Fraud monitoring method and apparatus based on interaction and psychological perception, device and medium

Abstract

The application provides an interaction and psychological perception-based fraud monitoring method, device, equipment and medium, which comprises: collecting rhythm data and fingertip pressure data during user human-computer interaction; calculating comprehensive deviation degree according to the rhythm data, pressure data and historical data and judging the size of the comprehensive deviation degree and the preset risk threshold; if the comprehensive deviation degree is greater than the preset risk threshold, collecting user intention data and voice data based on the preset fraud warning strategy; and generating a fraud monitoring result according to the intention data and voice data. The method can be applied to fraud monitoring scenes such as large-value transfer in financial technology, and the deviation degree of user operation behavior from the normal state is quantified through rhythm data and fingertip pressure data during user interaction; when the comprehensive deviation degree exceeds the threshold, user intention data and voice data are further collected for multidimensional verification to accurately identify whether the user is in a fraud scene, thereby solving the problem of low monitoring accuracy in existing fraud scenes.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and is applied to the field of financial technology. In particular, it relates to a fraud detection method, device, equipment, and medium based on interaction and psychological perception. Background Technology

[0002] With the rapid development of financial technology, financial institutions have established a multi-dimensional identity verification system centered on facial biometrics, dynamic passwords, and SMS verification codes, which effectively prevents traditional financial risks such as account theft and password cracking caused by unauthorized operations.

[0003] In telephone-induced fraud scenarios, criminals impersonate law enforcement, judicial, or bank employees, using intimidation and enticement to manipulate victims into completing money transfers while fully conscious and cooperative. In this situation, traditional technologies such as GPS location verification, device fingerprint recognition, and biometric verification all appear normal. Existing systems cannot penetrate the appearance of "personal operation" to identify the customer's true transaction intent, and cannot distinguish whether the transaction is driven by genuine business needs or is a fraudulent activity under duress or instruction.

[0004] The inability to accurately monitor whether users are being defrauded makes it difficult to effectively intercept fraudulent activities in the process of large-amount transfers, which not only causes significant financial losses to individual customers, but also significantly increases the claims rate and operating costs of financial institutions. Summary of the Invention

[0005] This invention provides a fraud detection method, apparatus, device, and medium based on interaction and psychological perception, aiming to solve the technical problem of low detection accuracy in existing financial transactions to determine whether a user is in a fraudulent scenario.

[0006] In a first aspect, embodiments of the present invention provide a fraud detection method based on interaction and psychological perception, comprising: collecting rhythm data and fingertip pressure data during user human-computer interaction; calculating a comprehensive deviation degree based on the rhythm data, the pressure data, and historical data, and determining the magnitude of the comprehensive deviation degree compared with a preset risk threshold; if the comprehensive deviation degree is greater than the preset risk threshold, collecting user intent data and voice data based on a preset fraud warning strategy; and analyzing the intent data and the voice data to generate fraud detection results.

[0007] Secondly, embodiments of the present invention also provide a fraud detection device based on interaction and psychological perception, which includes a unit for performing the above-described method.

[0008] Thirdly, embodiments of the present invention also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0009] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the above-described method.

[0010] This application provides a fraud detection method, device, equipment, and medium based on interaction and psychological perception. By collecting rhythmic data and fingertip pressure data during user human-computer interaction, it quantifies the degree of deviation between user behavior and normal state. When the overall deviation exceeds a threshold, it further collects user intent data and voice data for multi-dimensional verification to accurately identify whether the user is in a fraudulent state such as being misled. This application breaks through the limitation of traditional identity verification that only verifies whether it is the person in question, achieving a deep perception of the true intent behind the person's actions. It can effectively intercept telephone-induced fraud, significantly reducing customer financial losses and financial institution claims risks. 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 flowchart illustrating the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 2 A schematic diagram of a sub-process of the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 3 A schematic diagram of a sub-process of the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 4 A schematic diagram of a sub-process of the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 5 A schematic diagram of a sub-process of the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 6 A schematic diagram of a sub-process of the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 7A schematic diagram of a sub-process of the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention; Figure 8 A schematic block diagram of a fraud detection device based on interaction and psychological perception provided in an embodiment of the present invention; Figure 9 A schematic block diagram of a computer device provided for 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] Please see Figure 1This is a schematic flowchart illustrating a fraud detection method based on interaction and psychological perception provided in this invention. In this application, the fraud detection method based on interaction and psychological perception is applied to the field of artificial intelligence, particularly in anti-fraud scenarios using electronic channels such as mobile banking and online banking in the fintech field. For example, when a user initiates a large transfer through a mobile banking app, this method can silently collect the user's interaction rhythm and fingertip pressure data throughout the entire human-computer interaction process, including entering the transfer amount, transaction password, or fingerprint verification, and perform real-time deviation analysis based on their historical operating habits. Once an unusual operating pattern is detected, deep intent perception is triggered. By analyzing the user's information entered on the intent confirmation page displayed by the system and the user's voice communication with fraudsters, the system proactively identifies whether the customer is in a fraudulent scenario involving inducement or intimidation, thereby blocking the risk before funds are transferred. Ultimately, the risk control system has been upgraded from "policy identity protection" to "mental health and intent protection." Without increasing the user's additional operational burden, it effectively fills the gap in the existing risk control system's identification of "the person being defrauded" scenarios, improves the accuracy of fraud monitoring, avoids significant financial losses for individual customers, and significantly reduces the claims rate and operating costs of financial institutions.

