Method, device and medium for verifying authenticity of policy change intention

Abstract

The present application relates to the field of artificial intelligence technology, and discloses a life insurance policy change willingness authenticity verification method, device, equipment and medium, the method comprises: triggering a multi-modal data acquisition process; capture the facial video stream when the user performs the preset identity verification operation, extract the facial key point micro-motion features; in the process of reading the preset legal confirmation text, the voice signal is captured synchronously, and the acoustic features reflecting the psychological stress are extracted; in the last confirmation link, cognitive trap problems are presented to the user, and feedback behavior data is recorded; based on the above multi-modal data, the preset model is used to calculate the comprehensive confidence of the willingness authenticity; if it is lower than the threshold, it is determined that the non-real willingness expression is intercepted. It can be applied to the financial technology scene, the present application improves the financial business compliance verification from identity comparison to psychological authenticity audit level, can accurately identify the non-voluntary life insurance policy change behavior in the state of being coerced and being manipulated, and effectively prevent compliance risks.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a method, apparatus, equipment and medium for verifying the authenticity of an intention to change an insurance policy. Background Technology

[0002] With the development of fintech, online insurance service platforms have enabled end-to-end policy change processing. Insurance applications, such as Ping An Financial Manager, now support policyholders in completing key operations online, including beneficiary designation and policyholder changes, significantly improving efficiency and user experience. In the sensitive area of ​​beneficiary designation, to ensure the operation is performed by the individual, current technology typically employs biometric identification techniques, such as facial recognition and liveness detection (e.g., nodding, shaking, blinking), to verify the operator's identity, complying with regulatory requirements for "real-name registration" in financial transactions.

[0003] However, existing technologies have the following shortcomings and deficiencies in practical applications: First, "real identity" does not equal "real will." Existing liveness detection technology can only prove that the operator in front of the camera is the policyholder, but it cannot prove that the operation is a true manifestation of their free will. In extreme scenarios such as family disputes, inheritance disputes, or even physical coercion, the parties involved may be threatened with knives or verbally intimidated outside the camera, forced to complete liveness detection actions such as nodding or shaking their heads. The system only passes the verification based on the identity comparison result, failing to recognize the discrepancy between will and identity in this "manipulated" state. Second, the lack of physiological feedback information under stress. When a person is in a state of fear, anxiety, or performing actions against their will, their physiological state will undergo uncontrollable changes, which may manifest as subtle physiological manifestations. However, existing interactive interfaces only record the user's click confirmation result. No data reflecting the user's real-time physiological state is collected throughout the entire business process, resulting in the system being completely unaware of the user's true physical and mental state when confirming key terms, and unable to determine whether they are under abnormal pressure.

[0004] Furthermore, the existing system's business process design suffers from a lack of simplistic interaction logic. The current policy change process is linearly guided; the system simply collects user confirmation information according to a pre-defined sequence of steps, and users merely follow the on-screen prompts to complete the prescribed actions. This linear interaction design fails to detect abnormal user behavior when cognitive resources are excessively occupied. For manipulated or coerced users, their brains must simultaneously process external threats and complete operational tasks, resulting in a cognitive load far exceeding normal levels. When dealing with unconventional issues, they often exhibit delayed responses or logical contradictions, but the existing system, lacking any unconventional interaction elements, is completely unable to capture such abnormal behavioral characteristics. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for verifying the authenticity of policy change intentions, aiming to solve the technical problem that existing technologies cannot identify the true intention of users in handling sensitive business, leading to the risk of being coerced into handling business.

[0006] In a first aspect, embodiments of the present invention provide a method for verifying the authenticity of a policy change intention, including: Upon receiving a user's policy change request, a multimodal physiological feedback and behavioral data collection process is triggered. In response to a trigger command, during the user's execution of a preset authentication operation, the system continuously captures the user's facial video stream through an image acquisition device and extracts the microscopic motion features of key facial points. During the process of the user reading the preset legal confirmation text, the user's voice signal is captured synchronously through an audio acquisition device, and acoustic features reflecting the user's psychological stress are extracted from the voice signal. In the final confirmation stage after the user completes the policy change process, one or more preset cognitive trap questions are presented to the user, and the user's feedback behavior data on the cognitive trap questions is recorded. Based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, a comprehensive confidence level representing the authenticity of the user's intention is calculated through a preset multimodal fusion verification model. If the overall confidence level is lower than a preset threshold, it is determined that the current policy change request does not represent the user's true intention, and corresponding risk warnings and blocking instructions are output.

[0007] Secondly, embodiments of the present invention also provide a device for verifying the authenticity of a policy change intention, comprising: The request receiving module is used to trigger the multimodal physiological feedback and behavioral data collection process after receiving a user's policy change request. The micro-expression analysis module is used to respond to trigger commands and continuously capture the user's facial video stream through an image acquisition device during the user's execution of preset identity verification operations, and extract the micro-motion features of key facial points. The acoustic stress analysis module is used to capture the user's voice signal synchronously through an audio acquisition device while the user reads a preset legal confirmation text, and to extract acoustic features reflecting the user's psychological stress from the voice signal. The cognitive trap module is used to present one or more preset cognitive trap questions to the user during the final confirmation stage of the policy change process, and to record the user's feedback behavior data on the cognitive trap questions. The multimodal fusion decision module is used to calculate the comprehensive confidence level representing the authenticity of the user's intention based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, through a preset multimodal fusion verification model. The risk interception module is used to determine that the current policy change request does not represent the user's true intention if the overall confidence level is lower than a preset threshold, and to output corresponding risk warnings and interception instructions.

