Financial system entrance security detection method based on browser technology
Through the financial system entry security detection method based on browser technology, multiple verification is used to utilize user's historical typing actions and machine learning, the problem of insufficient security of traditional identity verification is solved, and higher user identity confirmation and financial system security is achieved. .
Patent Information
- Application Number
- CN202410915122.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-11-01
AI Technical Summary
Traditional financial system identity verification methods rely on accounts and passwords, and are susceptible to security threats caused by leakage and theft. Mobile phone SMS verification has security risks, making it difficult to effectively ensure the security of the financial system.
Adopt the financial system entry security detection method based on browser technology, and capture the user's historical typing actions, generate personal typing habit models, combine machine learning for real-time verification, and use typing comprehensive scores and image recognition verification to perform multiple verification to confirm user identity. .
It enhances the user's identity confirmation and the security of the financial system, prevents illegal login, and even if the attacker knows the account and password, it is difficult to pass additional verification, improving the security of the system.
Smart Images

Figure CN118885993A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data security technology, and in particular to a financial system entrance security detection method based on browser technology. Background Art
[0002] As the financial system's requirements for network security are increasingly strengthened, user identity verification technology has become a hot topic in the security field. Although traditional identity verification technologies, such as account and password verification, can verify user identities to a certain extent, many users now use the same account and password for multiple platform accounts for the convenience of memory. In this case, when the data of one platform is leaked, the user's password on other platforms will also face security risks. Therefore, it is difficult to ensure the authenticity of user identities and the security of financial systems by relying solely on account and password for user verification.
[0003] The existing technology uses mobile phone text messages to assist in login verification. However, mobile phone verification still has certain security risks and limitations. When the user's mobile phone is stolen or the text message data is intercepted or stolen due to the low security performance of the mobile phone, the security of the financial system is difficult to guarantee. Summary of the invention
[0004] In order to solve the above problems, the present invention provides a financial system entrance security detection method based on browser technology.
[0005] To achieve the above object, the technical solution adopted by the present invention is:
[0006] A financial system entrance security detection method based on browser technology, comprising:
[0007] Based on browser technology, the user's historical typing actions when logging into the financial system entrance are captured to collect the user's historical verification data; the historical typing actions include historical account typing actions and historical password typing actions;
[0008] Capture features based on the user's historical typing actions and perform machine learning to output a personal typing habit model;
[0009] Verify the account and password input in real time, perform weighted analysis on the real-time typing action according to the personal typing habit model, generate a comprehensive typing score, generate a personalized verification based on the comprehensive typing score, and collect the personalized verification results; the personalized verification includes the user's historical verification data and randomly generated verification data;
[0010] Log in and pass according to the individual verification result;
[0011] After logging in, the randomly generated verification data is recorded in the user's historical verification data.
[0012] Furthermore, the input action includes the time difference of inputting each character and the total input time.
[0013] Furthermore, the verification data includes an image recognition verification package.
[0014] Furthermore, the feature capture and machine learning based on the user's historical typing actions includes:
[0015] Create time series data based on the time difference of each character input and the total input time;
[0016] Taking time series data as input, training is performed based on the long short-term memory network to generate a personal typing habit model.
[0017] Further, generating the typing comprehensive score includes:
[0018] Scoring is performed based on the degree of match between the user's real-time typing action and the personal typing habit model, and outputting a real-time account typing score and a real-time password typing score;
[0019] The real-time account entry score and the real-time password entry score are weighted and synthesized to generate a comprehensive entry score.
[0020] Further, generating a personality verification based on the input comprehensive score includes:
[0021] When the input comprehensive score is greater than or equal to the preset value, the randomly generated verification data is used for verification;
[0022] When the comprehensive score entered is less than the preset value, the user's historical verification data is used for verification.
[0023] Further, the verification using randomly generated verification data includes:
[0024] When the verification input is correct, it is determined to be a normal user;
[0025] When the verification input is incorrect, re-verify using randomly generated verification data.
[0026] Further, the verification using the user's historical verification data includes:
[0027] Use image recognition verification packages used by several users in past logins for verification, and record the real-time verification time of each image recognition verification package;
[0028] Subtract the real-time verification time from the historical verification time of the corresponding image recognition verification package, calculate the time difference of each image recognition verification package and add them up to obtain the total difference;
[0029] When the total difference is positive or zero, it is determined to be an abnormal user;
[0030] When the total difference is negative, the user is determined to be a normal user.