[0018] This application provides a fraud detection method, device, computer equipment, and storage medium based on interaction and psychological perception. The fraud detection method based on interaction and psychological perception includes: collecting rhythm data and fingertip pressure data during user human-computer interaction; calculating a comprehensive deviation degree based on the rhythm data, the pressure data, and historical data, and determining the magnitude of the comprehensive deviation degree compared with a preset risk threshold; if the comprehensive deviation degree is greater than the preset risk threshold, collecting user intent data and voice data based on a preset fraud warning strategy; and generating fraud detection results based on the intent data and the voice data.

[0019] This application quantifies the degree of deviation between user behavior and normal state by collecting rhythmic and fingertip pressure data during human-computer interaction. When the overall deviation exceeds a threshold, it further collects user intent data and voice data for multi-dimensional verification. By integrating explicit operational intent, implicit physiological intent, and voice feature analysis, it accurately identifies whether a user is in a state of being induced into fraud. This application breaks through the limitation of traditional identity verification that only verifies whether it is the person in question, achieving a deep perception of the true intent behind the person's actions. It can effectively intercept telephone-induced telecommunications fraud, significantly reducing customer property losses and financial institution claims risks.

[0020] Figure 1 This is a flowchart illustrating the fraud detection method based on interaction and psychological perception provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S10-S40.

[0021] S10. Collect rhythm data and fingertip pressure data during user human-computer interaction; Specifically, the rhythm data refers to the time-series data generated when a user performs operations such as clicking, swiping, or long-pressing on the touchscreen while operating the financial transaction application interface. This data may include the interval time of key clicks, the timestamp sequence of swipe trajectories, or the overall rhythm pattern of password input.

[0022] The pressure data refers to the sequence of force values ​​applied by the fingertip to the screen in real time, collected by the device's pressure sensor or gyroscope when the user performs the aforementioned operations on the touchscreen. More specifically, in devices with pressure-sensitive touchscreens, a pressure-sensing layer is integrated beneath the touchscreen, capable of sensing changes in the force applied by the user's finger when pressing the screen and quantifying it into a sequence of pressure values. In ordinary capacitive touchscreens without dedicated pressure-sensing hardware, pressure data can be indirectly estimated by analyzing the contact area in touch events. Specifically, when a user presses the screen with their finger, the contact area between the fingertip and the screen increases with the pressure applied. The system calculates the change in contact area using the major and minor axis parameters of the contact ellipse in the touch event, thereby indirectly estimating the relative pressure value and obtaining the pressure data. Furthermore, the pressure data can also be collected in conjunction with wearable devices. The specific collection method can be set according to the actual usage scenario and is not limited here.

[0023] In practice, when a user enters a high-risk operation page of a financial transaction application, such as transferring funds, making payments, or modifying security settings, the system obtains the raw touch event stream in real time from the device's underlying driver through software interface calls or hardware signal monitoring. The system needs to format the continuous timestamp stream to form rhythmic data with a unified time dimension; simultaneously, it performs discrete sampling and noise reduction preprocessing on the continuous analog signals from the pressure sensor to form fingertip pressure data corresponding one-to-one with the interaction events. This step forms the data foundation for subsequent behavioral analysis.

[0024] This embodiment collects invisible behavioral biometrics, interaction rhythms, and fingertip pressure from users. Compared to traditional static biometrics such as passwords and faces, these features are internalized in the user's operating habits, making them extremely difficult for fraudsters to remotely imitate or hijack. Furthermore, the rhythm data and pressure data can objectively reflect changes in the user's behavior and psychological state. When a user is in a fraudulent scenario where they are being intimidated or misled, their operating rhythm will become disordered, and their fingertip pressure will also fluctuate abnormally. These features provide important evidence for subsequent fraud risk identification from two dimensions (behavioral and psychological).

[0025] S20. Calculate the comprehensive deviation based on the rhythm data, the pressure data, and the historical data, and determine the magnitude of the comprehensive deviation compared to the preset risk threshold. Specifically, the historical data refers to a pre-established behavioral benchmark model for the user, including stable operating patterns exhibited by the user in numerous past normal transactions, such as historical rhythm benchmark values ​​and historical stress benchmark values. The comprehensive deviation is a quantitative indicator used to measure the degree of difference between the user's current operating behavior and normal historical behavior; a larger value indicates a more significant difference and a higher risk of fraud. The preset risk threshold is a pre-set critical value used to determine the existence of fraud risk; the specific value can be adjusted according to actual business needs and risk control strategies.

