Verification code management method based on multimodality and reinforcement learning and computer system

Abstract

The invention relates to the field of communication verification management, in particular to a verification code management method based on multi-mode and reinforcement learning and a computer system.The method comprises the steps that after verification information sent by a third-party platform is responded, multi-mode features of the verification information are extracted firstly, then a preset semantic understanding engine is called, and a verification code is obtained; fusing and analyzing the multi-modal features to obtain a verification analysis result, and finally, executing a verification behavior according to the verification analysis result; or the verification analysis result is sent to the user side. Compared with the prior art, the method provided by the invention has the advantages that the automation of verification behaviors and verification forwarding is realized, and the verification efficiency and safety of enterprise-level verification management are improved.

Description

Technical Field

[0001] The present invention relates to the field of communication verification management. More specifically, the present invention relates to a verification code management method and computer system based on multimodal and reinforcement learning. Background Art

[0002] A CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) is an interactive security mechanism that distinguishes whether a user is a machine or a human. Widely used for website access verification, CAPTCHAs not only prevent malicious activities such as password cracking, ticket fraud, and forum flooding, but also effectively prevent hackers from using specific programs to brute-force login attempts against a specific registered user. Currently, to facilitate employee communication (customer service, e-commerce, or sales positions), companies register a large number of mobile phone numbers for employees in corresponding positions as shared mobile phone numbers. Employees can use these company-assigned numbers to log in to third-party platforms (such as Taobao, JD.com, or TikTok) through verification.

[0003] However, the following flaws exist when employees perform the above operations to obtain verification codes: 1. Inefficient verification management: The existing enterprise verification forwarding process is that employees first submit verification applications, the third-party platform sends the verification code to the enterprise docking system, and then manually forwards it to the corresponding employee. This method will lead to delayed verification responses.

[0004] 2. High security risks in verification management: Traditional code receiving platforms and methods cannot meet enterprise-level privacy protection requirements. During manual forwarding and verification, there is a high risk of verification code leakage. There is also a high risk of malicious repeated verification or incorrect input leading to the blacklisting of mobile phone numbers.

[0005] Therefore, existing enterprise-level verification management technology has technical problems such as low verification efficiency and high security risks. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems of low verification efficiency and high security risks, the present invention discloses a verification code management method and computer system based on multimodal and reinforcement learning.

[0007] In a first aspect, the present invention discloses a verification code management method based on multimodal and reinforcement learning, comprising: In response to verification information sent by the third-party platform, extract multimodal features of the verification information; Call the preset semantic understanding engine to fuse and analyze multimodal features to obtain verification and analysis results; Execute the verification behavior according to the verification and analysis result; or send the verification and analysis result to the user end.

[0008] Beneficial effect: After receiving the verification information sent by the third-party platform, the method of the present invention first extracts the multimodal features of the verification information, and then calls the semantic understanding engine to fuse and analyze the multimodal features to obtain the verification analysis results, and automatically executes the verification behavior or automatically forwards and distributes the verification code. Compared with the existing technology, the method of the present invention realizes the automation of verification behavior and the automation of verification forwarding, and improves the verification efficiency of enterprise-level verification management. The user end (employee) cannot know the initial verification information throughout the whole process. The entire verification process belongs to the verification processing process of the internal enterprise system, and no manual forwarding or input is required. This verification method solves the problem of high security risks in enterprise verification management.

[0009] Preferably, the multimodal features are fused and analyzed to obtain verification and analysis results, including: Configuring modal weights corresponding to the multimodal features; Call the extraction function corresponding to the multimodal feature; The modal weight and extraction function are placed into the preset multimodal fusion function to calculate the recognition confidence; The recognition confidence and multimodal features are placed into the semantic understanding engine based on the Transformer architecture, and the parsing is performed to obtain the verified parsing results.

[0010] Beneficial Effects: This method utilizes a Transformer-based semantic understanding engine to analyze multimodal features, which can improve the recognition rate of verification codes. Compared to existing technologies, this method can rely on the semantic understanding engine to dynamically adjust weights and support the parsing of logical expressions.