[0031] Furthermore, the passing according to the personality verification result includes:
[0032] When the result of personality verification is a normal user, log in and pass;
[0033] When the result of personality verification is that the user is not a normal user, login is blocked.
[0034] Further, recording the randomly generated verification data into the user's historical verification data includes:
[0035] The image recognition verification package and verification time of the randomly generated verification data are recorded and saved in the user's historical verification data.
[0036] The beneficial effects of the present invention are as follows: the present invention firstly captures the historical typing actions of the user when logging into the financial system entrance through the browser technology, which includes the historical data when the user types the account number and password. These data can provide a unique behavioral feature for each user, because the time interval of the input characters when each user enters the password will be different due to personal habits. The present invention performs feature capture based on the historical typing actions of these users and analyzes them through machine learning, thereby outputting a personal typing habit model for each user. When the user attempts to log in to the financial system, his typing action will be captured again and compared with the previously generated personal typing habit model. This comparison will generate a comprehensive typing score for each login attempt, which can reflect the degree of match between the current typing action and the historical data habit, ensuring that even if the attacker knows the user's account number and password, it is difficult to pass this additional layer of verification because their typing habits are different from those of the real user.
[0037] The user's identity is further confirmed through personal verification. The user's historical verification data is used for verification through the login (i.e., the matching degree is low and it is suspected that the user is not the user himself). The user has a certain impression of the image recognition authentication that has been done before, so the verification time will be equal to or lower than the historical verification time. The verification time of multiple historical image recognition verification packages is used to judge whether it is the user himself to achieve deep authentication. Then, the login with a high comprehensive score is verified using randomly generated verification data, and the image recognition verification package and verification time are recorded after the successful login, so as to increase the data source for future verification. The present invention uses the user's typing habits as an additional identity authentication method in addition to the account password, and further verifies through personal verification, thereby increasing the degree of confirmation of the user's identity and the security of the entire financial system, and achieving that even if a non-user knows the user's account password and can obtain the SMS verification code, it is difficult for him to enter the financial system and pose a threat to financial security. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a flow chart of the steps of a financial system entrance security detection method based on browser technology in the present invention.
[0039] Figure 2 It is a step flow chart of step S2 in the present invention.
[0040] Figure 3 It is a step flow chart of step S32 in the present invention. DETAILED DESCRIPTION
[0041] See also Figure 1-3 As shown, the present invention relates to a financial system entrance security detection method based on browser technology.
[0042] Specifically, the present invention provides a financial system entrance security detection method based on browser technology, comprising the following steps:
[0043] S1. Capturing the historical typing actions of the user when logging into the financial system entrance based on browser technology, and collecting the historical verification data of the user; the historical typing actions include historical account typing actions and historical password typing actions;
[0044] The typing action includes the time difference of inputting each character and the total input time, and the verification data includes an image recognition verification package.
[0045] Specifically, when a user logs in to the financial system, a capture program is started based on browser technology to monitor and record every typing action of the user. This data capture is not limited to the characters themselves, but also detailed to the time difference between typing each character, and the total time to complete the entire account or password input. This detailed typing data can reflect their personalized typing habits. For example, when a user enters the financial system account "fi nance123", there may be a 0.3 second interval between "f" and "i", and only a 0.1 second interval between "1" and "2", and the entire account input may take 2.5 seconds to complete. These specific and tiny time differences provide each user with an almost unique identity, like a fingerprint. At the same time, the technology also tracks historical verification data. In addition to the typing actions of the account and password, the historical verification data also includes the data of the image recognition verification package.
[0046] When it comes to the image recognition verification package, the system will show the user a set of images and ask the user to complete a specific recognition task. But unlike other image verification methods, here we do not focus on the order of user selection, but on the total time it takes for the user to complete the verification. For example, the user may need to select three pictures that match the description from five pictures. Some users may complete it quickly within 3 seconds, while other users may need 5 seconds. This difference in verification time further strengthens the verification process of the user's identity.
[0047] S2, based on the user's historical typing actions, feature capture and machine learning are performed to output a personal typing habit model; Figure 2 , step S2 comprises:
[0048] S21, creating time series data according to the time difference of each character input and the total input time;
[0049] Specifically, the system first captures the input time difference between each character and the total input time of the entire string for each typing action of the user. For example, consider the account number entered by the user as "fi nance321", and the total time from entering the first character "f" to the last character "1" may be 2.8 seconds. In this 2.8-second period, the time difference between each two consecutive characters, such as the time difference from "f" to "i", or the time difference from "c" to "e", is accurately captured.