[0026] In the specific implementation process, the system calculates the comprehensive deviation between the user's current operation and historical operations based on the rhythm data, stress data and historical data, using a preset calculation model from two dimensions: user interaction rhythm and psychological stress state. The comprehensive deviation is then compared with the preset risk threshold to determine whether there is a risk of fraud.

[0027] This embodiment quantifies the difference between a user's current behavior and normal behavior by calculating a comprehensive deviation score, rather than relying on static judgments based on fixed rules. It accurately captures subtle operational distortions caused by drastic changes in a user's psychological state (such as panic or being instructed), providing an initial effective risk score for identifying fraudulent intent under the user's own actions, thus achieving preliminary screening for fraud risk. When the comprehensive deviation score exceeds a preset risk threshold, further verification is triggered, ensuring the accuracy of fraud identification while avoiding excessive interference with legitimate users.

[0028] In one embodiment, reference is made to Figure 2 The historical data includes historical rhythm baseline values ​​and historical pressure baseline values, and step S20 includes steps S21-S24.

[0029] S21. Obtain the user's historical data, which includes historical rhythm baseline values ​​and historical pressure baseline values; S22. Calculate the rhythm deviation value based on the rhythm data and historical rhythm benchmark values; S23. Calculate the pressure deviation value based on the pressure data and historical pressure reference value; S24. Calculate the comprehensive deviation value based on the rhythm deviation value and the pressure deviation value.

[0030] Specifically, the rhythm deviation value measures the degree of difference between the user's current operating rhythm and the normal historical rhythm, such as the overall speed change when entering a password, or the change in the uniformity of key press intervals. The pressure deviation value measures the degree of difference between the user's current fingertip pressure characteristics and the normal historical pressure characteristics, such as the deviation of the pressure mean or the distortion of the pressure waveform. The comprehensive deviation value is a weighted sum of the rhythm deviation value and the pressure deviation value, and the weights can be dynamically adjusted according to different user groups and business scenarios.

[0031] In the specific implementation process, the system first acquires the user's historical data and calculates the user's historical rhythm baseline value and historical pressure baseline value. Then, based on a preset calculation model, it calculates the rhythm deviation value between the current rhythm data and the historical rhythm baseline value, and the pressure deviation value between the current pressure data and the historical pressure baseline value. Finally, the rhythm deviation value and the pressure deviation value are weighted and summed or obtained through other fusion algorithms to obtain the comprehensive deviation degree.

[0032] This embodiment performs refined deviation quantification from two independent dimensions: rhythm and pressure, and then merges them into a comprehensive index. Compared with judgment using only a single dimension, this dual-modal analysis strategy can effectively prevent false alarms caused by a single, accidental change in user operation (such as pressure changes due to hand fatigue, but with a normal rhythm). At the same time, it maintains high sensitivity to fraud scenarios where normal rhythm is disguised but pressure characteristics cannot be controlled. Even if the user is controlled, the rhythm of their physical operation and the pressure signals of their fingertips will betray the fraudster's whereabouts. It is expected to reduce telecommunications fraud claims by more than 15%.

[0033] In one embodiment, reference is made to Figure 3 Step S21 includes steps S211-S212.

[0034] S211. Calculate Shannon entropy based on the rhythm data; S212. Calculate the rhythm deviation value based on the Shannon entropy and the historical rhythm benchmark value.

[0035] Specifically, Shannon entropy refers to the degree of disorder or non-patternedness in a user interaction time series. When a user is in a calm, habitual operating state, their operating rhythm usually follows a certain internalized pattern, and the Shannon entropy value is low and stable. When a user is in a fraudulent state of being guided by telephone instructions, experiencing panic, or hesitating, their operating sequence will exhibit high disorder, such as being fast or slow, and involving scrambled and re-entered inputs, leading to a significant increase in Shannon entropy. The historical rhythm baseline value mentioned in this embodiment is the average Shannon entropy of the user's operating rhythm under normal conditions.

[0036] In the specific implementation process, the system will collect a series of time interval sequences of interactive events in step S10. As input; then the time interval sequence is discretized or its probability distribution is estimated to construct its state space. Next, the Shannon entropy calculation formula is applied. Calculate the real-time entropy value of this time series. . If a user's Shannon entropy is in a low range, it indicates that the user's actions are coherent and conform to the long-established "muscle memory" path; if the user's operation exhibits high entropy characteristics, such as random long pauses and unnatural input rhythm, it suggests that the user is receiving external instruction input, indicating that the user is in a state of being induced or deceived.

[0037] Finally, the system compares the Shannon entropy with the historical rhythm baseline value and calculates the difference between the two, which is the rhythm deviation value; or, the system reads the user's rhythm baseline entropy value from historical data and calculates the absolute difference or ratio between the two as the rhythm deviation value.