[0011] Furthermore, the multimodal fusion function is specifically:

[0012] Where, Indicates a given under conditions The recognition confidence of Indicates the The feature extraction function of each modality, express function, represents the bias term, Indicates the The weight of each mode.

[0013] Furthermore, the modal weight includes at least text weight, image weight and speech weight; wherein the image weight is greater than the speech weight and less than the text weight, and the sum of the text weight, image weight and speech weight is 1.

[0014] Beneficial effect: The method of the present invention can adjust the verification weight of the initial verification information from multiple dimensions such as text, image and voice, thereby improving the success rate of verification code parsing.

[0015] Preferably, before extracting the multimodal features of the verification information in response to the verification information sent by the third-party platform, the method of the present invention further includes: In response to the user's verification requirements, according to the preset number resource scheduling model, a mobile phone number for information verification is scheduled from the preset number pool to the user; Submit a verification request to a third-party platform in the name of a mobile phone number.

[0016] Compared with the existing technology, the method of the present invention can automatically assign a mobile phone number to the user terminal in response to the user terminal's verification requirements, without the user terminal's own mobile phone number for verification and login. On the one hand, it realizes the separation of public and private mobile phone numbers of the user terminal, reducing the troubles caused by corporate operations to personal information; on the other hand, it is conducive to corporate information management.

[0017] The number resource scheduling model is as follows:

[0018] Where, Indicates the actual allocation parameters of mobile phone numbers. Indicates Find the minimum value in the range and output the corresponding A function of the value, Indicates the The current task volume of mobile phone numbers, Indicates the The reputation score of each mobile number, represents the smoothing coefficient, Indicates the total number of mobile phone numbers. Indicates the allocation parameters of the current mobile phone number. Indicates the allocation parameters of the current mobile phone number. It means taking the square of the 2-norm.

[0019] Compared with the existing technology, the number resource scheduling scheme of the method of the present invention can flexibly allocate available mobile phone numbers to the user terminal according to the current task volume, credit score and the total number of numbers in the number pool of the mobile phone number, thereby improving verification efficiency.

[0020] Preferably, after performing a verification action according to the verification and analysis result; or sending the verification and analysis result to the user terminal, the method of the present invention further includes: Use the preset signal strength monitoring module to evaluate the communication strength of the mobile phone number under the current operator in real time; If the communication strength is lower than the communication threshold, the current operator of the mobile phone number is switched to the backup operator.

[0021] Preferably, before extracting the multimodal features of the verification information in response to the verification information sent by the third-party platform, the method of the present invention further includes: Use the preset fault prediction model to predict the failure probability of the corresponding SIM card in the preset number pool; If the failure probability of the corresponding SIM card is greater than the failure threshold, a self-recovery mechanism of the system or device to which the corresponding SIM card belongs is triggered.

[0022] Preferably, after sending the verification and analysis result to the user terminal, the method of the present invention further includes: Calculate the abnormality of the user's operation based on the user's operation records; The degree of operational abnormality serves as an indicator parameter for number pool security audit management.

[0023] In a second aspect, the present invention further discloses a computer system comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the verification code management method based on multimodal and reinforcement learning recorded in the first aspect is implemented.

[0024] The beneficial effects of the present invention are: (1) Compared with the existing technology, the method of the present invention realizes the automation of verification behavior and verification forwarding, and improves the verification efficiency of enterprise-level verification management. The user end (employee) cannot know the initial verification information throughout the whole process. The entire verification process belongs to the verification processing process of the enterprise internal system, and no manual forwarding or input is required. This verification method solves the problem of high security risks in enterprise verification management.

[0025] (2) Compared with the existing technology, the method of the present invention can adjust the verification weight of the initial verification information from multiple dimensions such as text, image and voice, thereby improving the success rate of verification code parsing.