[0008] Thirdly, embodiments of the present invention also provide a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-mentioned method for verifying the authenticity of the intention to change the policy.

[0009] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the above-mentioned method for verifying the authenticity of the intention to change the policy.

[0010] This invention collects facial micro-movement features when a user performs a preset identity verification operation and speech acoustic features when the user reads a legal confirmation text. It then combines this with multimodal fusion analysis of user feedback behavior data under cognitive trap conditions. This allows for a comprehensive assessment of the user's true psychological state from both physiological feedback and behavioral response dimensions. Unlike existing technologies that only verify identity, this solution captures uncontrollable muscle tremors, speech stress features, and behavioral abnormalities caused by cognitive overload under duress. This enables in-depth verification of the user's true intentions, fundamentally addressing the compliance risk of discrepancies between identity authenticity and intention. Furthermore, this solution expands the verification dimension from a single identity feature to a multi-feature space encompassing physiological stress and behavioral abnormalities. When the overall confidence level falls below a preset threshold, timely interception is achieved, effectively identifying and blocking involuntary policy changes in scenarios such as family disputes, inheritance battles, and personal coercion. This significantly enhances the risk control capabilities of financial businesses in sensitive areas. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic diagram of an application environment for the method for verifying the authenticity of policy change intentions provided in an embodiment of the present invention; Figure 2A flowchart illustrating the method for verifying the authenticity of policy change intentions provided in an embodiment of the present invention; Figure 3 A schematic diagram of the micro-expression analysis process provided in an embodiment of the present invention; Figure 4 A schematic diagram of the acoustic pressure analysis process provided in an embodiment of the present invention; Figure 5 A schematic diagram of the cognitive trap recording process provided in an embodiment of the present invention; Figure 6 A schematic diagram of a device for verifying the authenticity of policy change intentions provided in an embodiment of the present invention; Figure 7 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0015] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0017] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0018] The method for verifying the authenticity of policy change intentions provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can receive policy change requests from users through the client. Upon receiving the request, the server triggers a multimodal physiological feedback and behavioral data collection process. Responding to the trigger command, during the user's pre-defined identity verification process, the server continuously captures the user's facial video stream using an image acquisition device and extracts the microscopic motion features of key facial points. During the user's reading of a pre-defined legal confirmation text, the server simultaneously captures the user's voice signal using an audio acquisition device and extracts acoustic features reflecting the user's psychological stress from the voice signal. In the final confirmation stage after the user completes the policy change process, the server presents the user with one or more pre-defined cognitive trap questions and records the user's feedback behavior data on these questions. Based on the microscopic motion features of the key facial points, the acoustic features, and the feedback behavior data, a pre-defined multimodal fusion verification model is used to calculate the comprehensive confidence level representing the authenticity of the user's intentions. If the comprehensive confidence level is lower than a pre-defined threshold, the server determines that the current policy change request does not represent the user's true intentions and outputs corresponding risk warnings and interception commands. In this invention, by collecting facial micro-movement features when a user performs a preset identity verification operation and speech acoustic features when the user reads a legal confirmation text, and combining this with multimodal fusion analysis of the user's feedback behavior data under cognitive trap problems, the true psychological state of the user can be comprehensively judged from two dimensions: physiological feedback and behavioral response. Unlike existing technologies that can only verify whether it is the person in question, this technical solution achieves in-depth verification of whether it is the person's true intention by capturing muscle tremors that cannot be controlled involuntarily under duress, speech stress features, and behavioral abnormalities caused by cognitive overload. This fundamentally solves the compliance risk of discrepancies between identity authenticity and intention. This technical solution expands the verification dimension from a single identity feature to a multi-feature space that includes physiological stress and behavioral abnormalities. When the overall confidence level is lower than a preset threshold, it can promptly intercept and effectively identify and block involuntary policy change behaviors in scenarios such as family disputes, inheritance disputes, and personal coercion, significantly improving the risk control capabilities of financial business in sensitive areas. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a separate server or a server cluster composed of multiple servers. The present invention will now be described in detail through specific embodiments.

[0019] Please see Figure 2 This invention provides a method for verifying the authenticity of a policy change intention, including steps S1-S6: S1. Upon receiving a user's policy change request, the multimodal physiological feedback and behavioral data collection process is triggered.