[0038] This embodiment introduces Shannon entropy to quantify the disorder of operational rhythms, which can capture the fluctuations of users' psychological states from a deep statistical level. It has stronger anti-interference ability and biological interpretability than methods such as directly comparing simple average speeds.

[0039] In one embodiment, reference is made to Figure 4 Step S22 includes steps S221-S222.

[0040] S221. Extract the pressure waveform features of the pressure data; S222. Calculate the pressure deviation value based on the pressure waveform characteristics and historical pressure reference value.

[0041] Specifically, the pressure waveform characteristics refer to a set of abstract numerical features extracted from the continuous force signal of fingertip pressing, which can characterize the pressing posture and muscle control pattern. Examples include the pressure rise slope, peak duration, and pressure fall slope of a complete click, or the pressure distribution sequence along the sliding trajectory. In this embodiment, the historical pressure benchmark value is the average pressure waveform characteristics when the user performs similar operations under normal conditions.

[0042] In the specific implementation process, the system preprocesses the collected pressure waveform data, including noise reduction and smoothing, and then extracts the pressure waveform features. Users under psychological high pressure (being frightened, extremely tense) will have significant high-frequency harmonic noise generated by the tremor frequency of their fingers when they tap the screen.

[0043] The extracted current pressure waveform features are compared with historical pressure benchmark values, the deviation value of each feature is calculated, and then the overall pressure deviation value is obtained by weighted summation and other methods. If the "involuntary vibration intensity caused by anxiety" exceeds the preset threshold, the user's current psychological load is determined to be overloaded.

[0044] This embodiment extracts multiple feature parameters of the pressure waveform and calculates the comprehensive deviation value, which can comprehensively reflect the abnormal changes in the user's fingertip pressure and improve the accuracy and robustness of the pressure deviation value calculation. At the same time, it can capture the profound impact of involuntary stiffness, trembling or hesitation of the hand muscles when the user is under high tension or being manipulated, which has a profound impact on the profile of the pressure waveform. This level of fine features greatly improves the accuracy of fraud behavior identification.

[0045] S30. If the overall deviation is greater than the preset risk threshold, then collect user intent data and voice data based on the preset fraud warning strategy. Specifically, the preset risk threshold is the critical point that triggers the deep intent perception mechanism, i.e., the risk threshold for initially judging whether a user is in a fraud scenario. When the comprehensive deviation exceeds the standard (the comprehensive deviation is greater than the preset risk threshold), it means that the user's current operational behavior pattern has significantly deviated from the norm. Behavioral deviation alone is insufficient to draw a conclusion, and a higher-level intent confirmation and acoustic quality inspection directly related to the fraudulent content are needed to reconfirm whether the user is in a fraud scenario. In this embodiment, the preset risk threshold can be dynamically adjusted according to the stability of the user's historical behavior. For users with consistently stable behavior patterns, the threshold can be appropriately lowered; for users with large behavioral fluctuations, the threshold can be appropriately increased to balance the false positive rate and the false negative rate.

[0046] The preset fraud warning strategy refers to a series of data collection and risk verification measures taken by the system when a user is detected to be at risk of being defrauded. This embodiment is based on a dual strategy of interactive intent confirmation and forced delay and acoustic quality inspection, aiming to proactively intervene and obtain key evidence for fraud determination.

[0047] The intent data refers to information obtained by displaying a specially designed intent confirmation page on the terminal interface for the user to fill out, thereby acquiring information about the user's current operational motivation. This intent confirmation page employs targeted psychological suggestion design, directly pointing to typical scenarios of telecommunications fraud. For example, it sends questions to the interface such as, "Is this transfer to cooperate with a public security, procuratorate, or judicial investigation?", "Did the other party ask you not to disclose this transfer to others?", and "Have you received any notification of suspected illegal or criminal activity?" If the user selects "yes" to these guiding and discriminatory questions, the system directly obtains the intent data that the user states is in a fraudulent scenario.

[0048] Finally, the system uploads the intent data to the risk control platform for fraud scenario determination.

[0049] This strategy utilizes the principle of psychological suggestion to create a gap in the information cocoon constructed by fraudsters, guiding deeply brainwashed users to reveal themselves. This directly penetrates the appearance of cooperation, achieving rapid and accurate fraud confirmation and intercepting the transaction. It uses psychological suggestion to directly confirm the fraud scenario.

[0050] The voice data refers to the surrounding audio segments collected by the system when the terminal microphone is forcibly activated to capture the user's current call environment. When the overall deviation exceeds the standard, the system can automatically activate the microphone to collect voice data for a preset time period while simultaneously displaying the intent confirmation information page; or, when the system identifies the user as being in a fraudulent scenario based on the intent data, it can also automatically activate the microphone to collect voice data for a preset time period. Here, voice data specifically refers to the audio stream actively collected during the forced delay period, including the user's own voice, background commands from fraudsters, or abnormal environmental sounds. This forced delay strategy not only provides a data collection window for subsequent acoustic analysis but also serves as a cooling-off period, interrupting the continuous rhythm of fraudsters' phone commands and giving the user a moment to clear their head and reflect.

