Intelligent monitoring and diagnostic methods for user behavior on cloud platforms for risk assessment
By combining the collection of user behavior characteristics and certificate status for joint diagnosis, the problem of independent certificate verification and behavior detection has been solved, enabling accurate assessment and graded response to user behavior on the cloud platform, and improving the risk assessment capabilities of the cloud platform.
Patent Information
- Application Number
- CN202611424859.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-15
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, certificate verification and user behavior detection are independent of each other in cloud platform security protection, which cannot effectively assess the confidence of access sessions in complex scenarios. Furthermore, single-dimensional behavior detection is easily simulated and it is difficult to distinguish between operational negligence, phishing attacks, and adversarial spoofing access.
By collecting users' direct and automated operational behavior characteristics, such as mouse movement and keyboard input, multi-dimensional confidence scores are calculated and combined with certificate status for joint diagnosis, generating risk diagnosis conclusions and executing graded responses.
It enables accurate identification of automated operation behaviors in complex scenarios, improves the accuracy and comprehensiveness of the cloud platform in identifying malicious access, solves the limitations of single-dimensional detection, and enhances the risk assessment and control capabilities.
Smart Images

Figure CN123122964A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent monitoring technology, specifically to a cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment. Background Technology
[0002] In the field of cloud platform security, website trustworthiness assessment is a core step in ensuring secure user access. Currently, the industry primarily relies on verifying the validity of security certificates to assess website trustworthiness. When a certificate expires or becomes abnormal, browsers or clients typically display a unified security warning. Meanwhile, user behavior detection technologies are mostly applied independently in anti-scraping or anti-fraud scenarios.
[0003] However, in the actual operating environment of cloud platforms, certificate verification and behavior detection are often independent and operate independently. Certificate verification only focuses on static attributes, while behavior detection relies heavily on single-dimensional features. When faced with complex scenarios where certificates have expired or are abnormal, but a large number of users still initiate access, existing technologies reveal significant limitations: on the one hand, existing mechanisms cannot calculate the confidence level of the current access session generated by the AI agent, let alone jointly evaluate it with the certificate status; on the other hand, relying solely on the static certificate status makes it difficult to effectively distinguish between a legitimate website's expired certificate due to operational negligence and a phishing website's forged certificate abnormality.
[0004] Furthermore, with the evolution of AI agent technology, existing single-dimensional behavior detection is easily simulated and bypassed by automated programs, causing the detection of spoofed access to adversarial automated operations to fail.
[0005] Therefore, there is an urgent need for an intelligent monitoring and diagnostic method that can deeply integrate the security certificate status with the multi-dimensional behavioral characteristics of cloud platform users, accurately calculate the confidence level of automated operations, and accurately distinguish between operational negligence, phishing attacks, and adversarial spoofing access. Summary of the Invention
[0006] The purpose of this invention is to provide a cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment, thereby solving the above-mentioned technical problems.
[0007] The objective of this invention can be achieved through the following technical solutions:
[0008] A cloud platform user behavior intelligent monitoring and diagnostic method for risk assessment includes the following steps:
[0009] The system obtains the security certificate status of the target website. When the security certificate expires, it triggers the collection of characteristic behaviors of the user behavior in this access session. The user behavior includes at least one of direct operation behavior and automated operation behavior. Direct operation behavior refers to the user directly performing page access, form filling and click operations through an input device. Automated operation behavior refers to the programmatic tool performing direct operation behavior on behalf of the user. The characteristic behaviors include mouse movement characteristics and keyboard input characteristics.
[0010] Based on the aforementioned characteristic behaviors, the confidence level of this access session belonging to automated operation behavior is calculated. Based on the confidence level and the status of the security certificate, a joint risk diagnosis is performed and a risk diagnosis conclusion is generated. The risk diagnosis conclusion is used to distinguish at least one of the following risk types: operation and maintenance negligence-related certificate expiration, phishing attack-related certificate anomaly, and adversarial automated operation behavior disguised access. According to the risk diagnosis conclusion, a risk assessment response matching the risk level of the risk diagnosis conclusion is executed.