[0026] (3) Compared with the existing technology, the method of the present invention can automatically assign a mobile phone number to the user terminal in response to the user terminal's verification requirements, without the user terminal's own mobile phone number for verification and login. On the one hand, it realizes the separation of public and private mobile phone numbers of the user terminal, reducing the troubles caused by corporate operations to personal information; on the other hand, it is conducive to corporate information management. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flowchart of a verification code management method based on multimodality and reinforcement learning in Example 1 of the present invention; Figure 2 is a schematic diagram of the structure of a computer system in Embodiment 2 of the present invention; Figure 3 This is a framework diagram of the verification code management system based on multimodality and reinforcement learning in Example 3 of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The specific implementation methods of the present invention are described in detail below with reference to the accompanying drawings.

[0029] Example 1 like Figure 1 As shown, this embodiment discloses a verification code management method based on multimodality and reinforcement learning, which is suitable for mobile phone number management and automatic verification code processing in enterprise multi-account registration scenarios. The method includes: S10: In response to the verification information sent by the third-party platform, extract multimodal features of the verification information.

[0030] In this embodiment, the third-party platform may refer to domestic third-party platforms such as Taobao, JD.com, Douyin or Xiaohongshu that are required for daily operations of the enterprise. In the case that the enterprise has an overseas registered place, the third-party platform can also be an overseas third-party platform, such as youtube, X, Google or Amazon. After the user terminal submits a verification request in the enterprise's internal system, the enterprise system dispatches the mobile phone number and submits the verification to the third-party platform. Multimodal features include text, image and voice features. Text features involve text logical formulas such as "1+3=?", "According to the legend, please enter the second and fourth digits" or "Find the bicycle from the picture", etc. Image features involve images automatically generated by the third-party platform after receiving the verification request, such as "nine-square grid", "six-sided view" or "color difference map", etc. Voice features involve language or semantics. Multimodal features can be extracted using CNN convolutional neural network.

[0031] S20: Calling the preset semantic understanding engine to fuse and analyze the multimodal features to obtain verification and analysis results.

[0032] In this embodiment, the semantic understanding engine adopts a semantic understanding engine based on the Transformer architecture to obtain more accurate verification and parsing results.

[0033] S30: Execute verification behavior according to the verification and analysis result; or send the verification and analysis result to the user end.

[0034] In this embodiment, the user terminal generally refers to the communication device (such as a computer, mobile phone, or iPad) provided by the enterprise to employees. These user terminals are pre-installed with a script that automatically enters verification text or clicks a verification icon upon receiving the verification analysis result, thereby automatically executing the verification behavior. In some special scenarios, the verification analysis result can be directly sent to the user terminal's specific information receiving interface without the script being configured.

[0035] Through steps S10 to S30, the method of the present invention automates both verification and forwarding, improving the efficiency of enterprise-level verification management. The intermediate verification process is kept confidential within the enterprise system, and employees cannot maliciously enter the wrong verification code. This approach addresses the high security risk associated with enterprise verification management.

[0036] Preferably, if the verification information is SMS, TLS encryption can be used to directly connect to the carrier gateway (packet loss rate <0.01%) to extract features. If the verification information is voice, an integrated ASR engine with a WER <5% can be used to extract features to achieve mixed Chinese and English recognition. If the verification information is image, a ResNet-50 model (anti-distortion attack accuracy >99%) can be used to extract features.

[0037] Furthermore, the specific process of the above step S20 is as follows: S21: configuring modal weights corresponding to the multimodal features.

[0038] S22: Call the extraction function corresponding to the multimodal feature.

[0039] S23: The modal weight and the extraction function are placed into a preset multimodal fusion function, and the recognition confidence is calculated.

[0040] S24: The recognition confidence and multimodal features are placed into the semantic understanding engine based on the Transformer architecture, and the parsing is performed to obtain the verified parsing results.

[0041] In this embodiment, the modal weights include text weight, image weight, and speech weight; the image weight is greater than the speech weight and less than the text weight, and the sum of the text weight, image weight, and speech weight is 1. By properly configuring multiple weights, the accuracy of the verification and parsing results can be further improved.