[0020] In this step, when a user initiates a policy change operation involving the sensitive "beneficiary designation" step on an online platform (such as the Ping An Financial Manager APP), the system first identifies the high-risk nature of the operation and automatically enters an enhanced intent verification step to replace or supplement the existing standard identity verification process. Specifically, upon detecting that the user clicks on the "beneficiary designation" or "policyholder change" function entry, a prompt message immediately pops up on the front-end interface, informing the user that a composite intent verification involving facial recognition, voice collection, and simple Q&A will be conducted. Simultaneously, the mobile terminal's front-facing camera and microphone are automatically activated, and corresponding system resources are allocated for data collection preparation. This setup elevates compliance verification from simple biometric comparison to a psychological authenticity audit level, laying the foundation for subsequent multi-dimensional physiological feedback and behavioral characteristic analysis. This effectively addresses the risk of individuals being forced to conduct business in extreme scenarios such as family disputes, inheritance disputes, or personal coercion.

[0021] S2. In response to the trigger command, during the user's execution of the preset authentication operation, the user's facial video stream is continuously captured by the image acquisition device, and the microscopic motion features of the facial key points are extracted.

[0022] In this step, during the user's facial liveness detection or signature confirmation stage, the front-facing camera continuously captures facial video streams. Taking the Ping An Financial Manager APP as an example, when the user enters the beneficiary designation stage, the system first requires the user to complete standard liveness detection actions (such as nodding, shaking the head, and blinking). During this process, this invention not only records the identity verification result but also simultaneously collects facial key point movement data. It should be noted that even if the person is threatened with a knife or verbally intimidated outside the camera and forced to complete the liveness detection and reading actions, their facial muscles will still produce uncontrollable stress responses, which is precisely the key information that this invention aims to capture.

[0023] In a specific embodiment, the step of extracting the microscopic motion features of facial key points in step S2, such as... Figure 3 As shown, the specific steps include S21-S23: S21. Perform frame-by-frame analysis on the facial video stream to track and locate the coordinates of multiple preset key points on the face.

[0024] In this step, a facial keypoint detection algorithm is used to analyze the facial video stream frame by frame at a frequency of more than 30 frames per second, accurately locating the coordinates of 68 key points on the face in each frame. These key points cover the eyebrows (such as the outer corner of the left eyebrow, the brow peak, and the inner corner of the eyebrow), eyes (such as the corner of the eye and the edge of the eyelid), nose (such as the tip of the nose and the alar of the nose), mouth (such as the corner of the mouth and the midpoint of the upper and lower lips), and facial contours (such as the mandibular point and the cheekbone point). Special attention is paid to the core facial expression muscle areas, including the interbrow muscles (corresponding to the eyebrow area, reflecting emotions such as frowning and tension), the orbicularis oris muscles (corresponding to the corner of the mouth area, reflecting micro-expressions such as tension and fake smiles), and the orbicularis oculi muscles (corresponding to the periorbital area, reflecting the difference between a genuine smile and a fake smile). Through meticulous tracking of these areas, minute muscle movements that are difficult to detect with the naked eye can be captured.

[0025] S22. Based on the temporal coordinate changes of the key points, calculate the micro-motion amplitude and frequency of each key point.

[0026] In this step, for each key point, its displacement vector over a continuous time series is extracted to form the motion trajectory of that point. The focus is on analyzing involuntary, high-frequency, low-amplitude micro-tremors, typically with amplitudes between 0.1 mm and 2 mm and frequencies between 8 Hz and 20 Hz. Under normal circumstances, facial muscle movements are smooth, coordinated, and low-amplitude, exhibiting a regular movement pattern. However, under duress, when users attempt to conceal their fear, facial muscles exhibit a "high-frequency tremor after stiffness" characteristic—first stiffening due to tension, followed by uncontrollable high-frequency micro-tremors, making the movement pattern chaotic and disordered. Micro-motion characteristics of the eyebrow region, corner of the mouth region, and periorbital region are calculated separately to provide basic data for subsequent entropy value calculations.

[0027] S23. The amplitude and frequency of the micro-movements are spatiotemporally encoded to quantify the micro-vibration entropy value of facial muscles under involuntary control. The micro-vibration entropy value is used to characterize the degree of unnatural fluctuations in facial expressions.

[0028] In this step, the present invention calculates the information entropy of the facial motion sequence. This is used to quantify the degree of disorder. The higher the information entropy, the greater the randomness and disorder of facial muscle movements, and the more likely they are to be in an unnatural state. The calculation formula is: in, This represents the i-th micro-motion state pattern observed on the time axis. This represents the probability of the pattern occurring. In the specific calculation, the continuous motion trajectory is discretized into several motion patterns (such as upward micro-movement, downward micro-movement, leftward micro-movement, rightward micro-movement, stillness, etc.), and the probability of each pattern is statistically analyzed and substituted into the formula to calculate the entropy value. This characteristic is quantified by calculating the "micro-vibration entropy value." A significantly high entropy value deviating from the normal baseline is an important indicator of potential risk. For example, under normal conditions, a user's micro-vibration entropy value typically fluctuates between 0.3 and 0.5; however, when a user is under duress, due to involuntary high-frequency muscle tremors, the entropy value may surge to over 0.8. Based on this, forced designation behaviors against the individual's true will can be accurately identified, thereby effectively intercepting high-risk policy changes that could lead to disputes.

[0029] S3. During the process of the user reading the preset legal confirmation text, the user's voice signal is captured synchronously through the audio acquisition device, and acoustic features reflecting the user's psychological pressure are extracted from the voice signal.