[0011] As a further aspect of the present invention: the mouse movement feature includes the path curvature of the mouse movement trajectory and overshoot events, wherein the overshoot event refers to the event in which the cursor returns to the target position for correction after passing the target position during the mouse movement process;
[0012] The keyboard input features include at least one of text input methods and keystroke interval distribution, wherein the text input methods include key-by-key input and paste input.
[0013] As a further aspect of the present invention: the method for calculating the confidence level that the current access session belongs to an automated operation based on the aforementioned characteristic behavior includes:
[0014] Calculate the score for mouse movement dimension Where R is the path curvature, R0 is the preset path curvature threshold, and N over N is the number of overshoot events, and N0 is the preset overshoot event threshold.
[0015] Calculate the keyboard input dimension score N paste To paste the input count, N type The coefficient of variation (CV) represents the number of keystrokes per keystroke. k =σ / μ, where σ is the standard deviation of keystroke interval, μ is the mean of keystroke interval, and CV0 is the preset threshold for the coefficient of variation of keystroke interval;
[0016] Construct a certificate state vector T = (t1, t2, t3), where t1 ∈ {0, 1}, t1 = 1 indicates that the security certificate has expired, t2 ∈ {0, 1}, t2 = 1 indicates that the certificate authority is trustworthy, and t3 represents the normalized value of the domain name registration duration, and t3 = min(T con / T0,1),t3∈[0,1],T con T0 represents the length of time that has elapsed from the date of domain registration to the current moment, and is a preset reference threshold for domain registration duration.
[0017] Calculate the weight w of the i-th dimension i =w i 0 ×(1+α×t1×(1-t2)×(1-t3)), w i 0 Here, α is the preset baseline weight for the i-th dimension, and α is the preset adjustment coefficient, i=1,2;
[0018] Calculate the original confidence level c = (w1×s1+w2×s2) / (w1+w2), and correct the original confidence level c to obtain the confidence level C = c×(1-β×t1×t3), where β is the preset correction coefficient.
[0019] As a further aspect of the present invention: a method for performing joint risk diagnosis and generating risk diagnosis conclusions based on confidence level and security certificate status includes:
[0020] Calculate cross-dimensional dissimilarity And calculate the normalized cross-dimensional difference Ds=D / (1+γ×t1×t2), where γ is the preset normalization coefficient and γ>0;
[0021] When the confidence level C is lower than the preset first threshold C1 and the security certificate status is expired, and the target website's historical access records on the cloud platform meet the preset normal access conditions, a risk diagnosis conclusion of operation and maintenance negligence-related certificate expiration is generated.
[0022] When the confidence level C is higher than the preset second threshold C2 and the security certificate status is abnormal, a risk diagnosis conclusion of phishing attack type certificate abnormality is generated, where C2 > C1;
[0023] When the confidence level C is in (C1, C2), and the normalized cross-dimensional dissimilarity Ds exceeds the preset dissimilarity threshold D th At the same time, risk diagnostic conclusions are generated for adversarial automated operation behaviors that masquerade as access.
[0024] As a further aspect of the present invention: when the risk diagnosis conclusion is that the certificate expires due to operational negligence, the system will allow the process and display a first-level enhanced warning message. The first-level enhanced warning message includes a certificate expiration prompt and a chain of evidence. The chain of evidence includes at least one of the following: domain registration duration, historical security certificate type, and number of historical access users.
[0025] When the risk diagnosis conclusion is that the access is disguised by adversarial automated operation, perform two-factor authentication.
[0026] When the risk diagnosis conclusion is that the certificate is abnormal and a phishing attack is detected, interception is performed and a phishing attack alarm message is output.
[0027] As a further aspect of the present invention: obtaining the security certificate status of a website includes at least one of the following methods:
[0028] Query the certificate transparency log of the target website to obtain the historical certificate issuance records of the target website. The certificate transparency log refers to a public log system used for monitoring and auditing digital certificates.
[0029] Obtain the domain registration information of the target website, wherein the domain registration information includes at least one of the domain registration duration and domain registration entity information;
[0030] Obtain the certificate type information of the target website. The certificate type information is used to distinguish between domain verification certificates, organization verification certificates, and extended verification certificates.
[0031] As a further aspect of the present invention: when the security certificate has not expired, subsequent operations are stopped.