[0050] Next, the system uses the captured time difference data to construct the structure of the time series data. Time series data is used in many fields, such as finance, weather forecasting, and stock market analysis. It can capture the inherent patterns and trends of data over time. Here, the time series data records the specific moment when the user enters each character, thereby reflecting the rhythm and speed of the user's typing. For example, for the account "fi nance321", the time series data may be structured like this: [0 seconds, 0.3 seconds, 0.6 seconds, 0.9 seconds, 1.2 seconds, 1.6 seconds, 2 seconds, 2.3 seconds, 2.6 seconds, 2.8 seconds], which represents the time point when the user enters each character at each moment.
[0051] S22, using time series data as input, training based on long short-term memory network to generate a personal typing habit model;
[0052] Specifically, the long short-term memory network (LSTM) is selected as the training model because of its excellent performance in processing sequence data. LSTM is a special recurrent neural network (RNN) architecture that performs particularly well in learning long-term dependent data. Traditional RNNs have difficulty processing long sequence data due to the problem of gradient disappearance or gradient explosion. LSTM, through its unique gating mechanism, can effectively solve these problems, so that it can capture and learn deep patterns in time series data. In step S22, the time series data is first fed into the LSTM network. Considering that the data is the time points when the user continuously enters characters, LSTM can use its internal structure to capture the intrinsic patterns in these data, such as the user's input rhythm, speed changes, etc. After multiple iterations and training, the LSTM network generates a specific typing habit model for each user. This model is actually a collection of weights and parameters of the LSTM network, which can describe and reflect the user's unique habits when typing. A simplified example: Suppose Alice is typing a password. Her speed is relatively slow at the beginning, then speeds up in the middle, and finally slows down again. After training, the LSTM network will recognize this pattern and generate a corresponding typing habit model for Alice. When Alice tries to log in next time, her typing data will be compared with the model to verify her identity.
[0053] S3, verifying the account and password input in real time, performing weighted analysis on the real-time typing action according to the personal typing habit model, generating a comprehensive typing score, generating a personalized verification according to the comprehensive typing score, and collecting the personalized verification results; the personalized verification includes the user's historical verification data and randomly generated verification data;
[0054] S31, verifying the account and password input in real time;
[0055] Specifically, when the user starts to enter the account and password, the system will compare it with the data in the background database. The database stores the user's account information and the corresponding encrypted password. The system will perform the same encryption process on the password entered by the user and compare the two encryption results. If they match, the account and password verification is successful; if they do not match, the verification fails.
[0056] S32, performing weighted analysis on the real-time typing action according to the personal typing habit model to generate a comprehensive typing score; referring to Figure 3 , step S32 comprises:
[0057] S321, scoring the user's real-time typing action according to the matching degree with the personal typing habit model, and outputting the real-time account typing score and the real-time password typing score;
[0058] Specifically, after the account and password are verified, first, the system will record the time difference between each character of the account and password and the total input time of the account. The data collected in this way forms a time series that can reflect the user's typing rhythm and habits. Next, the time series data is input into the previously trained personal typing habit model. The model calculates the degree of match between the real-time typing data and the data predicted by the model. The higher the match, the more similar the user's typing habits are to the past, so the higher the score; conversely, if the match is low, it means that the user's typing habits this time are different from the past, and the score will be reduced accordingly.
[0059] For example, if in historical data, the average time for user Tom to enter "a" and then "b" is 0.5 seconds, and in real-time login process, Tom enters "a" and then "b" 0.52 seconds later, then the match between these two actions will be very high, thus obtaining a high score. On the contrary, if Tom enters "a" and then "b" 1.5 seconds later in real-time login, this is obviously different from his historical habits, thus the score will be lower.
[0060] S322, weighting and synthesizing the real-time account input score and the real-time password input score to generate a comprehensive input score;
[0061] Specifically, in step S322, the system aims to generate a comprehensive score, which is a "typing comprehensive score", based on the scores of the user's real-time account and password. This typing comprehensive score can provide a quantitative indicator for the system, reflecting the similarity between the current login attempt and the user's historical typing habits. In order to ensure the accuracy and fairness of the comprehensive score, different weights are usually given to the account and password, which is mainly based on two considerations: security, passwords are usually regarded as more confidential and important information than accounts, so verifying their typing habits may be more critical than accounts; differences, because the complexity of passwords is usually higher than that of accounts, the personalized characteristics shown by users when typing passwords may be more obvious. First, weights are determined for the real-time account typing score and the real-time password typing score. For example, a weight of 0.7 can be given to the password typing score, while a weight of 0.3 can be given to the account typing score. This means that when calculating the typing comprehensive score, the password will have a greater impact than the account. Explanation: The real-time account and password typing scores are weighted averaged using the determined weights. Finally, the system will output this comprehensive score, which will be used in the subsequent personality verification step to determine whether to verify the user's login request together with other verification data.