[0030] In this step, during the policy change process, a pre-set legal confirmation text will be displayed on the screen, such as "I confirm that the above beneficiary designation is my true intention and has not been coerced or deceived by any other party. I fully understand the legal consequences of this change and assume legal responsibility for all my actions during this process." The user must face the camera and clearly read this text at a normal speaking speed. During the reading, the system collects the voice signal through the microphone and performs real-time noise reduction to eliminate ambient noise. This invention proposes a voice acoustic pressure weighting method to analyze the pressure characteristics in speech. The theoretical basis of this method is that when a person faces fear, extreme anxiety, or makes a decision against their will, the autonomic nervous system becomes uncontrolled, leading to tension in the laryngeal muscles and unstable vocal cord vibration. These changes are reflected in the speech signal in the form of acoustic characteristics.

[0031] In a specific embodiment, step S3, which involves extracting acoustic features reflecting the user's psychological stress from the speech signal, is as follows: Figure 4 As shown, the specific steps include S31-S33: S31. Preprocess the speech signal and extract its fundamental frequency trajectory.

[0032] In this step, the acquired raw speech signal is preprocessed, including pre-emphasis, framing (25ms per frame, 10ms frame shift), windowing, etc., and then the fundamental frequency of each frame of speech signal is extracted using autocorrelation or cepstral method. This creates a continuous fundamental frequency trajectory. The fundamental frequency is the basic frequency of vocal cord vibration, typically between 80-200Hz for males and between 150-350Hz for females. Under normal circumstances, the fundamental frequency trajectory should be smooth and slowly decreasing when reading declarative sentences; however, under psychological stress, the fundamental frequency will fluctuate abnormally.

[0033] S32. Based on the fundamental frequency trajectory, calculate the fundamental frequency jitter parameter reflecting the instability of vocal cord vibration and the frequency offset parameter reflecting abnormal changes in tone.

[0034] In this step, fundamental frequency jitter is a parameter reflecting the minute changes in fundamental frequency between adjacent cycles, usually expressed as a percentage. Under psychological pressure, subtle tremors in the laryngeal muscles can lead to instability in the vocal cord vibration cycle, resulting in a significant increase in the jitter index. Normal jitter is typically between 0.5% and 1.0%, while it can exceed 2.0% under high stress. Simultaneously, the frequency shift of intonation is analyzed, i.e., the deviation of the instantaneous fundamental frequency from the smooth fundamental frequency trajectory is calculated. When coerced, users may exhibit two extreme behaviors due to extreme tension: one is an unnatural "monotonous flatness" in tone due to an attempt to control emotions, with almost no intonation; the other is a sudden "sudden rise" in intonation due to emotional loss of control, with abnormal pitch jumps at the end of sentences or specific words. The average jitter value and maximum frequency shift are calculated throughout the entire reading process.

[0035] S33. The base frequency jitter parameter and the frequency offset parameter are weighted and fused to generate a stress index for quantifying the psychological stress contained in the user's voice.

[0036] In this step, the deviation of the instantaneous fundamental frequency is calculated, and a stress index is generated to quantify the psychological stress behind the speech. The calculation formula is as follows: in, This represents the maximum deviation of the instantaneous fundamental frequency from the smooth fundamental frequency trajectory. The average fundamental frequency, These are preset weighting coefficients used to map the fundamental frequency deviation to the range of the pressure index. The value of can be set according to the actual application scenario and sensitivity requirements, and is usually obtained through experimental calibration. In a preferred embodiment, The value ranges from 0.5 to 2.0. For example, when the system is highly sensitive to risk and it is desirable to improve the detection rate of stress states, it can be set to... At this point, even a small fluctuation in the fundamental frequency can cause a significant increase in the stress index; when the system wants to reduce the false alarm rate and prioritize user experience, it can be set to... =0.8, at which point only significant fundamental frequency anomalies will trigger an alert. Under the default configuration, The value is set to 1.0, at which point the stress index represents the relative deviation of the fundamental frequency. The core of this formula lies in quantifying the degree of psychological stress in speech through the relative deviation of the fundamental frequency. Under psychological pressure, the contraction of the laryngeal muscles leads to unstable vocal cord vibration, manifesting as abnormal fluctuations in the fundamental frequency. This stress index is used to quantify the degree of psychological stress contained in the user's speech; the closer the index is to 1, the greater the psychological stress. It should be noted that the fundamental frequency jitter parameter, as an auxiliary judgment indicator, together with the frequency offset parameter, constitutes the basis for stress judgment. However, in quantization fusion, this embodiment uses the frequency offset as the primary quantification indicator, while the fundamental frequency jitter parameter is used for auxiliary verification and threshold adjustment.

[0037] S4. In the final confirmation stage after the user completes the policy change process, present the user with one or more preset cognitive trap questions and record the user's feedback behavior data on the cognitive trap questions.