[0032] As a further aspect of the present invention: a method for executing a risk assessment response that matches the risk level of the risk diagnosis conclusion, based on the risk diagnosis conclusion, includes:
[0033] The risk diagnosis conclusions are pushed to the security operations center of the cloud platform. When the risk diagnosis conclusion is that the certificate expired due to operational negligence, a security certificate renewal reminder is generated for the target website. When the risk diagnosis conclusion is that the certificate is abnormal due to phishing attack or that the target website is accessed by disguised access due to adversarial automated operation behavior, the characteristic information of the target website is added to the threat intelligence database of the cloud platform.
[0034] The beneficial effects of this invention are as follows: (1) By collecting multi-dimensional behavioral features such as mouse movement and keyboard input, calculating the dimensional scores and constructing dynamic weights, the corrected confidence level is finally obtained. Compared with existing single-dimensional behavior detection, this solution overcomes the simulation and bypassing of single features by AI agents, greatly improves the accuracy of cloud platforms in identifying automated operation behaviors in complex access scenarios, and provides accurate data support for subsequent joint risk assessment.
[0035] (2) By combining the certificate state vector with user behavior confidence and cross-dimensional difference for joint diagnosis, the limitation of certificate verification and behavior detection being independent of each other is broken. By accurately distinguishing three types of risks—certificate expiration due to operational negligence, certificate anomalies due to phishing attacks, and adversarial automated operations disguised as access—and executing graded responses, the problem of not being able to identify malicious phishing websites based solely on static certificates is effectively solved, significantly improving the comprehensiveness and control capabilities of cloud platform risk assessment. Attached Figure Description
[0036] The invention will now be further described with reference to the accompanying drawings.
[0037] Figure 1 This is a flowchart illustrating the cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to the present invention.
[0038] Figure 2 This is a schematic diagram illustrating the execution of the cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to the present invention. Detailed Implementation
[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0040] Please see Figure 1 As shown, this invention is a cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment, comprising the following steps:
[0041] The system obtains the security certificate status of the target website. When the security certificate expires, it triggers the collection of characteristic behaviors of the user behavior in this access session. The user behavior includes at least one of direct operation behavior and automated operation behavior. Direct operation behavior refers to the user directly performing page access, form filling and click operations through an input device. Automated operation behavior refers to the programmatic tool performing direct operation behavior on behalf of the user. The characteristic behaviors include mouse movement characteristics and keyboard input characteristics.
[0042] Based on the aforementioned characteristic behaviors, the confidence level of this access session belonging to automated operation behavior is calculated. Based on the confidence level and the status of the security certificate, a joint risk diagnosis is performed and a risk diagnosis conclusion is generated. The risk diagnosis conclusion is used to distinguish at least one of the following risk types: operation and maintenance negligence-related certificate expiration, phishing attack-related certificate anomaly, and adversarial automated operation behavior disguised access. According to the risk diagnosis conclusion, a risk assessment response matching the risk level of the risk diagnosis conclusion is executed.
[0043] It should be noted that the first step is to obtain the security certificate status of the target website. This can be achieved by querying at least one of the following methods: querying the target website's certificate transparency log, obtaining domain registration information, or obtaining certificate type information. The certificate transparency log refers to a publicly available log system used for monitoring and auditing digital certificates; querying this log allows access to the target website's historical certificate issuance records. Domain registration information includes key data such as the domain registration duration and the domain registrant's information. Certificate type information distinguishes between domain verification certificates, organization verification certificates, and extended verification certificates. After obtaining the above information, the system will perform a real-time assessment of the security certificate status. When the security certificate is determined to be valid, it indicates that the static trust foundation of the current website still exists. To conserve cloud platform computing resources and avoid unnecessary interference, the system will stop subsequent session behavior collection and risk assessment operations. Only when the security certificate expires will the feature collection of user behavior in the current access session be triggered. This triggering mechanism constitutes the starting point for the dynamic assessment of this solution.