[0042] As a preferred solution, the text weight is 0.6, the image weight is 0.3 and the speech weight is 0.1.

[0043] More specifically, the multimodal fusion function is:

[0044] Where, Indicates a given under conditions The recognition confidence of Indicates the The feature extraction function of each modality, express function, represents the bias term, Indicates the The weight of each mode. The range is 0-1.

[0045] It should be explained that after the multi-dimensional weights are initially configured, the semantic understanding engine will dynamically adjust the weights based on the understanding of the context and specific modal features.

[0046] Therefore, through the design of steps S21-S24, relying on the Transformer architecture's semantic understanding engine and the aforementioned multimodal fusion function, the present method achieves cross-language parsing of text, image, and voice verification codes. It also improves recognition accuracy in complex scenarios through contextual analysis (such as automatically marking the timeliness of "verification code valid for 5 minutes"). Using this method, the parsing accuracy can reach 99.2%.

[0047] Furthermore, in order to further improve the verification efficiency, before step S10, the method of the present invention further includes: S100: In response to the verification requirement of the user terminal, according to the preset number resource scheduling model, a mobile phone number for information verification is scheduled to the user terminal from a preset number pool.

[0048] S200: Submit a verification request to a third-party platform in the name of the mobile phone number.

[0049] It should be explained that the number pool refers to a collection of a large number of numbers (such as phone numbers, SMS verification code numbers, service numbers, etc.), which can be dynamically allocated, recycled and reused by the system.

[0050] Furthermore, the above number resource scheduling model is specifically as follows:

[0051] Where, Indicates the actual allocation parameters of mobile phone numbers. Indicates Find the minimum value in the range and output the corresponding A function of the value, Indicates the The current task volume of mobile phone numbers, Indicates the The reputation score of each mobile number, represents the smoothing coefficient, Indicates the total number of mobile phone numbers. Indicates the allocation parameters of the current mobile phone number. Indicates the allocation parameters of the current mobile phone number. It means taking the square of the 2-norm.

[0052] Through steps S100-S200, the method of the present invention can promptly retrieve available mobile phone numbers for employees to verify based on their actual login needs. During the number resource scheduling process, the number resource scheduling model employed fully considers the current workload, reputation score, and stability of the mobile phone number, preventing logged-in mobile phone numbers from being unnecessarily logged out, thereby ensuring seamless mobile phone number scheduling. Furthermore, the method of the present invention can screen out risky mobile phone numbers (those blacklisted by the operator), thereby improving the efficiency of number resource scheduling and further enhancing the efficiency of the verification process.

[0053] It should be explained that, in general, the smoothing coefficient is set to 0.1 to control the volatility of the allocation. The smoothing coefficient can also be adaptively adjusted according to the employee's job category. For example, if the employee's job is in customer service, the smoothing coefficient is set to 0.2. If the employee's job is in marketing, the smoothing coefficient is set to 0.05. When the smoothing coefficient is small, the scheduling plan will be more inclined to allocate based on the current number of tasks and the number credit score, which may result in larger allocation fluctuations. When the smoothing coefficient is large, the scheduling plan will pay more attention to maintaining consistency with the previous allocation plan, reducing the volatility of the allocation, and making the allocation of number resources more stable.

[0054] Furthermore, in order to avoid verification failure caused by SIM card communication failure during the verification process, before the above step S10, the method of the present invention further includes: A preset fault prediction model is used to predict the failure probability of the corresponding SIM card in the preset number pool.

[0055] If the failure probability of the corresponding SIM card is greater than the failure threshold, a self-recovery mechanism of the system or device to which the corresponding SIM card belongs is triggered.

[0056] Specifically, the algorithm of the above fault prediction model is:

[0057] Where, represents the probability of failure, Indicates the The product of the probabilities of class failures, for A subset of Indicates the Basic probability of type failure (if it is signal attenuation, Take 0.2; if the SIM card is oxidized, Take 0.5), Indicates the cumulative number of days of failure indicators.