[0038] In this step, during the final confirmation stage after the user completes the policy change process, a "cognitive trap" question completely unrelated to the current policy change transaction pops up. The question is designed based on the theory of "cognitive load" to test whether the user experiences delayed responses or contradictory thinking due to being "manipulated." Users who are manipulated or coerced have their cognitive resources heavily consumed by fear and external threats, resulting in a cognitive load far exceeding normal levels. They will exhibit abnormal reactions when dealing with such unpredictable questions.

[0039] In a specific embodiment, the cognitive trap problem includes problems unrelated to the current policy change business that require the user to access long-term memory or perform simple logical reasoning; step S4, which records the user's feedback behavior data on the cognitive trap problem, is as follows: Figure 5 As shown, the specific steps include S41-S43: S41. Record the response delay time between when the user receives the cognitive trap question and when they make their first response.

[0040] In this step, randomly presented questions include: "Please quickly state the city of residence you filled in when you purchased this policy three years ago within three seconds" or "Please calculate how old you were five years ago if you are 45 years old this year." These questions cannot be answered directly from the current interface; users need to access long-term memory or perform simple mental arithmetic. The response delay is precisely recorded from the moment the question is fully displayed on the screen until the user makes their first response (including clicking a screen button, touching an option, or beginning a verbal answer). For normal users in a relaxed state, the response delay for such simple questions is typically between 1.5 and 3.0 seconds; however, manipulated users, due to excessive cognitive load, may experience response delays exceeding 5 seconds or even longer.

[0041] S42. Record the accuracy of the user's answers to the cognitive trap questions.

[0042] In this step, the user's answers are compared with historical records stored in the database to determine their correctness. For example, for questions about the user's city of residence, the answer is compared to the address information the user entered when purchasing the policy; for age calculation questions, the correct answer is calculated based on the user's date of birth. Manipulated users, due to their anxious state and distracted attention, will have a significantly higher error rate. Normally, the accuracy rate for such simple questions should be above 95%, but the accuracy rate of manipulated users may drop below 60%.

[0043] S43. The response delay time and the accuracy of the answer are used as the feedback behavior data.

[0044] In this step, the data from the two dimensions mentioned above—response delay time and answer accuracy—are used as feedback behavioral data and input into the subsequent fusion model. Users who are being manipulated or in a state of high anxiety have their cognitive resources consumed by fear. When dealing with unpredictable questions that require calm recall or reasoning, their cognitive load peaks instantly, resulting in significantly longer response delays and extremely high error rates compared to normal users. This data is crucial evidence for determining whether their mental state is normal; therefore, users with excessively long response delays or incorrect answers are labeled as candidates for "abnormal expression of intent."

[0045] S5. Based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, calculate the comprehensive confidence level representing the authenticity of the user's intention through a preset multimodal fusion verification model.

[0046] In this step, data from all modalities are fused and analyzed. The fusion model can be a classifier based on logistic regression, support vector machines, or deep learning (such as multi-layer neural networks or attention mechanism networks). Taking a three-layer neural network as an example, the input layer includes features such as micro-vibration entropy, pressure index, response delay time, and answer accuracy; the hidden layer learns higher-order interaction relationships between features through nonlinear transformations; and the output layer uses the sigmoid function to output a "comprehensive confidence score" between 0 and 1. During the training phase, this model has already learned a large amount of sample data labeled with "true intention" and "false intention." The samples include users conducting normal business, volunteers simulating coercion, and actual cases from historical disputes. By inputting the three dimensions of the current user's features, the model can output a comprehensive confidence score; the higher the value, the greater the probability that the user has a true intention.

[0047] S6. If the overall confidence level is lower than the preset threshold, it is determined that the current policy change request does not represent the user's true intention, and corresponding risk warnings and blocking instructions are output.

[0048] In this step, a risk threshold, such as 0.8, is set based on business rules and risk preferences. Once the overall confidence level falls below 0.8, the current policy change process is immediately interrupted, and a red warning box pops up on the front-end interface: "This operation may involve involuntary risk; the change process has been automatically suspended. To continue, please contact customer service for manual review." Simultaneously, a risk work order is sent to the back-end review personnel, attaching the overall confidence score and an analysis summary of each modal characteristic. An interception command is automatically generated to prevent the change from taking effect, and the relevant data is stored in the high-risk operation database for subsequent auditing and model optimization. Through MEA and PSW analysis, forced designation behaviors that violate the insured's true will can be accurately identified, improving the interception efficiency of high-risk policy changes and effectively protecting the legitimate rights and interests of policyholders, preventing illegal appropriation.

[0049] In a specific embodiment, before the step of calculating the comprehensive confidence level representing the authenticity of the user's intention based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data through a preset multimodal fusion verification model, the method further includes the following step: A1. Construct an initial multimodal fusion verification model, and use historical business processing data with known intention authenticity labels as training samples to train the initial multimodal fusion verification model until the model converges; The training samples include facial micro-motion features, speech acoustic features, feedback behavior data on cognitive trap questions, and corresponding intention authenticity labels of sample users.

