A method and system for online monitoring of abnormal data in financial service applications
By analyzing user behavior data of financial service application system and combining with the isolated forest algorithm, the problem of low monitoring accuracy caused by the failure to utilize causal correlation in the existing technology is solved, and efficient abnormal data monitoring of financial service applications is achieved, and security is improved and monitoring accuracy.
Patent Information
- Application Number
- CN202410740166.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-07
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-06-07
AI Technical Summary
The existing financial service application abnormal data monitoring scheme fails to fully utilize the time series characteristics of the data and the causal correlation between events, resulting in low monitoring accuracy and inability to effectively deal with security risks in financial service applications.
By collecting the user login failure frequency, unauthorized access application frequency and access module frequency of the financial service application system, a security log vector is formed, data consistency score, operation sequence correlation coefficient and data change trend coefficient, and abnormality is performed in combination with the isolated forest algorithm. Online data monitoring reflects the characteristics of causal correlation.
It improves the accuracy and sensitivity of abnormal data monitoring of financial services applications, can promptly detect potential security risks, and reduce economic losses and user trust crises.
Smart Images

Figure CN118569871B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a method and system for online monitoring of abnormal data in financial services applications. Background Art
[0002] With the rapid development of financial technology, financial service applications have become an indispensable part of modern life. However, with the growth of the number of users and the diversification of service types, financial service applications are facing increasingly serious security challenges. Cybersecurity incidents such as hacker attacks, account theft, and fraudulent transactions occur frequently, seriously affecting the asset security and privacy protection of financial institutions and their customers. In order to maintain the sound operation of the financial system and the rights and interests of users, it is crucial to establish an efficient and accurate online monitoring method for abnormal data, which can timely detect and prevent potential security risks, reduce economic losses and user confidence crisis.
[0003] Currently, many financial service applications have adopted basic security protection measures, such as login verification, permission management, firewalls, etc. However, existing anomaly monitoring solutions fail to fully utilize the time series characteristics of data and the causal relationship between events, resulting in low accuracy of online monitoring of abnormal data. Summary of the invention
[0004] In order to solve the above technical problems, the purpose of this application is to provide a method and system for online monitoring of abnormal data in financial service applications. The technical solutions adopted are as follows:
[0005] In a first aspect, an embodiment of the present application provides a method for online monitoring of abnormal data in financial services applications, the method comprising the following steps:
[0006] Collect the user's daily login failure frequency, unauthorized access application frequency and access module frequency in the financial service application system to form the user's daily security log vector; obtain the user's daily login failure time set, access time set and unauthorized application time set;
[0007] Analyze the fluctuation and change trend of the user's login failure frequency, unauthorized access application frequency, and access module frequency respectively, and obtain the user's daily data consistency score;
[0008] Based on the difference between the elements in the user's daily login failure time set and the access time set, determine the user's daily operation sequence correlation coefficient;
[0009] Analyze the correlation between the frequency of user login failures and the frequency of unauthorized access applications, and obtain the user's daily data change trend coefficient;
[0010] Combining the operation sequence correlation coefficient with the data change trend coefficient and the difference between the unauthorized application time set and the login failure time set, the user's daily data correlation change index is obtained;
[0011] In combination with the data consistency score, the data association change index and the security log vector, an isolation forest algorithm is used to monitor abnormal data in financial services applications online.
[0012] In one embodiment, the module access frequency is the sum of the number of times a user accesses a module for which he / she has permission and the frequency of unauthorized access applications.
[0013] In one embodiment, determining the data consistency score includes:
[0014] The login failure frequencies of the user on any day and all days before the any day are combined into a login failure sequence for the any day, the login failure sequence is subjected to the moving average method to obtain each moving average to form a moving average sequence of the login failure sequence, and the data comprehensive fluctuation coefficient of the login failure frequency on the any day is determined based on the difference between adjacent elements in the moving average sequence of the login failure sequence;
[0015] The same calculation method as the data comprehensive fluctuation coefficient of the frequency of failed logins on any day is used to obtain the data comprehensive fluctuation coefficient of the frequency of unauthorized access applications and the frequency of access modules on any day;
[0016] Calculate the difference between the data comprehensive fluctuation coefficient of the access module frequency and the data comprehensive fluctuation coefficient of the unauthorized access application frequency, record it as the first difference, calculate the difference between the unauthorized access application frequency of any one day and the average of the unauthorized access application frequencies of all days before any one day, record it as the second difference;
[0017] Determine the data consistency score of any day by comprehensively analyzing the data comprehensive fluctuation coefficient of the login failure frequency, the first difference, and the second difference;
[0018] The data consistency score of any one day is negatively correlated with the data comprehensive fluctuation coefficient of the login failure frequency and the second difference, and is positively correlated with the first difference.