[0051] Finally, the system will upload the collected voice data to the cloud-based risk control platform in real time for acoustic analysis and keyword recognition to further determine whether the user is in a fraudulent scenario.

[0052] This embodiment employs a proactive fraud early warning strategy that combines interactive intent verification with forced-delay acoustic quality inspection, upgrading the defensive stance from passive observation to proactive verification. On one hand, it uses cleverly designed psychological cues to directly obtain information that the user is in a fraudulent scenario; on the other hand, it utilizes a forced-delay strategy to break the fraudsters' control loop from both temporal and informational dimensions, creating data conditions for acoustic quality inspection. This dual proactive verification mechanism directly dismantles the black box of intent under the user's own operation, significantly improving the accuracy and reliability of fraud identification and providing objective evidence for fraud confirmation, thus helping to alleviate claims disputes.

[0053] S40. Analyze the intent data and the voice data to generate fraud detection results.

[0054] Specifically, the fraud monitoring result refers to the final judgment made by the system after comprehensively analyzing the intent data and the voice data regarding whether the user is in a fraudulent scenario, including both normal transactions and the user being in a fraudulent scenario. When the fraud monitoring result indicates that the user is in a fraudulent scenario, the system will immediately trigger corresponding risk handling measures, including but not limited to transaction interception, pop-up warnings, SMS reminders, and human customer service intervention; and generate a warning report based on Shannon entropy, stress waveform characteristics, intent data, and voice data, so that financial institutions can provide clear evidence to customers explaining why their operation was intercepted at the last minute, which helps to alleviate claims disputes.

[0055] In its implementation, the system employs parallel analysis, processing both the intent data and the voice data simultaneously. If either analysis result indicates a fraud risk, the system generates a detection result indicating the user is in a fraudulent scenario. Only when both analysis results are normal will the system generate a detection result indicating a normal transaction, allowing the user to continue completing the transaction.

[0056] This embodiment, serving as the final decision-making stage for fraud detection, employs a multi-dimensional parallel verification mechanism. It comprehensively analyzes interactive intent and voice data, using an OR logic-based judgment rule. If fraudulent features are detected in any dimension, a risk warning is generated, ensuring high-precision fraud detection and achieving a high interception rate. This mechanism overcomes the limitations of traditional identity verification, which only verifies whether the person is the account holder, effectively solving the problem of identifying fraudulent intent under the account holder's own operation, and achieving accurate and rapid interception of deceptive telecommunications fraud.

[0057] In one embodiment, reference is made to Figure 5 Step S40 includes steps S41-S43.

[0058] S41. Based on the preset judgment rules and the intent data, determine whether the user is in a fraud scenario and generate a judgment result; S42. Extract speech features based on the speech data, and analyze whether the speech features are in an abnormal state to generate an abnormal result; S43. If the judgment result indicates a fraud scenario or the abnormal result indicates an anomaly, then a fraud monitoring result indicating that the user has been defrauded is generated.

[0059] Specifically, the preset judgment rules are a set of rules and judgment methods for parsing and judging the interactive feedback based on the intent data collected by the interactive intent confirmation channel in step S30. This rule set predefines an option mapping relationship that corresponds one-to-one with fraud scenarios, clearly identifying which user selections or inputs directly correspond to the user admitting to being in a fraudulent state of being induced or coerced, i.e., the user is in a fraud scenario; for example, if the user selects "This transfer is to cooperate with the public security, procuratorate, and judicial investigation," a judgment result of "in a fraud scenario" is immediately generated.

[0060] The speech features refer to a set of multidimensional acoustic parameters extracted from the collected speech data that reflect the speaker's psychological state, including but not limited to fundamental frequency distribution, pitch fluctuation amplitude, average speech rate, pause interval duration, rate of change of vocal energy, and formant trajectory features. These acoustic features are directly related to the activity of the human autonomic nervous system. When users are in a state of tension, fear, anxiety, or forced to lie, they will exhibit abnormal patterns that they cannot control. The model analyzes these speech features to determine whether the user is in an abnormal state such as extreme tension, fear, or lying, and then performs anomaly identification.

[0061] In practice, the system analyzes all user interactions (intent data) on the specially designed intent confirmation page in real time. If a user selects "yes" to any question pointing to a typical fraud scenario, or enters text containing keywords characteristic of fraud, the system immediately triggers the judgment rules and generates a judgment result indicating that the user is in a fraud scenario. The system does not need to wait for verification from other dimensions and directly determines that there is a fraud risk.

[0062] Simultaneously, the system preprocesses the collected speech data, including endpoint detection, environmental noise suppression, and speaker separation, extracting the user's own speech segments. The preprocessed speech is then input into a pre-trained speech emotion recognition model to extract the aforementioned multi-dimensional acoustic feature vectors. The system compares the extracted feature vectors with the user's personal normal speech baseline model to calculate a speech anomalousness score. When the speech anomalousness score exceeds a preset anomalous threshold, the system generates an abnormal result.