[0062] S33, generating personality verification according to the input comprehensive score, and collecting personality verification results;
[0063] Among them, the specific steps of generating personality verification based on the comprehensive score entered are:
[0064] When the input comprehensive score is greater than or equal to the preset value, the randomly generated verification data is used for verification;
[0065] When the comprehensive score entered is less than the preset value, the user's historical verification data is used for verification.
[0066] The verification using randomly generated verification data includes:
[0067] When the verification input is correct, it is determined to be a normal user;
[0068] When the verification input is incorrect, re-verify using randomly generated verification data.
[0069] Specifically, in step S33, the user's typing comprehensive score is first read and compared with a preset value. This preset value is set based on system experience and security requirements, and is used to distinguish between the normality and abnormality of the user's typing behavior. If the user's typing comprehensive score is greater than or equal to this preset value, it is considered that the user's typing behavior is highly matched with its historical data, so this login attempt is more trusted. At this time, the system will generate a random verification data as a verification method. For example, the system may provide the user with a randomly selected image and ask the user to select a description that matches the image from several options. The purpose of this verification method is to prevent automated attacks to a certain extent and provide additional security.
[0070] If the comprehensive typing score is lower than the preset value, it means that the current login attempt is significantly different from the user's historical habits, and it may be because the user is not the one trying to log in. At this time, the system will use the user's historical verification data for verification. The system expects that users will have an impression of the image recognition authentication they have done before, so the time they take to complete the verification should be equal to or lower than the historical verification time. This method takes advantage of the user's memory of the image verification they have experienced before, ensuring that even if the attacker knows the account password, it is difficult to pass the verification smoothly due to the lack of familiarity with the historical verification data.
[0071] Specifically, use the image recognition verification packages used by several users in past logins for verification, and record the real-time verification time of each image recognition verification package respectively;
[0072] Subtract the real-time verification time from the historical verification time of the corresponding image recognition verification package, calculate the time difference of each image recognition verification package and add them up to obtain the total difference;
[0073] When the total difference is positive or zero, it is determined to be an abnormal user;
[0074] When the total difference is negative, the user is determined to be a normal user.
[0075] Specifically, several image recognition verification packages that the user has used in past login attempts are selected. These contain the user's previous verification history information, and the user has a certain understanding and memory of them. The user needs to complete the verification task of each selected image recognition verification package. The system will carefully record the real-time time required for the user to complete each verification package. For each image recognition verification package, the system will calculate the difference between the real-time verification time and the time taken by the verification package in the historical verification. Each difference obtained will be recorded and accumulated to obtain a total difference. If the calculated total difference is a positive value or zero, the system will determine that the current login attempt is an abnormal user. This means that the user spent more time overall than the historical verification when completing these verification tasks, which may be due to the user's lack of familiarity with these verification packages. If the calculated total difference is a negative value, the system will determine that the current login attempt is a normal user. This means that the user spent less time overall than the historical verification when completing the verification task, indicating that the user has a certain degree of familiarity and memory of the verification package. Through such a verification mechanism, the system can more accurately identify and confirm the identity of the user, ensuring that only genuine users can successfully log in. This not only makes use of the user's historical verification data, but also fully considers the user's familiarity and memory of the verification package, further improving the security of the system.
[0076] S4. Log in and pass according to the individual verification result;
[0077] When the result of personality verification is a normal user, log in and pass;
[0078] When the result of personality verification is that the user is not a normal user, login is blocked.
[0079] Specifically, if the result of the personality verification is displayed as "normal user", it means that all previous verifications have been passed, the user's login habits match the historical data, and his performance during the image recognition verification is also consistent with the historical records. In this case, the system allows the user to continue and successfully complete the login process, thereby obtaining full access to the financial system. If the result of the personality verification is displayed as "abnormal user", it means that there is a mismatch or abnormal behavior in a previous verification step. Possible reasons are: the typing habits do not match the historical data, or the user's performance does not match the historical records when completing the image recognition verification. In this case, for the security of the system, the system will immediately stop the login process and give an error message or prompt to tell the user that his login request has been rejected. This method not only relies on traditional username and password verification, but also adds a personalized verification mechanism to ensure that even if the attacker knows the user's credentials, it is difficult to pass the system's multiple verification defense lines. This provides a higher level of security and protection for the financial system.