[0019] In one embodiment, the process of determining the comprehensive fluctuation coefficient of the data of the login failure frequency is as follows:
[0020] The difference between each element and the adjacent element in the moving average sequence of the login failure sequence is calculated, recorded as the adjacent difference, all the adjacent differences in the moving average sequence of the login failure sequence are forward fused, and the forward fusion result is used as the data comprehensive fluctuation coefficient of the login failure frequency.
[0021] In one embodiment, the process of determining the operation sequence correlation coefficient is as follows:
[0022] If all elements in the daily access time set are less than or equal to the minimum value in the daily login failure time set, the daily operation sequence correlation coefficient is the first preset value; otherwise, the daily operation sequence correlation coefficient is the second preset value, wherein the first preset value is less than the second preset value.
[0023] In one embodiment, the process of determining the data change trend coefficient includes:
[0024] Calculate the correlation between the daily failed login sequence and the unauthorized access application sequence, calculate the growth rate of each element in the daily failed login sequence relative to the previous element, record it as the first growth rate, calculate the growth rate of each element in the daily unauthorized access application sequence relative to the previous element, record it as the second growth rate, forward fuse the difference between the first growth rate and the second growth rate of all the same positions, and the correlation, and use the forward fusion result as the daily data change trend coefficient.
[0025] In one embodiment, the process of determining the data association change index includes:
[0026] Calculate the difference between the mean values of all elements in the unauthorized application time set and the login failure time set every day, record it as the third difference, and determine the data association change index every day by combining the third difference, the data change trend coefficient, and the operation sequence correlation coefficient every day;
[0027] The daily data association change index is positively correlated with the daily data change trend coefficient, the third difference, and the operation sequence correlation coefficient.
[0028] In one embodiment, the online monitoring of abnormal data of financial services applications includes:
[0029] The daily security log vector is used as the input of the isolation forest algorithm. The weight of each isolated tree is determined based on the data consistency score and data association change index distribution of the security log vector corresponding to the number of days contained in each isolated tree in the isolation forest algorithm, as well as the depth of the left and right subtrees of each isolated tree.
[0030] The output of the isolation forest algorithm is the anomaly score of the security log vector of each day. When the anomaly score is less than a preset threshold, the corresponding security log vector is determined to be abnormal data; otherwise, the corresponding security log vector is determined to be normal data.
[0031] In one embodiment, the weight of each isolated tree is determined as follows:
[0032] Calculate the discrete degree of the data consistency score of all security log vectors corresponding to the days contained in each isolated tree, calculate the average value of the data association change index of all security log vectors corresponding to the days contained in each isolated tree, and calculate the difference between the maximum and minimum depths of the left and right subtrees of each isolated tree, which is recorded as the fourth difference;
[0033] Determining a weight of each isolated tree based on the dispersion degree, the average value and the fourth difference;
[0034] The weight of each isolated tree is positively correlated with the dispersion degree, the average value and the fourth difference.
[0035] In a second aspect, an embodiment of the present application also provides an online monitoring system for abnormal data of financial services applications, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.
[0036] This application has at least the following beneficial effects:
[0037] This application collects the user's daily login failure frequency, unauthorized access application frequency and access module frequency in the financial service application system to form the user's daily security log vector, obtains the user's daily login failure time set, access time set and unauthorized application time set, analyzes the fluctuation and change trend of the user's login failure frequency, unauthorized access application frequency and access module frequency respectively, and obtains the user's daily data consistency score, which reflects the similarity of the user's login failure frequency, unauthorized access application frequency and access module frequency on all days before any day, and then judges the security of the user's financial service application data on any day, thereby improving the reliability of abnormal data monitoring of financial service applications; based on the difference between the elements in the user's daily login failure time set and the access time set, the user's daily operation sequence correlation coefficient is determined, which reflects the chronological relationship between the time points in the login failure time set and the access time set, and reflects the possibility of the user's account being attacked by attackers. In the process of abnormal data monitoring of financial service applications, causal correlation features are added to improve the accuracy of abnormal data monitoring of financial service applications; the correlation between the user's login failure frequency and the unauthorized access application frequency is analyzed to obtain the user's daily data change trend coefficient, and the difference between the operation sequence correlation coefficient and the data change trend coefficient, as well as the difference between the unauthorized application time set and the login failure time set, to obtain the user's daily data association change index, which reflects the possibility of the user's financial service application system being attacked and abnormal data appearing every day. Combined with the data consistency score, the data association change index and the security log vector, the isolation forest algorithm is used to complete the online monitoring of abnormal data of financial service applications. The value and timeliness of abnormal monitoring of security log vectors are reflected in the isolation forest algorithm. While ensuring the accuracy of abnormal data monitoring, the sensitivity of abnormal monitoring is also improved, effectively responding to risk challenges in financial service application systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0039] Figure 1 A flowchart of a method for online monitoring of abnormal data in financial services applications provided by an embodiment of the present application;
[0040] Figure 2 Construct a flow chart for the weights of isolated trees. DETAILED DESCRIPTION
[0041] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following is a detailed description of the method and system for online monitoring of abnormal data in financial services applications proposed in the present application, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0042] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
[0043] The following is a detailed description of a specific scheme of an online monitoring method and system for abnormal data in financial services applications provided by the present application in conjunction with the accompanying drawings.