[0044] After triggering the data collection, the system will comprehensively acquire user behavior characteristics from the current access session on the target website. These user behaviors include at least one of direct operation behaviors and automated operation behaviors. Direct operation behaviors refer to users directly performing page access, form filling, and click operations through input devices, such as real users clicking links with a mouse or entering account passwords via a keyboard. Automated operation behaviors, on the other hand, refer to direct operation behaviors performed by programmatic tools on behalf of users, such as clicks and inputs simulated by web crawler scripts or AI agents. To distinguish between these two types of behavior, the feature behaviors collected in this embodiment specifically include mouse movement characteristics and keyboard input characteristics. The mouse movement characteristics include the path curvature of the mouse movement trajectory and overshoot events. An overshoot event refers to the event where, during mouse movement, the cursor returns to the target position after passing it, correcting its position. Logically, when real human users move the mouse, due to the physiological characteristics of their hand muscles, they often decelerate and make minor adjustments when approaching the target, inevitably resulting in path curvature and occasional overshoot phenomena such as passing the target and then pulling back. In contrast, automated programs typically use linear interpolation or Bézier curves to generate trajectories, resulting in extremely smooth or mechanically regular path curvatures. Furthermore, due to the lack of physiological feedback, unexpected overshoot corrections are usually absent. Therefore, these two characteristics effectively characterize the operator's biological attributes.
[0045] Furthermore, the keyboard input features include at least one of text input methods and keystroke interval distribution. The text input methods include sequential keystrokes and paste input. In specific implementation details, sequential keystrokes reflect the user's actual operation of pressing each key individually, while paste input typically means the user directly fills in content via the clipboard, a behavior extremely common in automated scripts. Keystroke interval distribution refers to the statistical regularity of the time difference between consecutive keystrokes. The keystroke intervals of real users usually conform to a specific statistical distribution, exhibiting some random fluctuations; while the keystroke intervals of automated tools are often fixed values or follow a simple uniform distribution. Based on the multi-dimensional features collected above, the system will further calculate the confidence level that this access session belongs to automated operation behavior. The calculation process of this confidence level integrates multi-dimensional behavior scores and adaptive weights of certificate status, and its specific calculation logic is as follows.
[0046] First, the system calculates the score for the mouse movement dimension. Where R is the path curvature, R0 is the preset path curvature threshold, and N... over The number of overshoot events is denoted by N0, which is a preset overshoot event threshold. The min function truncates the upper limit of each indicator to prevent any single indicator from becoming abnormally high and distorting the overall score. Equally weighting the path curvature and overshoot count provides a comprehensive measure of the anthropomorphism of mouse movement. If both curvature and overshoot count are very small, it indicates that the mouse movement is too mechanical, resulting in a low s1 score, suggesting a tendency towards automated operation. Next, the keyboard input dimension score is calculated. N paste To paste the input count, N type The coefficient of variation (CV) represents the number of keystrokes per keystroke. k =σ / μ, where σ is the standard deviation of keystroke interval, μ is the mean of keystroke interval, and CV0 is the preset threshold for the coefficient of variation of keystroke interval. The logic is: N paste Divide by N paste +N type This represents the proportion of paste operations in total input operations. A higher proportion indicates that the input behavior is less consistent with human typing habits, therefore, the smaller the value of 1 minus this proportion, ultimately leading to a lower S2 score. Meanwhile, CV... k Dividing by CV0 reflects the comparison between the actual keystroke interval variability and the preset human variability threshold. If the actual coefficient of variation CV0... k If the value is much smaller than the preset threshold CV0, it indicates that the keystroke interval is too uniform and has strong machine characteristics. The max function ensures that this penalty term will not have a negative value. Through the above two formulas, the system transforms complex biological behavioral characteristics into quantified scores s1 and s2.