[0058] Specifically, when When it is greater than 0.7, the self-recovery mechanism of the system or device to which the corresponding SIM card belongs is triggered.

[0059] Compared to existing technologies, this method can calculate the probability of complex failures, resulting in more reliable predicted failure probabilities. Furthermore, it can trigger preventative maintenance, predicting the risk of SIM card oxidation 48 hours in advance.

[0060] During the above step S30, in order to ensure communication quality and reduce delays in the verification code filling process, the method of the present invention further includes: Use the preset signal strength monitoring module to evaluate the communication strength of the mobile phone number under the current operator in real time.

[0061] If the communication strength is lower than the communication threshold, the current operator of the mobile phone number is switched to the backup operator.

[0062] In this embodiment, the strength monitoring module is used to calculate the RSRP value, triggering the handover delay of the three networks (China Telecom, China Mobile and China Unicom) to be less than 500ms. The communication threshold is -110dBm.

[0063] Furthermore, after sending the verification and analysis results to the user terminal, the method of the present invention further includes: Calculate the abnormality of the user's operation based on the user's operation records.

[0064] The degree of operational abnormality serves as an indicator parameter for number pool security audit management.

[0065] Specifically, the algorithm for calculating the abnormality of the user terminal operation is as follows:

[0066] Where, Indicates the degree of operational abnormality. Indicates the verification operation score of enterprise employees, Represents the mean of abnormal operation data of enterprise employees, Indicates the standard deviation of abnormal operation data of enterprise employees, Indicative function indicating whether the operator’s blacklist is hit, represents the penalty term, Represents the operator for taking the L2 norm.

[0067] In this embodiment, the penalty term is set to 0.5.

[0068] Through the above technical solution, after completing the verification behavior, the method of the present invention will calculate the operation abnormality degree as an indicator parameter for security audit management, thereby realizing audit traceability and timely detecting abnormal verification behavior, thereby further reducing the security risks of enterprise verification management.

[0069] Furthermore, in order to control the communication delay within a relatively reasonable range under the premise of low maintenance cost in enterprise verification management, the method of the present invention introduces a multi-dimensional routing decision model to combine maintenance cost and realize dynamic channel quality scoring. The specific calculation method is as follows:

[0070] Where, represents the State-Action Value Function, Indicates the number of successful historical verification code receptions. Indicates the total number of historical verification applications. Indicates the preset channel cost threshold, represents the total cost of the historical verification process, Indicates the time interval between the current time and the most recent failure. represents the first weight coefficient, represents the second weight parameter, represents the third weight coefficient, e represents the base of the logarithm of natural numbers, Represents the attenuation coefficient.

[0071] Preferably, .

[0072] Compared with the existing technology, the above method introduces cost budget constraints , so that enterprises can set cost thresholds (such as the cost of a single verification code < 0.05 yuan) to control verification costs. A time decay factor is also introduced. To improve the dynamic adaptability of the score calculation, so as to facilitate the adjustment of the sensitivity of the operator's strategy (for example, when it is detected that the operator's "mobile" frequently blocks the account, its weight will be gradually reduced).

[0073] Example 2 like Figure 2As shown, this embodiment discloses a computer system, including a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the verification code management method based on multimodal and reinforcement learning recorded in Example 1 is implemented.

[0074] Example 3 like Figure 3 As shown, this embodiment discloses a verification code management system based on multimodal and reinforcement learning, which includes a hardware layer, a data processing layer, a decision layer, an application layer and a security layer.

[0075] The hardware layer is used to implement the loading of the physical SIM card, monitor the signal strength of the SIM card, and execute the self-recovery mechanism.

[0076] The data processing layer is used to execute steps S10 to S30 described in the first embodiment.

[0077] The decision layer is used to execute the channel quality dynamic scoring process and number resource scheduling process recorded in the first embodiment.

[0078] The application layer is used to implement API docking with enterprise systems and automatically fill in verification fields.