[0063] Finally, the system uses OR logic as the final fraud determination rule. That is, as long as the intent data judgment result indicates that the user is in a fraud scenario, or the voice feature analysis result is abnormal, the system will generate the final fraud monitoring result that the user has been defrauded.

[0064] If a fraud detection result is generated indicating that a user has been defrauded, the system will immediately terminate the current transaction process, generate a standardized fraud detection report, and automatically trigger multi-level risk handling measures, including real-time transaction interception, a full-screen red warning pop-up, emergency SMS reminders, and automatic connection to the anti-fraud customer service hotline.

[0065] If a fraud detection result is generated for a normal transaction, the transaction restriction will be lifted and the user will be allowed to continue the operation.

[0066] This embodiment uses preset fraud scenario mapping rules to quickly analyze user interaction intent feedback and identify fraud signals in milliseconds. By extracting acoustic features such as base frequency, speech rate, and energy distribution, it objectively perceives the tension and fear emotions that users cannot control. Finally, it adopts a conservative judgment based on OR logic. This architecture effectively solves the problem of users being coerced into concealing information, ensuring a high interception rate while controlling the fraud false judgment rate within a reasonable range, thus achieving accurate and rapid interception of inducement fraud.

[0067] Furthermore, this embodiment utilizes an AI assistant for real-time acoustic quality inspection.

[0068] In one embodiment, reference is made to Figure 6 The intent data includes explicit information intent and implicit physiological intent, and step S41 includes steps S411-S413.

[0069] S411. Determine whether the user is in a fraud scenario based on the displayed information intent and the preset intent determination rule, and generate a first determination result; S412. Determine whether the user is in a fraud scenario based on the implicit physiological intent and preset outliers, and generate a second judgment result; S413. If the first judgment result is that the user is in a fraud scenario or the second judgment result is that the user is in a fraud scenario, then a judgment result of being in a fraud scenario is generated.

[0070] Specifically, the intent data can be further divided into explicit information intent and implicit physiological intent. Explicit information intent refers to user intent data that is directly expressed and can be clearly read, such as the user's interactive selection results on the intent confirmation page. Implicit physiological intent refers to physiological and behavioral cues that are unconsciously revealed by the user and can indirectly reflect their psychological state and intent. For example, in multitasking, when urged by fraudulent instructions, a user may frequently and incorrectly switch between a banking app and a chat app instructed by the fraudster; micro-expressions and heart rate variability captured by the device's front-facing camera (under strict compliance and user authorization), or abnormal hand tremors captured by the gyroscope, can all be considered implicit physiological intent data.

[0071] The preset intent determination rule refers to a predefined mapping rule with fraud scenarios, used to map the user's specific interaction feedback to the determination conclusion.

[0072] In practice, users assess fraud scenarios through both explicit and implicit physiological intent channels. First, the system reads the user's interactive selections on the intent confirmation page. This page features several discriminatory options designed around common fraudulent rhetoric, such as "This transfer is to cooperate with a police investigation" or "I am following instructions from someone else over the phone." Pre-defined intent judgment rules map these options as high-risk intent markers. When a user selects any of these options, the system generates a primary judgment that the user is in a fraudulent scenario. This channel directly obtains key evidence for fraud determination from the user's subjective self-report.

[0073] Simultaneously, when the fraud warning strategy is triggered, the system activates the implicit physiological intent acquisition module. This module uses the device's underlying sensor interfaces to acquire accelerometer and gyroscope signals in real time to extract hand tremor characteristics. It also analyzes the short-term variability of pressure sensor data to extract pressure waveform irregularities and obtains application switching frequency from system logs. The system inputs this multi-dimensional data into an anomaly detection model and compares it with preset anomaly values. If the hand tremor amplitude exceeds the normal baseline, the pressure waveform irregularity significantly increases, or the application switching frequency abnormally increases, it indicates that the user is in a highly stressed state, and the second judgment result is generated: the user is in a fraud scenario. This channel provides objective evidence from the perspective of the user's uncontrollable physiological stress response.

[0074] Finally, as long as the explicit intent channel determines that the user admits to being induced, or the implicit physiological intent channel captures an uncontrollable stress physiological response, the system generates a comprehensive judgment result that the user is in a fraud scenario, and transmits it to step S43 for final fraud monitoring result fusion.

[0075] This embodiment constructs a dual-channel intent perception system by subdividing intent data into two independent dimensions: explicit intent and implicit physiological intent. This system is complementary and difficult to bypass simultaneously. Even if fraudsters instruct users to feign explicit intent, implicit physiological signals such as physiological tremors and operational disorder caused by tension and fear will still be captured by the second channel. This effectively penetrates the feigned cooperation of the user, solving the problem in existing technologies where intent cannot be effectively identified when the user is operating the user themselves. This significantly improves the robustness and accuracy of fraud determination and enables precise identification of deep-seated fraud.

[0076] In one embodiment, reference is made to Figure 7 Step S40 also includes steps S44-S46.