[0047] After obtaining the behavioral dimension score, a certificate state vector T = (t1, t2, t3) is constructed, where t1 ∈ {0, 1}, t1 = 1 indicates that the security certificate has expired, t2 ∈ {0, 1}, t2 = 1 indicates that the certificate issuing authority is trustworthy, and t3 represents the normalized value of the domain name registration duration, and t3 = min(T con / T0,1),t3∈[0,1],T con T0 represents the elapsed time from the domain's registration date to the current moment, where T0 is a preset reference threshold for domain registration duration. Logically, the T vector incorporates the website's static reputation attribute to dynamically adjust the weights of behavioral characteristics. Specifically, it calculates the weight w of the i-th dimension. i =w i 0 ×(1+α×t1×(1-t2)×(1-t3)), w i 0 Let w be the preset baseline weight for the i-th dimension, and α be the preset adjustment coefficient, i=1,2. The logic of this weight calculation formula is crucial: when the certificate expires (t1=1), the issuing authority is untrustworthy (t2=0), and the domain name registration time is extremely short (t3 approaches 0), both (1 minus t2) and (1 minus t3) approach 1. At this time, the weight w... i This will result in maximum amplification. This means that in extreme cases where the website's static trust foundation collapses, the system will automatically increase its sensitivity to mouse and keyboard behavior, thus more rigorously scrutinizing whether the operator is an automated program. Conversely, if the certificate has expired but the issuing authority is trustworthy and the domain has been registered for a long time (e.g., t2=1 or t3=1), the weight will not be amplified, avoiding unintended harm to established, legitimate websites.
[0048] Based on the aforementioned weights, the system calculates the original confidence score c = (w1×s1 + w2×s2) / (w1 + w2). This formula uses a weighted average to combine the behavioral performance of both the mouse and keyboard. Subsequently, to further eliminate the interference of static certificate status on the behavioral confidence score, the system corrects the original confidence score c, obtaining the corrected confidence score C = c × (1 - β × t1 × t3). Here, β is a preset correction coefficient. The logic is as follows: When the certificate expires (t1 = 1) and the certificate authority is trustworthy (t2 = 1), t1 multiplied by t2 equals 1. In this case, the corrected confidence score C is multiplied by a decay factor of (1 - β). This is because a website with a certificate issued by a trustworthy authority experiencing certificate expiration is most likely due to negligence by maintenance personnel, rather than a malicious phishing website. In this scenario, if the behavioral detection results in a high automation confidence score c due to the user using certain automation plugins, directly classifying it as an attack would lead to a false positive. By multiplying by a decay factor, the system effectively reduces the oversensitivity to automated behavior in the "trusted institution and certificate expiration" scenario, making the final corrected confidence level C more objective and accurate.
[0049] After calculating the corrected confidence level C, the system will perform a joint risk diagnosis based on the confidence level and the status of the security certificate, and generate a risk diagnosis conclusion. This conclusion is used to distinguish at least one of the following: certificate expiration due to operational negligence, certificate anomaly due to phishing attacks, and adversarial automated operation behavior masquerading as access. To achieve accurate classification, the system first calculates the cross-dimensional dissimilarity. Mathematically, this formula essentially calculates the standard deviations of the two dimensions s1 and s2, reflecting the degree of dispersion between mouse and keyboard behavior. If the AI agent only simulates mouse clicks while ignoring the anthropomorphism of keyboard input, or only simulates keyboard input while the mouse trajectory is mechanical, then a huge difference will appear between s1 and s2, leading to a significant increase in the D value. Further, the system calculates the normalized cross-dimensional variance Ds = D / (1 + γ × t1 × t2), where γ is a preset normalization coefficient and γ > 0. The introduction of the certificate status factor (1 + γ × t1 × t2) is to appropriately smooth or reduce the sensitivity of the cross-dimensional variance when the certificate expires and the institution is trustworthy (which may be due to operational negligence), preventing misjudgments.
[0050] Based on the above calculation results, the system executes specific joint risk diagnosis logic. When the confidence level C is lower than the preset first threshold C1, and the security certificate status is expired, and the target website's historical access records on the cloud platform meet the preset normal access conditions, the system generates a risk diagnosis conclusion of certificate expiration due to operational negligence. Logically, a low C value indicates that the current access behavior is highly likely to be a real person's operation, and normal historical access indicates that the website usually has a good reputation. Combined with the certificate expiration, it can be inferred that the website administrator forgot to renew it. When the confidence level C is higher than the preset second threshold C2, and the security certificate status is abnormal, the system generates a risk diagnosis conclusion of certificate abnormality due to phishing attack. Logically, a high C value indicates that the current operation is highly likely to be an automated program (such as a batch phishing script), and the certificate abnormality further confirms its malicious nature. When the confidence level C is between (C1, C2), and the normalized cross-dimensional difference Ds exceeds the preset difference threshold D... th At that time, the system generates a risk diagnosis conclusion regarding adversarial automated operation behavior masquerading as access. Logically, the C value is in the middle range, indicating that the automation confidence is not extreme enough, but the Ds value is extremely high, indicating that the mouse and keyboard behavior characteristics are extremely mismatched. This is a typical characteristic of advanced AI agents attempting to masquerade as access by mixing real operations with machine operations.