[0050] In this step, the initial multimodal fusion validation model is trained using a large amount of labeled historical data to ensure that the model can accurately identify the multimodal feature differences between genuine and insincere expressions of intent. Training samples include: normal business processing data (labeled as genuine intent), data from volunteers in psychological experiments who were asked to simulate coercion (labeled as insincere intent), and confirmed past dispute case data (labeled as insincere intent). Cross-validation and early stopping are used during training to prevent overfitting and ensure the model's accuracy and robustness. After model training, A / B testing is conducted in real-world business scenarios to continuously optimize model parameters and threshold settings.

[0051] In a specific embodiment, the method for verifying the authenticity of policy change intentions further includes the following steps: S6a. When it is determined that the current policy change request does not represent the user's true intention, a sincerity assessment report is generated and stored. The intention authenticity assessment report includes at least the facial video stream, the voice signal, the calculated overall confidence level, and an analytical summary of the various micro-motion and acoustic features used to indicate non-genuine intention.

[0052] The generated "Intention Authenticity Assessment Report" in this step can serve as electronic evidence, providing indelible supporting evidence in future insurance contract disputes. The report includes: operation time, operator identification information, complete video and audio recordings of the reading, frame-by-frame analysis of facial key point motion trajectories, fundamental frequency variation curves, radar charts comparing various dimensional feature values ​​with the normal baseline, a final comprehensive confidence score, and detailed analysis conclusions. The report is automatically timestamped and digitally signed, and stored in a compliant evidence storage system in PDF format. This mechanism significantly enhances the legal validity and enforceability of insurance contracts; in the event of a dispute, the insurance company can use this report to prove that it has fulfilled its due diligence obligations.

[0053] In a specific embodiment, after the step of outputting the corresponding risk warning and interception command, step S7 is further included: All data from this verification process, including the facial video stream, voice signal, feedback behavior data, overall confidence level, and evaluation report, will be packaged into an immutable electronic record and uploaded to a blockchain or designated record platform for subsequent auditing and verification.

[0054] In this step, for all high-risk operations, especially intercepted requests, an electronic evidence preservation process will be initiated. All raw data (video streams, audio streams), intermediate processing results (key point coordinate sequences, fundamental frequency sequences), calculated indicators (micro-vibration entropy, pressure index, response delay time), final results (comprehensive confidence level, assessment report), and metadata (timestamps, device information, network environment) will be packaged into a complete, digitally signed electronic evidence package. This evidence package is uploaded to a consortium blockchain (such as Ping An's self-built blockchain platform) or a designated evidence preservation platform via a smart contract, obtaining a blockchain evidence preservation number and timestamp. This evidence is non-repudiable and traceable, providing strong supporting evidence for potential future legal disputes. Furthermore, this technology is particularly suitable for elderly people or customers with declining cognitive abilities. Elderly people often experience facial muscle relaxation and significant changes in voice characteristics, making accurate identification difficult with traditional algorithms. The multimodal fusion method of this invention can establish a specialized baseline model for the elderly, creating an intelligent anti-fraud shield for them. This effectively prevents policy fraud and property misappropriation targeting the elderly, reflecting the human touch of financial technology and fulfilling the social responsibility and consumer protection obligations of financial enterprises.

[0055] In summary, this invention elevates Ping An Financial Manager's compliance verification from biometric comparison to a new level of psychological authenticity auditing. It effectively protects the legitimate rights and interests of policyholders, preventing illegal appropriation in scenarios such as family disputes and inheritance battles. By generating tamper-proof electronic evidence, it enhances the legal validity and enforceability of insurance contracts. Furthermore, by establishing dedicated baseline models for special groups such as the elderly, it fulfills the social responsibility and consumer rights protection obligations of financial institutions, setting a new benchmark for compliance technology in the industry.

[0056] like Figure 6 As shown, this embodiment of the invention also provides a device for verifying the authenticity of a policy change intention, comprising: The request receiving module 10 is used to trigger the multimodal physiological feedback and behavioral data collection process after receiving the user's policy change request. The micro-expression analysis module 20 is used to continuously capture the user's facial video stream through an image acquisition device and extract the micro-motion features of key facial points in response to a trigger command during the user's execution of a preset identity verification operation. The acoustic stress analysis module 30 is used to capture the user's voice signal synchronously through an audio acquisition device during the user's reading of a preset legal confirmation text, and to extract acoustic features reflecting the user's psychological stress from the voice signal. The cognitive trap module 40 is used to present one or more preset cognitive trap questions to the user at the final confirmation stage of the user's policy change process, and record the user's feedback behavior data on the cognitive trap questions. The multimodal fusion decision module 50 is used to calculate the comprehensive confidence level representing the authenticity of the user's intention based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, through a preset multimodal fusion verification model. The risk interception module 60 is used to determine that the current policy change request does not represent the user's true intention if the overall confidence level is lower than a preset threshold, and to output corresponding risk warnings and interception instructions.

[0057] In a specific embodiment, the micro-expression analysis module 20 is specifically used for: The facial video stream is analyzed frame by frame to track and locate the coordinates of multiple preset key points on the face; Based on the temporal coordinate changes of the key points, the micro-motion amplitude and frequency of each key point are calculated. The amplitude and frequency of the micro-movements are spatiotemporally encoded to quantify the micro-vibration entropy value of facial muscles under involuntary control. The micro-vibration entropy value is used to characterize the degree of unnatural fluctuations in facial expressions.