[0044] See also Figure 1 , which shows a flowchart of a method for online monitoring of abnormal data in financial service applications provided by an embodiment of the present application, the method comprising the following steps:
[0045] S1, collects the user's daily login failure frequency, unauthorized access application frequency and access module frequency in the financial service application system to form the user's daily security log vector; obtains the user's daily login failure time set, access time set and unauthorized application time set.
[0046] Download the user's security operation log from the security operation log module of the financial service application system, and count the data for each day according to a one-day cycle. Statistically obtain the user's daily login failure frequency, unauthorized access application frequency, and access module frequency, where the login failure frequency represents the number of times the user fails to log in every day, the unauthorized access application frequency represents the number of times the user applies to access the unauthorized module every day, and the access module frequency represents the sum of the number of times the user accesses the authorized module and the number of times the user applies to access the unauthorized module every day. The user's daily login failure frequency, unauthorized access application frequency, and access module frequency are combined into the user's daily security log vector. For example, the user's login failure frequency on day a is 4, the unauthorized access application frequency is 1, and the access module frequency is 2, then the user's security log vector on day a is (4, 1, 2).
[0047] The time interval between each login failure of the user and the zero o'clock of each day is taken as the unit of seconds to form the login failure time set of the user. In the same way, the time interval between the starting time point of each access to any module by the user and the zero o'clock of each day is taken as the unit of seconds to form the access time set of the user every day. The time interval between the starting time point of each unauthorized access application by the user and the zero o'clock of each day is taken as the unit of seconds to form the unauthorized application time set of the user every day. Taking the login failure time set of the user's day a as an example, if the login failure time point of the user's day a is 0:02:20, the corresponding element in the login failure time set is 140, which is the number of seconds contained in 2 minutes and 20 seconds.
[0048] If a user's account has no operations within a day, the account is considered dormant on that day, and the value of each component in the security log vector corresponding to the dormant day is the average value of each component in the security log vector corresponding to all non-dormant days of the user.
[0049] S2, analyze the fluctuation and change trend of the user's login failure frequency, unauthorized access application frequency and access module frequency, and obtain the user's daily data consistency score.
[0050] Further, as an example of the present application, the process of constructing the user's daily data consistency score is as follows:
[0051] First, when users operate the financial service application system, due to the development of user habits, the security operation log data collected in most cases are highly similar. For example, there are almost no or very few login failures, and the operated modules are often modules for which the users have permissions, and unauthorized access requests are not frequent.
[0052] Taking the ath day as an example, the login failure frequency of the previous a days is arranged in chronological order to form a login failure sequence for the ath day, and then the login failure sequence and the subsequence length δ when performing moving average are used as input, and the moving average method is used to obtain each moving average. In this example, the subsequence length δ during the moving average is 5, and the implementer can set it according to the actual situation, and this application does not limit it here. The moving average method is an existing well-known technology, and the specific process is not described in detail in this application; all the moving averages obtained are arranged in chronological order to form a moving average sequence L of the login failure frequency of the ath day a , using the same calculation method as the moving average sequence of the frequency of failed logins on day a, calculate the moving average sequence P of the frequency of unauthorized access applications on day a a and the moving average sequence Q of the access module frequency a .
[0053] The moving average sequence L of the number of failed logins on day a aFor example, calculate the moving average sequence L a The difference between each element and the previous element is recorded as the adjacent difference, and the moving average sequence L a All the adjacent differences in the data are forward fused, and the forward fusion result is used as the data comprehensive fluctuation coefficient of the frequency of login failures on the ath day.
