[0176] S205: Identify users with the first and second tags as potential candidates for using peer-to-peer content delivery networks.

[0177] S206: Extract behavioral features of candidate users from log data and determine the initial score for each behavioral feature.

[0178] The initial score indicates the degree to which behavioral characteristics deviate from the baseline of normal user behavior.

[0179] S207: Based on log data and behavioral characteristics, adjust the initial score to obtain the adjusted score for each behavioral characteristic.

[0180] S208: The adjusted scores corresponding to each behavioral feature are weighted and summed to obtain the target score.

[0181] The target score is used to indicate the confidence level of a candidate user in actually using a peer-to-peer content delivery network.

[0182] S209: If the target score meets the score requirements, the candidate users are identified as users who actually use the peer-to-peer content delivery network.

[0183] In this embodiment, data is collected from multiple data sources, resulting in more comprehensive and richer data that can characterize user behavior from multiple dimensions. Furthermore, by jointly analyzing the interaction between the user terminal and the PCDN server, as well as the uplink and downlink traffic of the user terminal, suspected PCDN users are identified. This joint analysis avoids the limitations of single analysis and improves the accuracy of suspected user identification. Finally, a multi-source correlation causal fusion method is used to determine the comprehensive score of various characteristics of suspected PCDN users. This method, through comprehensive judgment across multiple dimensions, makes the comprehensive score more scientific and reasonable, thereby accurately identifying PCDN users and reducing the possibility of misjudgment and missed judgment.

[0184] Based on the same inventive concept, this disclosure also provides a user identification system corresponding to the user identification method. Since the principle of the system in this disclosure for solving the problem is similar to the user identification method described above in this disclosure, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0185] Reference Figure 3The diagram shown is a schematic representation of a user identification system provided in an embodiment of this disclosure. The system includes: a log acquisition module 30, a data extraction module 31, a user identification module 32, a calculation module 33, and a user determination module 34; wherein: The log acquisition module is used to acquire log data; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data; The data extraction module is used to extract interaction data between users and peer-to-peer content delivery network services from log data; The user identification module is used to identify potential users of peer-to-peer content delivery networks based on traffic usage data and interaction data. The calculation module is used to extract behavioral features of candidate users from log data and calculate a target score based on the behavioral features; the target score is used to indicate the confidence level of the candidate user's actual use of the peer-to-peer content delivery network. The user identification module is used to identify candidate users as actual users of the peer-to-peer content delivery network, provided that the target score meets the score requirements.

[0186] In one possible implementation, the system is also used for: Determine the initial score for each behavioral feature; the initial score indicates the degree to which the behavioral feature deviates from the baseline of normal user behavior. Adjust the initial scores based on log data and behavioral features to obtain the adjusted scores for each behavioral feature. Perform a weighted summation of the adjusted scores for each behavioral feature to obtain the target score.

[0187] In one possible implementation, the system is also used for: A directed weighted graph is constructed based on log data. In this graph, nodes are network node identifiers, directed edges indicate traffic transmission relationships between nodes, and edge weights indicate the proportion of traffic transmission between nodes. Network node identifiers include at least one of the following: user network address, autonomous system number. A causal relationship matrix is ​​constructed based on behavioral features. Rows and columns of the causal relationship matrix correspond to behavioral features, and element values ​​indicate whether a causal relationship exists between the behavioral feature in the row and column. Path scores for anomaly propagation paths are calculated based on the directed weighted graph, and the feature influence strength between behavioral features is determined based on the causal relationship matrix. Feature influence strength is represented by element values ​​in the causal relationship matrix. Initial scores are adjusted based on path scores and feature influence strength to obtain adjusted scores for each behavioral feature.

[0188] In one possible implementation, the system is also used for: Nodes exhibiting abnormal flow behavior in the directed weighted graph are identified as anomalous nodes. A flood propagation algorithm is used to trace the propagation path of the anomalous nodes along the edges of the directed weighted graph. The anomalousness of nodes along the propagation path, the weights between adjacent nodes, and the path depth are determined. A path score is calculated based on the anomalousness, the weights between adjacent nodes, and the path depth.

[0189] In one possible implementation, the system is also used for: Extract time-series data for any two behavioral features within a time window. Perform a Granger causality test on the time-series data and obtain the test results; the test results indicate whether historical data of one behavioral feature has predictive power for current data of another behavioral feature. Determine whether a causal relationship exists between the behavioral features based on the test results, and construct a causal relationship matrix based on the obtained determination results.