[0079] The security layer is used to record the historical operation records of the enterprise system and execute the security audit management process recorded in the first embodiment based on the historical operation records.

[0080] Although this specification has shown and described several embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and substitutions without departing from the idea and spirit of the present invention.

Claims

1. A verification code management method based on multimodal and reinforcement learning, characterized in that: include: In response to verification information sent by the third-party platform, extract multimodal features of the verification information; Calling a preset semantic understanding engine to fuse and analyze the multimodal features to obtain a verification and analysis result; Execute verification behavior according to the verification and analysis results; Alternatively, the verification and analysis result is sent to the user end.

2. The verification code management method based on multimodal and reinforcement learning according to claim 1 is characterized in that: The multimodal features are integrated and analyzed to obtain verification and analysis results, including: Configuring modal weights corresponding to the multimodal features; Calling the extraction function corresponding to the multimodal feature; Putting the modal weight and the extraction function into a preset multimodal fusion function to calculate the recognition confidence; The recognition confidence and the multimodal features are placed in a semantic understanding engine based on the Transformer architecture, and parsed to obtain the verification parsing result.

3. The verification code management method based on multimodal and reinforcement learning according to claim 2 is characterized in that: The multimodal fusion function is specifically: Where, Indicates a given under conditions The recognition confidence of Indicates the The feature extraction function of each modality, express function, represents the bias term, Indicates the The weight of each mode.

4. The verification code management method based on multimodal and reinforcement learning according to claim 2 is characterized in that: The modality weight includes at least a text weight, an image weight and a voice weight; wherein the image weight is greater than the voice weight and less than the text weight, and the sum of the text weight, the image weight and the voice weight is 1.

5. The verification code management method based on multimodal and reinforcement learning according to claim 1 is characterized in that: Before extracting the multimodal features of the verification information in response to the verification information sent by the third-party platform, the method further includes: In response to the verification requirement of the user terminal, according to the preset number resource scheduling model, scheduling a mobile phone number for information verification from a preset number pool to the user terminal; Submit a verification request to the third-party platform in the name of the mobile phone number.

6. The verification code management method based on multimodal and reinforcement learning according to claim 5 is characterized in that: The number resource scheduling model is specifically as follows: Where, Indicates the actual allocation parameters of mobile phone numbers. Indicates Find the minimum value in the range and output the corresponding A function of the value, Indicates the The current task volume of mobile phone numbers, Indicates the The reputation score of each mobile number, represents the smoothing coefficient, Indicates the total number of mobile phone numbers. Indicates the allocation parameters of the current mobile phone number. Indicates the allocation parameters of the current mobile phone number. It means taking the square of the 2-norm.

7. The verification code management method based on multimodal and reinforcement learning according to claim 5 is characterized in that: Executing verification behavior according to the verification and analysis result; Alternatively, the verification and analysis result is sent to the user terminal, and the method further includes: Using a preset signal strength monitoring module, the communication strength of the mobile phone number under the current operator is evaluated in real time; If the communication strength is lower than the communication threshold, the current operator of the mobile phone number is switched to a backup operator.

8. The verification code management method based on multimodal and reinforcement learning according to claim 1 is characterized in that: Before extracting the multimodal features of the verification information in response to the verification information sent by the third-party platform, the method further includes: Use the preset fault prediction model to predict the failure probability of the corresponding SIM card in the preset number pool; If the failure probability of the corresponding SIM card is greater than the failure threshold, a self-recovery mechanism of the system or device to which the corresponding SIM card belongs is triggered.

9. The verification code management method based on multimodal and reinforcement learning according to claim 1, characterized in that: After sending the verification and analysis result to the user terminal, the method further includes: Calculating the abnormality of the operation of the user terminal according to the operation record of the user terminal; The operation abnormality degree is used as an indicator parameter for number pool security audit management.

10. A computer system, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the verification code management method based on multimodal and reinforcement learning according to any one of claims 1 to 9 is implemented.

Images