[0080] S5. After logging in, the randomly generated verification data is recorded in the user's historical verification data;
[0081] Specifically, the image recognition verification package and verification time of the randomly generated verification data can be recorded and saved in the user's historical verification data. After the user successfully passes the login verification of the financial system, the system enters the data post-processing stage. This stage is crucial to ensure that the user's historical verification data is continuously updated and enhanced, because this will further improve the accuracy of verification and the overall security of the system. The system first captures the randomly generated verification data used in this login process. These verification data are mainly a series of image recognition verification packages, which are designed to test and confirm the identity of the user. For example, a randomly generated verification data may require the user to identify a series of pictures containing specific objects, such as "select all pictures containing red cars". Then, the system measures and records the time required for the user to complete the image recognition verification. Considering that each user may have their own unique processing verification time for similar verification tasks, this time information is of great significance for future personalized verification. Finally, these newly captured verification data, including the image recognition verification package and its corresponding verification time, will be integrated and saved into the user's historical verification data. In this way, when the user tries to log in again, these new data provide a richer and more comprehensive context to help the system more accurately determine the identity of the login user.
[0082] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.
Claims
1. A financial system entrance security detection method based on browser technology, characterized in that: The following steps are involved: Based on browser technology, the user's historical typing actions when logging into the financial system entrance are captured to collect the user's historical verification data; the historical typing actions include historical account typing actions and historical password typing actions; Capture features based on the user's historical typing actions and perform machine learning to output a personal typing habit model; Verify the account and password input in real time, perform weighted analysis on the real-time typing action according to the personal typing habit model, generate a comprehensive typing score, generate a personalized verification based on the comprehensive typing score, and collect the personalized verification results; the personalized verification includes the user's historical verification data and randomly generated verification data; Log in and pass according to the individual verification result; After logging in, the randomly generated verification data is recorded in the user's historical verification data.
2. According to claim 1, a financial system entrance security detection method based on browser technology is characterized in that: The input action includes the time difference of each character input and the total input time.
3. According to claim 1, a financial system entrance security detection method based on browser technology is characterized in that: The verification data includes an image recognition verification package.
4. According to claim 2, a financial system entrance security detection method based on browser technology is characterized in that: The feature capture and machine learning based on the user's historical typing actions include: Create time series data based on the time difference of each character input and the total input time; Taking time series data as input, training is performed based on the long short-term memory network to generate a personal typing habit model.
5. A financial system entrance security detection method based on browser technology according to claim 4, characterized in that: Generating a comprehensive typing score includes: Scoring is performed based on the degree of match between the user's real-time typing action and the personal typing habit model, and outputting a real-time account typing score and a real-time password typing score; The real-time account entry score and the real-time password entry score are weighted and synthesized to generate a comprehensive entry score.
6. A financial system entrance security detection method based on browser technology according to claim 1, characterized in that: The generating of personality verification according to the input comprehensive score includes: When the input comprehensive score is greater than or equal to the preset value, the randomly generated verification data is used for verification; When the comprehensive score entered is less than the preset value, the user's historical verification data is used for verification.
7. A financial system entrance security detection method based on browser technology according to claim 6, characterized in that: The verification using randomly generated verification data includes: When the verification input is correct, it is determined to be a normal user; When the verification input is incorrect, re-verify using randomly generated verification data.
8. A financial system entrance security detection method based on browser technology according to claim 7, characterized in that: The verification using the user's historical verification data includes: Use image recognition verification packages used by several users in past logins for verification, and record the real-time verification time of each image recognition verification package; Subtract the real-time verification time from the historical verification time of the corresponding image recognition verification package, calculate the time difference of each image recognition verification package and add them together to obtain the total difference; When the total difference is positive or zero, it is determined to be an abnormal user; When the total difference is negative, the user is determined to be a normal user.
9. A financial system entrance security detection method based on browser technology according to claim 8, characterized in that: The passing according to the personality verification result includes: When the result of personality verification is a normal user, log in and pass; When the result of personality verification is that the user is not a normal user, login is blocked.
10. According to the financial system entrance security detection method based on browser technology in claim 9, recording the randomly generated verification data into the user's historical verification data comprises: The image recognition verification package and verification time of the randomly generated verification data are recorded and saved in the user's historical verification data.