[0054] It should be noted that the difference indicates the degree of difference between two variables, which can be calculated by difference, ratio, etc. Forward fusion means combining multiple variables in a way to enhance the overall effect, which can be calculated by addition, multiplication, etc.
[0055] In one embodiment of the present application, the moving average sequence L is calculated. a The difference between each element and the previous element in the moving average sequence L a The sum of all the differences in the data is taken as the comprehensive fluctuation coefficient of the frequency of failed logins on day a.
[0056] The same calculation method as the data comprehensive fluctuation coefficient of the frequency of failed logins on day a is used to calculate the data comprehensive fluctuation coefficient of the frequency of unauthorized access applications on day a and the data comprehensive fluctuation coefficient of the frequency of access modules on day a.
[0057] Calculate the difference between the data comprehensive fluctuation coefficient of the access module frequency and the data comprehensive fluctuation coefficient of the unauthorized access application frequency, record it as the first difference, calculate the difference between the unauthorized access application frequency on any day and the average of the unauthorized access application frequencies on all days before the any day, record it as the second difference;
[0058] Determine the data consistency score of any day by comprehensively analyzing the data comprehensive fluctuation coefficient of the login failure frequency, the first difference, and the second difference;
[0059] The data consistency score of any one day is negatively correlated with the data comprehensive fluctuation coefficient of the login failure frequency and the second difference, and is positively correlated with the first difference.
[0060] It should be understood that a positive correlation means that the dependent variable will increase as the independent variable increases, and the dependent variable will decrease as the independent variable decreases. Specifically, it can be an additive relationship, a multiplicative relationship, or the power of an exponential function. A negative correlation means that the dependent variable will decrease as the independent variable increases, and the dependent variable will increase as the independent variable decreases. Specifically, it can be a subtractive relationship, a division relationship, etc.
[0061] In one embodiment of the present application, taking the ath day as an example, the sum of the data comprehensive fluctuation coefficient of the frequency of unauthorized access applications on the ath day and a preset adjustment factor greater than 0 is calculated, the ratio of the data comprehensive fluctuation coefficient of the access module frequency on the ath day to the summation result, i.e., the first difference, is calculated, the subtraction result of the frequency of unauthorized access applications on the ath day and the average frequency of unauthorized access applications on all days before the ath day, i.e., the second difference, the subtraction result is used as the independent variable of the sign function, when the independent variable of the sign function is less than or equal to 0, the function value of the sign function is 0, otherwise, the function value of the sign function is the value of the independent variable, the function value of the sign function is used as the exponent of an exponential function with a natural constant as the base, the calculation result of the exponential function and the product of the data comprehensive fluctuation coefficient of the frequency of failed logins on the ath day are calculated, the reciprocal of the sum of the product and the preset adjustment coefficient greater than 0 is calculated, and the product of the reciprocal and the ratio is used as the data consistency score for the ath day.
[0062] It should be noted that the adjustment factor preset to be greater than 0 and the adjustment coefficient preset to be greater than 0 are both for avoiding the denominator being 0 and causing the calculation to be impossible. As an embodiment of the present application, the adjustment factor preset to be greater than 0 and the adjustment coefficient preset to be greater than 0 are both artificially set to 0.01, and the implementer can set them according to the actual situation. The present application does not limit it here. The subtraction result of calculating the frequency of unauthorized access applications on the a-th day and the average frequency of unauthorized access applications on the previous a-th day is used as the independent variable of the sign function, and the function value of the sign function is used as the exponent of the exponential function with the natural constant as the base. The purpose is to measure whether the frequency of unauthorized access applications on the a-th day is normal. If the frequency of unauthorized access applications on the a-th day is less than the average frequency of unauthorized access applications on the previous a-th day, it means that the frequency of unauthorized access applications on the a-th day may be normal. At this time, the influence of the frequency of unauthorized access applications on the a-th day on the data consistency score of the a-th day is reduced. Otherwise, it means that the frequency of unauthorized access applications on the a-th day may be abnormal, and the influence of the frequency of unauthorized access applications on the a-th day on the data consistency score of the a-th day should be magnified.
[0063] When there are no anomalies in the security operation log data of the current day a, it means that the risk of the account is relatively small, and the security operation log data of each day are highly similar. Therefore, after moving average, each moving average fluctuates around a fixed value, but this fluctuation is very small and after positive and negative offset, the comprehensive data fluctuation coefficient is small or close to 0, so the first difference is large, and the comprehensive data fluctuation coefficient of the frequency of failed logins on the ath day is small; and because the security operation log data of each day are highly similar, the frequency of unauthorized access applications on the ath day is close to the average frequency of unauthorized access applications on the previous a day, so the calculation result of the exponential function is small. Therefore, when there are no anomalies in the security operation log data of the current day a, the data consistency score of the ath day is larger.