[0058] In a specific embodiment, the acoustic pressure analysis module 30 is specifically used for: The speech signal is preprocessed to extract its fundamental frequency trajectory; Based on the fundamental frequency trajectory, calculate the fundamental frequency jitter parameter reflecting the instability of vocal cord vibration and the frequency offset parameter reflecting abnormal changes in pitch. The fundamental frequency jitter parameter and the frequency offset parameter are weighted and fused to generate a stress index for quantifying the psychological stress contained in the user's voice.

[0059] In the cognitive trap module 40, the cognitive trap problems include those unrelated to the current policy change business, which require the user to call long-term memory or perform simple logical reasoning; Cognitive Trap Module 40 is specifically used for: Record the response delay time between when the user receives the cognitive trap question and when they make their first response; Record the accuracy of the answers given by users to the aforementioned cognitive trap questions; The response delay time and the accuracy of the answer are used as the feedback behavior data.

[0060] In a specific embodiment, the policy change intention authenticity verification device further includes a verification model construction unit, used for: An initial multimodal fusion verification model is constructed, and historical business processing data with known intention authenticity labels are used as training samples to train the initial multimodal fusion verification model until the model converges. The training samples include facial micro-motion features, speech acoustic features, feedback behavior data on cognitive trap questions, and corresponding intention authenticity labels of sample users.

[0061] In a specific embodiment, the device for verifying the authenticity of policy change intentions further includes an assessment report generation module, used for: When it is determined that the current policy change request does not represent the user's true intention, a sincerity assessment report is generated and stored. The intention authenticity assessment report includes at least the facial video stream, the voice signal, the calculated overall confidence level, and an analytical summary of the various micro-motion and acoustic features used to indicate non-genuine intention.

[0062] In a specific embodiment, the device for verifying the authenticity of policy change intentions further includes an electronic evidence generation module, used for: All data from this verification process, including the facial video stream, voice signal, feedback behavior data, overall confidence level, and evaluation report, will be packaged into an immutable electronic record and uploaded to a blockchain or designated record platform for subsequent auditing and verification.

[0063] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned policy change intention authenticity verification device and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0064] The aforementioned device for verifying the authenticity of policy change intentions can be implemented as a computer program, which can, for example... Figure 7 It runs on the computer device shown.

[0065] Please see Figure 7 , Figure 7 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a terminal or a server. The terminal can be an electronic device with communication functions, such as a smartphone, tablet, laptop, desktop computer, personal digital assistant, or wearable device. The server can be a standalone server or a server cluster composed of multiple servers.

[0066] See Figure 7 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0067] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. When the computer program 5032 is executed, it causes the processor 502 to execute a method for verifying the authenticity of a policy change intention.

[0068] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0069] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for verifying the authenticity of the intention to change the policy.

[0070] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0071] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps: S1. Upon receiving a user's policy change request, trigger the multimodal physiological feedback and behavioral data collection process; S2. In response to the trigger command, during the user's execution of the preset authentication operation, the user's facial video stream is continuously captured by the image acquisition device, and the microscopic motion features of the facial key points are extracted. S3. During the process of the user reading the preset legal confirmation text, the user's voice signal is captured synchronously through the audio acquisition device, and acoustic features reflecting the user's psychological pressure are extracted from the voice signal. S4. In the final confirmation stage after the user completes the policy change process, present the user with one or more preset cognitive trap questions and record the user's feedback behavior data on the cognitive trap questions. S5. Based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, calculate the comprehensive confidence level representing the authenticity of the user's intention through a preset multimodal fusion verification model; S6. If the overall confidence level is lower than the preset threshold, it is determined that the current policy change request does not represent the user's true intention, and corresponding risk warnings and blocking instructions are output.

[0072] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0073] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program may be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0074] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program. When executed by a processor, the computer program causes the processor to perform the following steps: S1. Upon receiving a user's policy change request, trigger the multimodal physiological feedback and behavioral data collection process; S2. In response to the trigger command, during the user's execution of the preset authentication operation, the user's facial video stream is continuously captured by the image acquisition device, and the microscopic motion features of the facial key points are extracted. S3. During the process of the user reading the preset legal confirmation text, the user's voice signal is captured synchronously through the audio acquisition device, and acoustic features reflecting the user's psychological pressure are extracted from the voice signal. S4. In the final confirmation stage after the user completes the policy change process, present the user with one or more preset cognitive trap questions and record the user's feedback behavior data on the cognitive trap questions. S5. Based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, calculate the comprehensive confidence level representing the authenticity of the user's intention through a preset multimodal fusion verification model; S6. If the overall confidence level is lower than the preset threshold, it is determined that the current policy change request does not represent the user's true intention, and corresponding risk warnings and blocking instructions are output.

[0075] The storage medium is a physical, non-transient storage medium, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk, or any other physical storage medium capable of storing program code.

[0076] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0077] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0078] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0079] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0080] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0081] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Since these modifications and variations fall within the scope of the claims and their equivalents, this invention also intends to include these modifications and variations.