[0077] S44. Analyze and extract the voice data to generate keywords; S45. Match the keywords with the preset dynamic inducement word library to generate matching results; S46. If the matching result is higher than the preset matching threshold, a fraud detection result indicating that the user is in a fraud scenario is generated.

[0078] Specifically, the keywords refer to high-risk words extracted from voice data that point to fraud scenarios. The matching result refers to a quantitative score generated by comparing the keywords with a fraud inducement database, used to measure the degree of similarity between the call content and known fraud inducement patterns. The preset matching threshold is a critical value used to determine whether the matching result meets the fraud confirmation standard.

[0079] The preset dynamic inducement word library is a database containing a large number of commonly used inducement words and phrases in fraud scenarios, including but not limited to "public security, procuratorate and court", "safe account", "confidentiality", "don't tell anyone", "verification code", "transfer", "freeze", and "fraud". This database is dynamically updated in real time based on the latest fraud cases to ensure coverage of the latest fraudulent tactics. The matching result refers to the degree of matching between keywords extracted from the speech and words in the dynamic inducement word library, usually expressed by the number and weight of matching keywords.

[0080] In the specific implementation process, the system first preprocesses the collected voice data, including environmental noise suppression, speaker separation, and endpoint detection, to extract clear call voice segments. Then, it converts the voice into text using automatic speech recognition technology, segments the text using word segmentation algorithms, and extracts semantic keywords based on a pre-trained natural language processing model. It focuses on selecting several words such as nouns, verbs, and phrases related to fraud as keywords, and also extracts named entities such as amounts and accounts involved, and includes them in the keyword set.

[0081] Then, all keywords are matched with a dynamic inducement word library, using a composite scoring strategy that includes exact matching, synonym matching, and word vector semantic similarity matching, and the matching results are generated by weighted summation.

[0082] The matching result is then compared with a preset matching threshold. When the matching result is higher than the preset matching threshold, it indicates that there is a sufficient density of misleading instructions in the call content. The system then directly generates a fraud detection result and triggers risk blocking.

[0083] Furthermore, in this embodiment, the keyword matching channel runs in parallel with the intent judgment and acoustic emotion analysis channels, and blocking is triggered upon confirmation of any one of the channels.

[0084] This embodiment utilizes keyword matching and a dynamic inducement word database mechanism to anchor anti-fraud judgments from vague emotional anomalies to specific inducement instructions, providing objective and interpretable blocking evidence. The dynamic update mechanism ensures rapid response to new rhetoric, forming a multi-level verification network together with explicit intent, implicit physiological, and acoustic emotional channels. This effectively solves the problems of missed and false positives in single-dimensional detection, improving the accuracy of fraud monitoring.

[0085] This application quantifies the degree of deviation between user behavior and normal state by collecting rhythmic and fingertip pressure data during user interaction. When the deviation exceeds a threshold, it further collects user intent data and voice data for multi-dimensional verification. By integrating explicit operational intent, implicit physiological intent, and voice feature analysis, it accurately identifies whether a user is in a state of being induced into fraud. This application breaks through the limitation of traditional identity verification that only verifies whether it is the person in question, achieving a deep perception of the true intent behind the person's actions. It can effectively intercept telephone-induced telecommunications fraud, significantly reduce customer property losses and financial institution claims risks, and is applicable to transaction risk control in various financial scenarios such as banks and payment institutions.

[0086] Figure 8 This is a schematic block diagram of a fraud detection device 300 based on interaction and psychological perception provided in an embodiment of the present invention. Figure 8 As shown, corresponding to the above-described fraud detection method based on interaction and psychological perception, the present invention also provides a fraud detection device 300 based on interaction and psychological perception. This fraud detection device 300 includes a unit for performing the above-described fraud detection method based on interaction and psychological perception, and the device can be configured in a computer device. Specifically, please refer to... Figure 8 The fraud monitoring device 300 based on interaction and psychological perception includes a data acquisition unit 301, a judgment unit 302, a data collection unit 303, and a data generation unit 304.

[0087] The acquisition unit 301 is used to collect rhythmic data and fingertip pressure data during user human-computer interaction. The judgment unit 302 is used to calculate the comprehensive deviation degree based on the rhythm data, the pressure data and historical data, and to determine the magnitude of the comprehensive deviation degree and the preset risk threshold. The collection unit 303 is used to collect user intent data and voice data based on a preset fraud warning strategy if the overall deviation is greater than the preset risk threshold. The generation unit 304 is used to analyze the intent data and the voice data to generate fraud detection results.

[0088] In one embodiment, the determination unit 302 includes an acquisition unit, a first calculation unit, a second calculation unit, and a third calculation unit.

[0089] The acquisition unit is used to acquire the user's historical data, which includes historical rhythm baseline values ​​and historical pressure baseline values. The first calculation unit is used to calculate the rhythm deviation value based on the rhythm data and the historical rhythm benchmark value; The second calculation unit is used to calculate the pressure deviation value based on the pressure data and the historical pressure reference value. The third calculation unit is used to calculate the comprehensive deviation value based on the rhythm deviation value and the pressure deviation value.