[0064] On the contrary, if there is an abnormality in the frequency of failed logins, its moving average value will be larger, and the comprehensive fluctuation coefficient of the frequency of failed logins on day a will be larger, which will ultimately lead to a smaller data consistency score; or if the frequency of unauthorized access applications increases, the calculation result of the exponential function will be larger, which will ultimately lead to a smaller data consistency score.
[0065] S3, determining the correlation coefficient of the user's daily operation sequence based on the difference between the elements in the user's daily login failure time set and the access time set.
[0066] For the time points in the user's daily login failure time set, it is very likely that the attacker is unfamiliar with the financial service application system and constantly tries to crack the user's password, resulting in login failure. The time points in the access time set represent the starting time points for accessing any module in the financial service application system after the user successfully logs in. Therefore, analyzing the sequence of time points in the login failure time set and the access time set, that is, determining the correlation coefficient of the user's daily operation sequence, can assist in judging the user's account security.
[0067] If all elements in the daily access time set are less than or equal to the minimum value in the daily login failure time set, the daily operation sequence correlation coefficient is the first preset value; otherwise, the daily operation sequence correlation coefficient is the second preset value, wherein the first preset value is less than the second preset value.
[0068] Among them, the first preset value is manually set to 1, and the second preset value is manually set to 2. The implementer can set it according to the actual situation, and this application does not impose any restrictions on this.
[0069] When all elements in the daily access time set are less than or equal to the minimum value in the daily login failure time set, it means that the time points in the access time set are all before the time points in the login failure time set, reflecting that when the user's account is attacked by the attacker, there is no successful login phenomenon, indicating that the user account is relatively safe at this time, and the user's operation sequence correlation coefficient is small. Otherwise, it means that after the login failure occurs, the module in the financial service application system is also accessed. At this time, it is possible that the attacker has successfully cracked the financial service application system, and the module in the financial service application system may be accessed by the attacker, which means that the user's account is at risk and the user's operation sequence correlation coefficient is large.
[0070] S4, analyze the correlation between the user's login failure frequency and the frequency of unauthorized access application, and obtain the user's daily data change trend coefficient.
[0071] When a user's account is attacked by an attacker, the attacker first needs to continuously try to crack the user's password, which will increase the frequency of the user's login failures. When the login is successful, because the attacker is not familiar with the financial service application system and is not clear about the user's own permissions, the attacker will continuously access various modules to view the user's permission modules and information, which will increase the frequency of module access and unauthorized access applications. The attacker will also try to use the account's existing permissions to gain unauthorized access by finding and exploiting loopholes in the permission management of the financial service application system. In this process, the frequency of unauthorized access applications will increase faster than the frequency of module access. When users use the financial service application system, they are more familiar with the financial service application system and are clear about their own permissions. Therefore, they perform purposeful operations in the financial service application system and will not frequently click on various modules or unauthorized access modules.
[0072] Therefore, based on the above analysis, the correlation between the daily failed login sequence and the unauthorized access application sequence is calculated, the growth rate of each element in the daily failed login sequence relative to the previous element is calculated, recorded as the first growth rate, and the growth rate of each element in the daily unauthorized access application sequence relative to the previous element is calculated, recorded as the second growth rate. For the daily failed login sequence and unauthorized access application sequence, the difference between the first growth rate and the second growth rate in all the same positions, as well as the correlation are forward fused, and the forward fusion result is used as the daily data change trend coefficient.
[0073] It should be noted that the correlation represents the correlation between sequences, and can be specifically calculated using Pearson correlation coefficient, cosine similarity, Spearman rank correlation coefficient, etc.
[0074] In one embodiment of the present application, the Pearson correlation coefficient of the daily login failure sequence and the unauthorized access application sequence is calculated. For the daily login failure sequence and the unauthorized access application sequence, the difference between the first growth rate and the second growth rate at the same position is recorded as the trend difference, and the trend difference is used as the independent variable of the symbol function. When the independent variable of the symbol function is less than or equal to 0, the function value of the symbol function is 0, otherwise, the function value of the symbol function is the value of the independent variable. The cumulative sum of the function values of all symbol functions at the same position and the sum of the Pearson correlation coefficient are used as the user's daily data change trend coefficient.