[0051] Finally, after generating a risk diagnosis conclusion, the system will execute a risk assessment response matching the risk level of the conclusion. Specifically, when the risk diagnosis conclusion is an expired certificate due to operational negligence, the system will allow access and display a Level 1 enhanced warning message. This message includes a certificate expiration notification and a chain of evidence, which includes at least one of the following: domain registration duration, historical security certificate types, and the number of historical users accessing the site. This informs the user of the risk and allows them to prove their innocence. When the risk diagnosis conclusion is an adversarial automated operation disguised as access, the system will perform two-factor authentication, adding verification dimensions to block further operations by the AI agent. When the risk diagnosis conclusion is a phishing attack-related certificate anomaly, the system will intercept and output a phishing attack alarm. Furthermore, in terms of response mechanisms, the system can also push the risk diagnosis conclusion to the cloud platform's security operations center. When the conclusion is an expired certificate due to operational negligence, a security certificate renewal reminder for the target website will be generated; when the conclusion is a phishing attack-related certificate anomaly or an adversarial automated operation disguised as access, the target website's characteristic information will be added to the cloud platform's threat intelligence database. Through the aforementioned hierarchical response and data linkage mechanism, this embodiment achieves a complete closed loop from behavior monitoring and confidence calculation to joint diagnosis and intelligent response, significantly improving the cloud platform's risk assessment capabilities and security protection level in complex network environments.
[0052] In another preferred embodiment of the present invention, the mouse movement feature includes the path curvature of the mouse movement trajectory and overshoot events. The overshoot event refers to the event in which the cursor returns to the target position to correct itself after passing the target position during the mouse movement process.
[0053] The keyboard input features include at least one of text input methods and keystroke interval distribution, wherein the text input methods include key-by-key input and paste input.
[0054] In another preferred embodiment of the present invention, the method for calculating the confidence level that the current access session belongs to an automated operation behavior based on the characteristic behavior includes:
[0055] Calculate the score for mouse movement dimension Where R is the path curvature, R0 is the preset path curvature threshold, and N over N is the number of overshoot events, and N0 is the preset overshoot event threshold.
[0056] Calculate the keyboard input dimension score N paste To paste the input count, N type The coefficient of variation (CV) represents the number of keystrokes per keystroke. k =σ / μ, where σ is the standard deviation of keystroke interval, μ is the mean of keystroke interval, and CV0 is the preset threshold for the coefficient of variation of keystroke interval;
[0057] Construct a certificate state vector T = (t1, t2, t3), where t1 ∈ {0, 1}, t1 = 1 indicates that the security certificate has expired, t2 ∈ {0, 1}, t2 = 1 indicates that the certificate authority is trustworthy, and t3 represents the normalized value of the domain name registration duration, and t3 = min(T con / T0,1),t3∈[0,1],T con T0 represents the length of time that has elapsed from the date of domain registration to the current moment, and is a preset reference threshold for domain registration duration.
[0058] Calculate the weight w of the i-th dimension i =w i 0 ×(1+α×t1×(1-t2)×(1-t3)), w i 0 Here, α is the preset baseline weight for the i-th dimension, and α is the preset adjustment coefficient, i=1,2;
[0059] Calculate the original confidence level c = (w1×s1+w2×s2) / (w1+w2), and correct the original confidence level c to obtain the confidence level C = c×(1-β×t1×t3), where β is the preset correction coefficient.