[0082] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0083] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for verifying the authenticity of an intention to change a policy, characterized in that, include: Upon receiving a user's policy change request, a multimodal physiological feedback and behavioral data collection process is triggered. In response to a trigger command, during the user's execution of a preset authentication operation, the system continuously captures the user's facial video stream through an image acquisition device and extracts the microscopic motion features of key facial points. During the process of the user reading the preset legal confirmation text, the user's voice signal is captured synchronously through an audio acquisition device, and acoustic features reflecting the user's psychological stress are extracted from the voice signal. In the final confirmation stage after the user completes the policy change process, one or more preset cognitive trap questions are presented to the user, and the user's feedback behavior data on the cognitive trap questions is recorded. Based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, a comprehensive confidence level representing the authenticity of the user's intention is calculated through a preset multimodal fusion verification model. If the overall confidence level is lower than a preset threshold, it is determined that the current policy change request does not represent the user's true intention, and corresponding risk warnings and blocking instructions are output.

2. The method for verifying the authenticity of policy change intention according to claim 1, characterized in that, The step of extracting the microscopic motion features of key facial points specifically includes: The facial video stream is analyzed frame by frame to track and locate the coordinates of multiple preset key points on the face; Based on the temporal coordinate changes of the key points, the micro-motion amplitude and frequency of each key point are calculated. The amplitude and frequency of the micro-movements are spatiotemporally encoded to quantify the micro-vibration entropy value of facial muscles under involuntary control. The micro-vibration entropy value is used to characterize the degree of unnatural fluctuations in facial expressions.

3. The method for verifying the authenticity of policy change intention according to claim 1, characterized in that, The step of extracting acoustic features reflecting the user's psychological stress from the speech signal specifically includes: The speech signal is preprocessed to extract its fundamental frequency trajectory; Based on the fundamental frequency trajectory, calculate the fundamental frequency jitter parameter reflecting the instability of vocal cord vibration and the frequency offset parameter reflecting abnormal changes in pitch. The fundamental frequency jitter parameter and the frequency offset parameter are weighted and fused to generate a stress index for quantifying the psychological stress contained in the user's voice.

4. The method for verifying the authenticity of policy change intention according to claim 1, characterized in that, The cognitive trap questions include those unrelated to the current policy change business, requiring the user to access long-term memory or perform simple logical reasoning; the step of recording user feedback behavior data on the cognitive trap questions specifically includes: Record the response delay time between when the user receives the cognitive trap question and when they make their first response; Record the accuracy of the answers given by users to the aforementioned cognitive trap questions; The response delay time and the accuracy of the answer are used as the feedback behavior data.

5. The method for verifying the authenticity of policy change intention according to any one of claims 1 to 4, characterized in that, Before the step of calculating the comprehensive confidence level representing the authenticity of the user's intention based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data through a preset multimodal fusion verification model, the method further includes: An initial multimodal fusion verification model is constructed, and historical business processing data with known intention authenticity labels are used as training samples to train the initial multimodal fusion verification model until the model converges. The training samples include facial micro-motion features, speech acoustic features, feedback behavior data on cognitive trap questions, and corresponding intention authenticity labels of sample users.

6. The method for verifying the authenticity of policy change intention according to claim 1, characterized in that, The method further includes: When it is determined that the current policy change request does not represent the user's true intention, a sincerity assessment report is generated and stored. The intention authenticity assessment report includes at least the facial video stream, the voice signal, the calculated overall confidence level, and an analytical summary of the various micro-motion and acoustic features used to indicate non-genuine intention.

7. The method for verifying the authenticity of policy change intention according to claim 1 or 6, characterized in that, Following the step of outputting the corresponding risk warning and blocking instructions, the following is also included: All data from this verification process, including the facial video stream, voice signal, feedback behavior data, overall confidence level, and evaluation report, will be packaged into an immutable electronic record and uploaded to a blockchain or designated record platform for subsequent auditing and verification.

8. A device for verifying the authenticity of a policy change intention, characterized in that, include: The request receiving module is used to trigger the multimodal physiological feedback and behavioral data collection process after receiving a user's policy change request. The micro-expression analysis module is used to respond to trigger commands and continuously capture the user's facial video stream through an image acquisition device during the user's execution of preset identity verification operations, and extract the micro-motion features of key facial points. The acoustic stress analysis module is used to capture the user's voice signal synchronously through an audio acquisition device while the user reads a preset legal confirmation text, and to extract acoustic features reflecting the user's psychological stress from the voice signal. The cognitive trap module is used to present one or more preset cognitive trap questions to the user during the final confirmation stage of the policy change process, and to record the user's feedback behavior data on the cognitive trap questions. The multimodal fusion decision module is used to calculate the comprehensive confidence level representing the authenticity of the user's intention based on the micro-motion features of the facial key points, the acoustic features, and the feedback behavior data, through a preset multimodal fusion verification model. The risk interception module is used to determine that the current policy change request does not represent the user's true intention if the overall confidence level is lower than a preset threshold, and to output corresponding risk warnings and interception instructions.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the method for verifying the authenticity of the policy change intention as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, can implement the method for verifying the authenticity of policy change intentions as described in any one of claims 1 to 7.

Images