[0075] If the user's data change trend coefficient on day a is larger, it means that the user's account is more likely to be attacked by an attacker on day a, the possibility that the account is at risk is greater, and the possibility of abnormal data in the financial service application is greater.
[0076] S5, combining the operation sequence correlation coefficient and the data change trend coefficient, as well as the difference between the unauthorized application time set and the login failure time set, to obtain the user's daily data correlation change index.
[0077] Calculate the difference between the mean values of all elements in the unauthorized application time set and the login failure time set every day, record it as the third difference, and determine the data association change index every day by combining the third difference, the data change trend coefficient, and the operation sequence correlation coefficient every day;
[0078] The daily data association change index is positively correlated with the daily data change trend coefficient, the third difference, and the operation sequence correlation coefficient.
[0079] In one embodiment of the present application, the difference between the mean of all elements in the daily unauthorized application time set and the mean of all elements in the login failure time set, that is, the third difference, is calculated and recorded as the time difference, and the product of the daily data change trend coefficient, the time difference and the operation sequence correlation coefficient is used as the daily data correlation change index.
[0080] The larger the difference between the mean of all elements in the daily unauthorized application time set and the mean of all elements in the login failure time set is, the greater the possibility that the user's system is at risk and the data is likely to be abnormal. The greater the user's data association change index is, the greater the data change trend coefficient and the operation sequence correlation coefficient are, the greater the possibility of data abnormality is, and the greater the user's data association change index is.
[0081] S6, combining the data consistency score, the data association change index and the security log vector, using the isolation forest algorithm, to complete the online monitoring of abnormal data in financial service applications.
[0082] The security log vectors of the most recent N days in the user's security operation log downloaded from the security operation log module of the financial service application system are used as inputs to the isolation forest algorithm. In one embodiment of the present application, the necessary parameters in the isolation forest algorithm, that is, the number of samples drawn each time is manually set to 50, and the number of isolated trees in the isolation forest algorithm is manually set to 100, N=400. The number of samples, the number of isolated trees and the value of N can be set by the implementer according to actual conditions, and this application does not impose any restrictions on this.
[0083] In the original isolation forest algorithm, the weight of each isolated tree is the same, but according to the construction process of the isolation tree, if the sample data selected when constructing the isolation tree has a high similarity, then the value of this isolated tree in subsequent data anomaly detection is lower; if the depth of the left and right subtrees of an isolated tree is relatively close, then the time overhead corresponding to this isolated tree in data anomaly detection is large, which is also not conducive to the online monitoring of abnormal data. Therefore, in this application, the weight of the isolated tree in the isolation forest algorithm is adjusted in combination with the user's daily data consistency score and data association change index, specifically:
[0084] Calculate the discrete degree of the data consistency score of all security log vectors corresponding to the days contained in each isolated tree, calculate the average value of the data association change index of all security log vectors corresponding to the days contained in each isolated tree, and calculate the difference between the maximum and minimum depths of the left and right subtrees of each isolated tree, which is recorded as the fourth difference;
[0085] Determining a weight of each isolated tree based on the dispersion degree, the average value and the fourth difference;
[0086] The weight of each isolated tree is positively correlated with the dispersion degree, the average value and the fourth difference.
[0087] It should be noted that the degree of dispersion represents the distribution characteristics and changing trends of data distribution, which can be calculated using variance, standard deviation, etc.
[0088] In one embodiment of the present application, taking the i-th isolated tree as an example, the variance of the data consistency scores of all days corresponding to all security log vectors in the i-th isolated tree is calculated, and the variance is used as the exponent of an exponential function with a natural constant as the base, recorded as a first exponential function, and the mean of the data association change index of all days corresponding to all security log vectors in the i-th isolated tree is calculated, and the mean of the data association change index is used as the independent variable of the sign function, recorded as the first sign function. When the independent variable of the first sign function is less than or equal to 0, the function value of the first sign function is 0, otherwise, the function value of the first sign function is the value of the independent variable. The ratio of the maximum value to the minimum value of the depth of the left and right subtrees of the i-th isolated tree is calculated, that is, the fourth difference, and the product of the calculation result of the first exponential function, the function value of the first sign function, and the ratio of the maximum value to the minimum value is used as the weight of the i-th isolated tree.