[0190] In one possible implementation, the system is also used for: For any given behavioral feature, determine the mean of the path scores associated with that feature. Based on the causal relationship matrix, determine the influence strength of other behavioral features on that feature, and calculate the product between the influence strength and the initial scores of other behavioral features. Perform a weighted summation on the mean and the product to obtain the processed result. Finally, sum the initial score and the processed result for that behavioral feature to obtain the adjusted score.

[0191] In one possible implementation, the system is also used for: The weighting coefficients are dynamically determined based on historical data corresponding to each behavioral feature. The adjusted scores for each behavioral feature and their corresponding weighting coefficients are then summed in a weighted manner to obtain the target score.

[0192] In one possible implementation, the system is also used for: If the historical false alarm rate exceeds the false positive threshold, the weight coefficient of that behavioral feature is reduced. If the proportion exceeds the proportion threshold, the weight coefficient of that behavioral feature is increased.

[0193] In one possible implementation, the system is also used for: If, based on traffic usage data, it is determined that a user's uplink traffic usage is greater than their downlink traffic usage, the user is assigned a first label. If, based on interaction data, it is determined that there is interaction between the user and the peer-to-peer content delivery network service, the user is assigned a second label. Users with both the first and second labels are identified as candidate users.

[0194] In one possible implementation, the system is also used for: Extract the domain name information requested by the user from the interaction data. If the domain name information matches the preset peer-to-peer content delivery network service domain name database, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is marked with a second tag.

[0195] In one possible implementation, the system is also used for: Extract the DNS resolution address corresponding to the user's requested domain name and the target address actually accessed by the user from the interaction data. If the DNS resolution address and the target address match successfully, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is then marked with a second tag.

[0196] In one possible implementation, the system is also used for: Behavioral characteristics include at least one of the following: traffic scale characteristics reflecting the user's bandwidth usage, port characteristics reflecting the user's access port behavior, protocol characteristics reflecting the user's protocol type, preset interaction characteristics reflecting the user and peer-to-peer content delivery network services, activity characteristics reflecting the user's traffic time distribution pattern, address association characteristics reflecting the consistency between the user's actual access address and the domain name resolution result, and traffic flow characteristics reflecting whether the user's traffic flows across domains.

[0197] In one possible implementation, the system is also used for: Application layer data from the deep packet inspection system is stored as application layer logs. Traffic usage data from the authentication, authorization, and accounting system is stored as traffic metering logs. Network layer traffic transmission data from the network flow analysis system and network path data from the border gateway protocol system are stored as traffic transmission logs.

[0198] The description of the processing flow of each module in the system and the interaction flow between each module can be found in the relevant descriptions in the above method embodiments, and will not be detailed here.

[0199] Corresponding to Figure 1 In addition to the user identification method in this disclosure, this embodiment also provides an electronic device 40, such as... Figure 4 The diagram shown is a structural schematic of an electronic device 40 provided in an embodiment of this disclosure, including: The system includes a processor 41, a memory 42, and a bus 43. The memory 42 stores execution instructions and includes main memory 421 and external memory 422. The main memory 421, also called internal memory, temporarily stores the computational data in the processor 41, as well as data exchanged with external memory such as a hard disk. The processor 41 exchanges data with the external memory 422 through the main memory 421. When the electronic device 40 is running, the processor 41 communicates with the memory 42 through the bus 43, causing the processor 41 to execute the following instructions: Acquire log data; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data. Extract interaction data between users and the peer-to-peer content delivery network service from the log data. Based on traffic usage data and interaction data, identify candidate users who are suspected of using the peer-to-peer content delivery network. Extract behavioral characteristics of the candidate users from the log data, and calculate a target score based on the behavioral characteristics; wherein the target score is used to indicate the confidence level of the candidate user's actual use of the peer-to-peer content delivery network. If the target score meets the scoring requirements, determine the candidate user as a user who is actually using the peer-to-peer content delivery network.