[0060] In another preferred embodiment of the present invention, the method for performing joint risk diagnosis based on confidence level and security certificate status and generating risk diagnosis conclusion includes:
[0061] Calculate cross-dimensional dissimilarity And calculate the normalized cross-dimensional difference Ds=D / (1+γ×t1×t2), where γ is the preset normalization coefficient and γ>0;
[0062] When the confidence level C is lower than the preset first threshold C1 and the security certificate status is expired, and the target website's historical access records on the cloud platform meet the preset normal access conditions, a risk diagnosis conclusion of operation and maintenance negligence-related certificate expiration is generated.
[0063] When the confidence level C is higher than the preset second threshold C2 and the security certificate status is abnormal, a risk diagnosis conclusion of phishing attack type certificate abnormality is generated, where C2 > C1;
[0064] When the confidence level C is in (C1, C2), and the normalized cross-dimensional dissimilarity Ds exceeds the preset dissimilarity threshold D th At the same time, risk diagnostic conclusions are generated for adversarial automated operation behaviors that masquerade as access.
[0065] In another preferred embodiment of the present invention, when the risk diagnosis conclusion is that the certificate expires due to operational negligence, the process is to allow the process and display a first-level enhanced warning message. The first-level enhanced warning message includes a certificate expiration prompt and a chain of evidence display. The chain of evidence display includes at least one of the following: domain name registration duration, historical security certificate type, and number of historical accessing users.
[0066] When the risk diagnosis conclusion is that the access is disguised by adversarial automated operation, perform two-factor authentication.
[0067] When the risk diagnosis conclusion is that the certificate is abnormal and a phishing attack is detected, interception is performed and a phishing attack alarm message is output.
[0068] In another preferred embodiment of the present invention, obtaining the security certificate status of the website includes at least one of the following methods:
[0069] Query the certificate transparency log of the target website to obtain the historical certificate issuance records of the target website. The certificate transparency log refers to a public log system used for monitoring and auditing digital certificates.
[0070] Obtain the domain registration information of the target website, wherein the domain registration information includes at least one of the domain registration duration and domain registration entity information;
[0071] Obtain the certificate type information of the target website. The certificate type information is used to distinguish between domain verification certificates, organization verification certificates, and extended verification certificates.
[0072] In another preferred embodiment of the present invention, subsequent operations are stopped when the security certificate has not expired.
[0073] In another preferred embodiment of the present invention, a method for executing a risk assessment response that matches the risk level of the risk diagnosis conclusion includes:
[0074] The risk diagnosis conclusions are pushed to the security operations center of the cloud platform. When the risk diagnosis conclusion is that the certificate expired due to operational negligence, a security certificate renewal reminder is generated for the target website. When the risk diagnosis conclusion is that the certificate is abnormal due to phishing attack or that the target website is accessed by disguised access due to adversarial automated operation behavior, the characteristic information of the target website is added to the threat intelligence database of the cloud platform.
[0075] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.
Claims
1. A cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment, characterized in that, Includes the following steps: The system obtains the security certificate status of the target website. When the security certificate expires, it triggers the collection of characteristic behaviors of the user in this access session. The user behaviors include at least one of direct operation behaviors and automated operation behaviors. The direct operation behaviors refer to the user directly performing page access, form filling and click operations through the input device. The automated operation behaviors refer to the programmatic tools performing direct operation behaviors on behalf of the user. The characteristic behaviors include mouse movement characteristics and keyboard input characteristics; Based on the aforementioned characteristic behaviors, the confidence level of this access session belonging to automated operation behavior is calculated. Based on the confidence level and the status of the security certificate, a joint risk diagnosis is performed and a risk diagnosis conclusion is generated. The risk diagnosis conclusion is used to distinguish at least one of the following risk types: operation and maintenance negligence-related certificate expiration, phishing attack-related certificate anomaly, and adversarial automated operation behavior disguised access. According to the risk diagnosis conclusion, a risk assessment response matching the risk level of the risk diagnosis conclusion is executed.
2. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, The mouse movement features include the path curvature of the mouse movement trajectory and overshoot events. The overshoot event refers to the event in which the cursor returns to the target position to correct itself after passing the target position during mouse movement. The keyboard input features include at least one of text input methods and keystroke interval distribution, wherein the text input methods include key-by-key input and paste input.
3. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, The methods for calculating the confidence level that this access session belongs to automated operation behavior based on the aforementioned characteristic behaviors include: Calculate the score for mouse movement dimension Where R is the path curvature, R0 is the preset path curvature threshold, and N over N is the number of overshoot events, and N0 is the preset overshoot event threshold. Calculate the keyboard input dimension score N paste To paste the input count, N type The coefficient of variation (CV) represents the number of keystrokes per keystroke. k =σ / μ, where σ is the standard deviation of keystroke interval, μ is the mean of keystroke interval, and CV0 is the preset threshold for the coefficient of variation of keystroke interval; Construct a certificate state vector T = (t1, t2, t3), where t1 ∈ {0, 1}, t1 = 1 indicates that the security certificate has expired, t2 ∈ {0, 1}, t2 = 1 indicates that the certificate authority is trustworthy, and t3 represents the normalized value of the domain name registration duration, and t3 = min(T con / T0,1),t3∈[0,1],T con T0 represents the length of time that has elapsed from the date of domain registration to the current moment, and is a preset reference threshold for domain registration duration. Calculate the weight w of the i-th dimension i =w i 0 ×(1+α×t1×(1-t2)×(1-t3)), w i 0 Here, α is the preset baseline weight for the i-th dimension, and α is the preset adjustment coefficient, i=1,2; Calculate the original confidence level c = (w1×s1+w2×s2) / (w1+w2), and correct the original confidence level c to obtain the confidence level C = c×(1-β×t1×t3), where β is the preset correction coefficient.
4. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, Methods for joint risk diagnosis and generation of risk diagnosis conclusions based on confidence level and security certificate status include: Calculate cross-dimensional dissimilarity And calculate the normalized cross-dimensional difference Ds=D / (1+γ×t1×t2), where γ is the preset normalization coefficient and γ>0; When the confidence level C is lower than the preset first threshold C1 and the security certificate status is expired, and the target website's historical access records on the cloud platform meet the preset normal access conditions, a risk diagnosis conclusion of operation and maintenance negligence-related certificate expiration is generated. When the confidence level C is higher than the preset second threshold C2 and the security certificate status is abnormal, a risk diagnosis conclusion of phishing attack type certificate abnormality is generated, where C2 > C1; When the confidence level C is in (C1, C2), and the normalized cross-dimensional dissimilarity Ds exceeds the preset dissimilarity threshold D th At the same time, risk diagnostic conclusions are generated for adversarial automated operation behaviors that masquerade as access.
5. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, When the risk diagnosis conclusion is that the certificate expired due to operational negligence, the process is allowed and a first-level enhanced warning message is displayed. The first-level enhanced warning message includes a certificate expiration prompt and a chain of evidence. The chain of evidence includes at least one of the following: domain registration duration, historical security certificate type, and number of historical accessing users. When the risk diagnosis conclusion is that the access is disguised by adversarial automated operation, perform two-factor authentication. When the risk diagnosis conclusion is that the certificate is abnormal and a phishing attack is detected, interception is performed and a phishing attack alarm message is output.
6. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, Obtaining the website's security certificate status includes at least one of the following methods: Query the certificate transparency log of the target website to obtain the historical certificate issuance records of the target website. The certificate transparency log refers to a public log system used for monitoring and auditing digital certificates. Obtain the domain registration information of the target website, wherein the domain registration information includes at least one of the domain registration duration and domain registration entity information; Obtain the certificate type information of the target website. The certificate type information is used to distinguish between domain verification certificates, organization verification certificates, and extended verification certificates.
7. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, Stop subsequent operations if the security certificate has not expired.
8. The cloud platform user behavior intelligent monitoring and diagnosis method for risk assessment according to claim 1, characterized in that, Based on the risk diagnosis conclusion, methods for implementing a risk assessment response that matches the risk level of the risk diagnosis conclusion include: The risk diagnosis conclusions are pushed to the security operations center of the cloud platform. When the risk diagnosis conclusion is that the certificate expired due to operational negligence, a security certificate renewal reminder is generated for the target website. When the risk diagnosis conclusion is that the certificate is abnormal due to phishing attack or that the target website is accessed by disguised access due to adversarial automated operation behavior, the characteristic information of the target website is added to the threat intelligence database of the cloud platform.