[0089] If all samples of the ith isolated tree, that is, all security log vectors, have a high similarity, then the smaller the variance of the data consistency score of the ith isolated tree, the smaller the possibility that all security log vectors of the ith isolated tree are abnormal, and the smaller the weight of the ith isolated tree. The purpose of calculating the first exponential function is to amplify the variance of the data consistency score. When the variance of the data consistency score is larger, the weight of the ith isolated tree is larger. If the mean value of the data association change index of all days corresponding to all security log vectors in the ith isolated tree is larger, it means that all security log vectors in the ith isolated tree are risky, that is, the possibility of abnormal data is greater, and the weight of the ith isolated tree is larger. The first symbol function is calculated to ensure the non-negativity of the weight of the ith isolated tree. The larger the difference between the maximum and minimum values of the depths of the left and right subtrees of the ith isolated tree, the greater the difference, which means that the ith isolated tree can quickly distinguish whether the data is abnormal, and has a greater effect on online monitoring of abnormal data. Therefore, the larger the weight of the ith isolated tree, the higher the value.
[0090] The weight of each isolated tree in the isolation forest algorithm is calculated in the same way as the weight of the i-th isolated tree, so as to complete the training of the isolation forest. The flowchart of the weight construction of the isolated tree is as follows: Figure 2 shown.
[0091] The security log vector monitored daily is used as the input of the trained isolation forest algorithm, and the output is the abnormal score of the corresponding security log vector. When the abnormal score is less than the preset threshold, the corresponding security log vector is determined to be abnormal data, otherwise, the corresponding security log vector is determined to be normal data, wherein the isolation forest algorithm is an existing well-known technology, and this application will not be described in detail here. In one embodiment of the present application, the preset threshold is artificially set to 0.5, and the implementer can set it according to the actual situation, and this application does not limit it here.
[0092] Based on the same inventive concept as the above method, an embodiment of the present application also provides an online monitoring system for abnormal data of financial service applications, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above-mentioned methods for online monitoring of abnormal data of financial service applications are implemented.
[0093] In summary, this application collects the user's daily login failure frequency, unauthorized access application frequency and access module frequency in the financial service application system to form the user's daily security log vector, obtains the user's daily login failure time set, access time set and unauthorized application time set, analyzes the fluctuation and change trend of the user's login failure frequency, unauthorized access application frequency and access module frequency respectively, and obtains the user's daily data consistency score, which reflects the similarity of the user's login failure frequency, unauthorized access application frequency and access module frequency of all days before any day, and then judges the security of the user's financial service application data on any day, thereby improving the reliability of abnormal data monitoring of financial service applications; based on the difference between the elements in the user's daily login failure time set and the access time set, the user's daily operation sequence correlation coefficient is determined, which reflects the chronological relationship between the time points in the login failure time set and the access time set, and reflects the possibility of the user account being attacked by attackers. In the process of monitoring abnormal data of financial service applications, causal correlation features are added to improve the accuracy of abnormal data monitoring of financial service applications; the correlation between the frequency of user login failures and the frequency of unauthorized access applications is analyzed to obtain the user's daily data change trend coefficient, and the difference between the operation sequence correlation coefficient and the data change trend coefficient, as well as the difference between the unauthorized application time set and the login failure time set, to obtain the user's daily data association change index, which reflects the possibility of the user's financial service application system being attacked and abnormal data appearing every day. Combined with the data consistency score, the data association change index and the security log vector, the isolation forest algorithm is used to complete the online monitoring of abnormal data of financial service applications. The value and timeliness of abnormal monitoring of security log vectors are reflected in the isolation forest algorithm, which ensures the accuracy of abnormal data monitoring while also improving the sensitivity of abnormal monitoring, effectively responding to risk challenges in financial service application systems.
[0094] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0095] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0096] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present application should be included in the protection scope of the present application.