[0200] This disclosure also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the user identification method described in the above-described method embodiments. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0201] This disclosure also provides a computer program product 50, such as... Figure 5 The diagram shown is a schematic diagram of the structure of a computer program product 50 provided in an embodiment of this disclosure. The computer program product 50 carries a computer program 51. The program included in the computer program 51 can be used to execute the steps of the user identification method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0202] The user identification method, system, device, medium, and product according to embodiments of the present disclosure have been described above with reference to the accompanying drawings. In these embodiments, log data is first acquired; the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data. Interaction data between the user and the peer-to-peer (PCDN) content delivery network service is extracted from the log data. Then, based on the traffic usage data and interaction data, candidate users suspected of using the PCDN are identified. Behavioral characteristics of the candidate users are extracted from the log data, and a target score is calculated based on these behavioral characteristics; the target score indicates the confidence level of the candidate user's actual use of the PCDN. Finally, if the target score meets the scoring requirements, the candidate user is determined to be a user actually using the PCDN. Through the above processing method, the present disclosure acquires information related to PCDN user behavior from multiple dimensions through log data and identifies candidate users through traffic usage data and interaction data, achieving joint determination of traffic and behavioral dimensions, effectively reducing the probability of misidentifying normal users as PCDN users. Furthermore, the system measures the likelihood of a user possessing PCDN user behavior characteristics by calculating a target score. Finally, when the target score meets the preset score requirements, the user is determined to be a user who actually uses PCDN services, thereby ensuring the reliability of the determination process and further improving the accuracy of user identification.

[0203] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0204] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0205] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0206] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0207] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0208] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0209] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A user identification method characterized by, include: Obtain log data; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data; Extract the interaction data between the user and the peer-to-peer content delivery network service from the log data; Based on the traffic usage data and the interaction data, candidate users suspected of using peer-to-peer content delivery networks are identified; The behavioral characteristics of the candidate users are extracted from the log data, and a target score is calculated based on the behavioral characteristics; wherein, the target score is used to indicate the confidence level of the candidate user's actual use of the peer-to-peer content delivery network; If the target score meets the score requirements, the candidate user is determined to be a user who actually uses the peer-to-peer content delivery network.

2. The method of claim 1, wherein, The calculation of the target score based on the behavioral characteristics includes: Determine an initial score for each of the aforementioned behavioral characteristics; wherein the initial score is used to indicate the degree to which the behavioral characteristic deviates from the baseline of normal user behavior; Based on the log data and the behavioral characteristics, the initial score is adjusted to obtain the adjusted score for each behavioral characteristic; The adjusted scores corresponding to each of the aforementioned behavioral features are weighted and summed to obtain the target score.

3. The method of claim 2, wherein, The adjustment of the initial score based on the log data and the behavioral characteristics to obtain the adjusted score for each behavioral characteristic includes: A directed weighted graph is constructed based on the log data; wherein, the nodes of the directed weighted graph are network node identifiers, the directed edges of the directed weighted graph are used to indicate the traffic transmission relationship between nodes, the edge weights of the directed weighted graph are used to indicate the proportion of traffic transmission between nodes, and the network node identifiers include at least one of the following: user network address, autonomous system number; A causal relationship matrix is ​​constructed based on the behavioral features; wherein, the rows and columns of the causal relationship matrix correspond to the behavioral features respectively, and the element values ​​in the causal relationship matrix are used to indicate whether there is a causal relationship between the behavioral features corresponding to the row and the behavioral features corresponding to the column. The path score of the anomaly propagation path is calculated based on the directed weighted graph, and the feature influence strength between behavioral features is determined based on the causal relationship matrix; wherein the feature influence strength is represented by the element values ​​in the causal relationship matrix. The initial score is adjusted based on the path score and the feature influence strength to obtain the adjusted score for each behavioral feature.

4. The method of claim 3, wherein, The path score for calculating the anomaly propagation path based on the directed weighted graph includes: Nodes exhibiting abnormal flow behavior in the directed weighted graph are identified as abnormal nodes. The flood propagation algorithm is used to trace the anomaly propagation path from the anomaly node along the edge direction of the directed weighted graph; Determine the degree of anomaly of the nodes in the anomaly propagation path, the weight between adjacent nodes, and the path depth; The path score is calculated based on the degree of anomaly, the weights between adjacent nodes, and the path depth.

5. The method of claim 3, wherein, The construction of the causal relationship matrix based on the behavioral features includes: Extract time-series data of any two behavioral features within a time window; A Granger causality test is performed on the time-series data to obtain the test result; wherein the test result is used to indicate whether the historical data of one behavioral feature has a predictive effect on the current data of another behavioral feature; Based on the test results, determine whether there is a causal relationship between the behavioral characteristics, and construct the causal relationship matrix based on the obtained judgment results.