Claims
1. A method for online monitoring of abnormal data in financial service applications, characterized in that: The method comprises the following steps: Collect the user's daily login failure frequency, unauthorized access application frequency and access module frequency in the financial service application system to form the user's daily security log vector; obtain the user's daily login failure time set, access time set and unauthorized application time set; Analyze the fluctuation and change trend of the user's login failure frequency, unauthorized access application frequency, and access module frequency respectively, and obtain the user's daily data consistency score; Based on the difference between the elements in the user's daily login failure time set and the access time set, determine the user's daily operation sequence correlation coefficient; Analyze the correlation between the frequency of user login failures and the frequency of unauthorized access applications, and obtain the user's daily data change trend coefficient; Combining the operation sequence correlation coefficient with the data change trend coefficient and the difference between the unauthorized application time set and the login failure time set, the user's daily data correlation change index is obtained; Combining the data consistency score, the data association change index and the security log vector, using an isolation forest algorithm, online monitoring of abnormal data for financial services applications; The login failure frequencies of the user on any day and all days before the said day are combined into a login failure sequence for the said day; The determination process of the operation sequence correlation coefficient is as follows: If all elements in the daily access time set are less than or equal to the minimum value in the daily login failure time set, the daily operation sequence correlation coefficient is the first preset value, otherwise, the daily operation sequence correlation coefficient is the second preset value, wherein the first preset value is less than the second preset value; The process of determining the data change trend coefficient includes: Calculate the correlation between the daily login failure sequence and the unauthorized access application sequence, calculate the growth rate of each element in the daily login failure sequence relative to the previous element, record it as the first growth rate, calculate the growth rate of each element in the daily unauthorized access application sequence relative to the previous element, record it as the second growth rate, perform forward fusion on the difference between the first growth rate and the second growth rate of all the same positions, and the correlation, and use the forward fusion result as the daily data change trend coefficient; The determination of the data consistency score includes: The login failure sequence is subjected to a moving average method to obtain each moving average to form a moving average sequence of the login failure sequence, and based on the difference between adjacent elements in the moving average sequence of the login failure sequence, a comprehensive fluctuation coefficient of the frequency of login failures on any day is determined; The same calculation method as the data comprehensive fluctuation coefficient of the frequency of failed logins on any day is used to obtain the data comprehensive fluctuation coefficient of the frequency of unauthorized access applications and the frequency of access modules on any day; Calculate the difference between the data comprehensive fluctuation coefficient of the access module frequency and the data comprehensive fluctuation coefficient of the unauthorized access application frequency, record it as the first difference, calculate the difference between the unauthorized access application frequency of any day and the average of the unauthorized access application frequencies of all days before any day, record it as the second difference; Determine the data consistency score of any day by comprehensively analyzing the data comprehensive fluctuation coefficient of the login failure frequency, the first difference, and the second difference; The data consistency score of any day is negatively correlated with the data comprehensive fluctuation coefficient of the login failure frequency and the second difference, and is positively correlated with the first difference; The determination process of the comprehensive fluctuation coefficient of the data of the login failure frequency is as follows: Calculate the difference between each element and the adjacent element in the moving average sequence of the login failure sequence, record it as the adjacent difference, perform forward fusion on all the adjacent differences in the moving average sequence of the login failure sequence, and use the forward fusion result as the data comprehensive fluctuation coefficient of the login failure frequency; The process of determining the data association change index includes: Calculate the difference between the mean values of all elements in the unauthorized application time set and the login failure time set every day, record it as the third difference, and determine the data association change index every day by combining the third difference, the data change trend coefficient and the operation sequence correlation coefficient every day; The daily data association change index is positively correlated with the daily data change trend coefficient, the third difference, and the operation sequence correlation coefficient.
2. The method for online monitoring of abnormal data in financial service applications according to claim 1, characterized in that: The module access frequency is the sum of the number of times a user accesses a module for which he / she has permission and the frequency of unauthorized access applications.
3. The method for online monitoring of abnormal data in financial service applications according to claim 1, characterized in that: The online monitoring of abnormal data of financial services applications includes: The daily security log vector is used as the input of the isolation forest algorithm. The weight of each isolated tree is determined based on the data consistency score and data association change index distribution of the security log vector corresponding to the number of days contained in each isolated tree in the isolation forest algorithm, as well as the depth of the left and right subtrees of each isolated tree. The output of the isolation forest algorithm is the anomaly score of the security log vector of each day. When the anomaly score is less than a preset threshold, the corresponding security log vector is determined to be abnormal data; otherwise, the corresponding security log vector is determined to be normal data.
4. A method for online monitoring of abnormal data in financial service applications as claimed in claim 3, characterized in that: The process of determining the weight of each isolated tree is as follows: Calculate the discrete degree of the data consistency score of all security log vectors corresponding to the days contained in each isolated tree, calculate the average value of the data association change index of all security log vectors corresponding to the days contained in each isolated tree, and calculate the difference between the maximum and minimum depths of the left and right subtrees of each isolated tree, which is recorded as the fourth difference; Determining a weight of each isolated tree based on the dispersion degree, the average value and the fourth difference; The weight of each isolated tree is positively correlated with the dispersion degree, the average value and the fourth difference.
5. A system for online monitoring of abnormal data in financial services applications, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.
Citation Information
Patent Citations
User abnormal access behavior detection method based on isolated forest
CN114090402A
Abnormal login behavior management method and device, storage medium and electronic equipment
CN117692216A
Account monitoring and early warning management method based on data analysis
CN117852003A
Abnormal login detection method and system based on isolated forest
CN117978461A
Detecting behavior anomalies of cloud users for outlier actions
US20200336503A1