6. The method of claim 3, wherein, The adjustment of the initial score based on the path score and the feature influence strength to obtain the adjusted score for each behavioral feature includes: For any given behavioral feature, determine the mean of the path scores associated with that behavioral feature; Based on the causal relationship matrix, determine the influence strength of other behavioral features on the behavioral feature, and calculate the product between the influence strength of the feature and the initial score of the other behavioral features; The mean and the product are weighted and summed to obtain the processing result; The initial score and the processing result of the behavioral feature are summed to obtain the adjusted score of the behavioral feature.

7. The method of claim 2, wherein, The step of weighted summing of the adjusted scores corresponding to each of the aforementioned behavioral features to obtain the target score includes: The weighting coefficients are dynamically determined based on the historical data corresponding to each of the aforementioned behavioral characteristics. The target score is obtained by weighting and summing the adjusted scores and corresponding weight coefficients for each behavioral feature.

8. The method of claim 7, wherein, The historical data includes the historical false alarm rate of the behavioral feature and the proportion of the initial score of the behavioral feature in the path score; The dynamic determination of weight coefficients based on historical data corresponding to each of the aforementioned behavioral features includes: If the historical false alarm rate exceeds the false judgment threshold, the weight coefficient of this behavioral feature is reduced. If the percentage exceeds the percentage threshold, the weight coefficient of that behavioral feature is increased.

9. The method of claim 1, wherein, The process of identifying potential users of a peer-to-peer content delivery network based on the traffic usage data and the interaction data includes: If, based on the traffic usage data, it is determined that a user's uplink traffic usage is greater than their downlink traffic usage, then the user is marked as a first user. If, based on the interaction data, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, the user is marked with a second tag. Users with the first tag and the second tag are identified as candidate users.

10. The method of claim 9, wherein, If, based on the interaction data, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, the user is marked a second time, including: Extract the domain name information requested by the user from the interaction data; If the domain name information matches successfully with the preset peer-to-peer content delivery network service domain name database, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is marked with a second tag.

11. The method of claim 9, wherein, If, based on the interaction data, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, the user is marked a second time, including: Extract the DNS return address corresponding to the user's requested domain name and the target address actually accessed by the user from the interaction data; If the parsed return address and the target address match successfully, it is determined that there is an interaction between the user and the peer-to-peer content delivery network service, and the user is marked with a second tag.

12. The method of claim 1, wherein, The behavioral characteristics include at least one of the following: traffic scale characteristics reflecting the user's bandwidth usage, port characteristics reflecting the user's access port behavior, protocol characteristics reflecting the user's protocol type, preset interaction characteristics reflecting the user and peer-to-peer content delivery network services, activity characteristics reflecting the user's traffic time distribution pattern, address association characteristics reflecting the consistency between the user's actual access address and the domain name resolution result, and traffic flow characteristics reflecting whether the user's traffic flows across domains.

13. The method of claim 1, wherein, After obtaining the log data, the method further includes: Application layer data from the deep packet inspection system is stored as application layer logs; Traffic usage data from the authentication, authorization, and billing system is stored as traffic metering logs; Network layer traffic transmission data from the network flow analysis system and network path data from the border gateway protocol system are stored as traffic transmission logs.

14. A subscriber identification system, characterized in that include: The log acquisition module is used to acquire log data; wherein the log data includes at least one of the following: application layer data, network layer traffic transmission data, network path data, and traffic usage data; The data extraction module is used to extract interaction data between users and peer-to-peer content delivery network services from the log data; The user identification module is used to identify potential users of the peer-to-peer content delivery network based on the traffic usage data and the interaction data. A calculation module is used to extract behavioral features of the candidate users from the log data and calculate a target score based on the behavioral features; wherein, the target score is used to indicate the confidence level of the candidate user's actual use of the peer-to-peer content delivery network; The user identification module is used to identify the candidate user as a user who actually uses the peer-to-peer content delivery network, provided that the target score meets the score requirements.

15. An electronic device, comprising: include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is in operation, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the user identification method as described in any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the user identification method as described in any one of claims 1 to 13.

17. A computer program product, characterised in that, The computer program product is stored in a storage medium, and the program product is executed by at least one processor to implement the user identification method as described in any one of claims 1 to 13.