[0090] In one embodiment, the first computing unit includes an entropy computing unit and a fourth computing unit.

[0091] An entropy calculation unit is used to calculate Shannon entropy based on the rhythm data; The fourth calculation unit is used to calculate the rhythm deviation value based on the Shannon entropy and the historical rhythm benchmark value.

[0092] In one embodiment, the second computing unit includes a feature extraction unit and a fifth computing unit.

[0093] The feature extraction unit is used to extract the pressure waveform features of the pressure data; The fifth calculation unit is used to calculate the pressure deviation value based on the pressure waveform characteristics and historical pressure reference values.

[0094] In one embodiment, the generation unit 304 includes a first judgment unit, a speech analysis unit, and a first generation unit.

[0095] The first judgment unit is used to determine whether the user is in a fraud scenario based on preset judgment rules and the intent data, and to generate a judgment result. The speech analysis unit is used to extract speech features based on the speech data and analyze whether the speech features are in an abnormal state to generate an abnormal result; The first generation unit is used to generate a fraud monitoring result indicating that the user has been defrauded if the judgment result indicates that the user is in a fraud scenario or the abnormal result is abnormal.

[0096] In one embodiment, the generation unit 304 further includes a keyword extraction unit, a matching unit, and a second generation unit.

[0097] The keyword extraction unit is used to analyze and extract keywords from the voice data to generate keywords; The matching unit is used to match the keywords with a preset dynamic inducement word library to generate matching results; The second generation unit is used to generate a fraud detection result indicating that the user is in a fraud scenario if the matching result is higher than a preset matching threshold.

[0098] In one embodiment, the first determination unit includes a second determination unit, a third determination unit, and a third generation unit.

[0099] The second judgment unit is used to determine whether the user is in a fraud scenario based on the displayed information intent and the preset intent judgment rule, and to generate a first judgment result. The third judgment unit is used to determine whether the user is in a fraud scenario based on the implicit physiological intention and preset outlier values ​​and to generate a second judgment result. The third generation unit is used to generate a judgment result indicating that the user is in a fraud scenario if the first judgment result indicates that the user is in a fraud scenario or the second judgment result indicates that the user is in a fraud scenario.

[0100] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned fraud detection device based on interaction and psychological perception 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.

[0101] The aforementioned fraud detection device 300 based on interaction and psychological perception can be implemented as a computer program, which can, for example... Figure 9 It runs on the computer device shown.

[0102] Please see Figure 9 , Figure 9 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.

[0103] See Figure 9 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.

[0104] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a fraud detection method based on interaction and psychological perception.

[0105] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0106] 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 fraud detection method based on interaction and psychological perception.

[0107] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 9 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.

[0108] The processor 502 is used to run a computer program 5032 stored in the memory to implement the steps of the above-described fraud detection method based on interaction and psychological perception.

[0109] 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.

[0110] 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 includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0111] 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, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the steps of the above-described fraud detection method based on interaction and psychological perception.

[0112] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0113] 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.

[0114] 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.

[0115] 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.

[0116] 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.

[0117] 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.

[0118] 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 fraud detection method based on interaction and psychological perception, characterized in that, The method includes: Collect rhythmic data and fingertip pressure data during user human-computer interaction; The comprehensive deviation is calculated based on the rhythm data, the pressure data, and the historical data, and the magnitude of the comprehensive deviation is determined relative to the preset risk threshold. If the overall deviation is greater than the preset risk threshold, user intent data and voice data are collected based on the preset fraud warning strategy. The intent data and the voice data are analyzed to generate fraud detection results.

2. The method according to claim 1, characterized in that, The step of calculating the comprehensive deviation based on the rhythm data, the pressure data, and historical data, and determining the magnitude of the comprehensive deviation relative to a preset risk threshold, includes: Obtain the user's historical data, which includes historical rhythm baseline values ​​and historical pressure baseline values; Calculate the rhythm deviation value based on the rhythm data and historical rhythm baseline values; Calculate the pressure deviation value based on the pressure data and historical pressure benchmark values; The combined deviation value is calculated based on the rhythm deviation value and the pressure deviation value.

3. The method according to claim 2, characterized in that, The step of calculating the rhythm deviation value based on the rhythm data and historical rhythm baseline values ​​includes: Calculate Shannon entropy based on the rhythm data; The rhythm deviation value is calculated based on the Shannon entropy and the historical rhythm baseline value.

4. The method according to claim 2, characterized in that, The step of calculating the pressure deviation value based on the pressure data and historical pressure benchmark values ​​includes: Extract the pressure waveform features from the pressure data; The pressure deviation value is calculated based on the pressure waveform characteristics and historical pressure reference values.

5. The method according to claim 1, characterized in that, The step of analyzing the intent data and the voice data to generate fraud detection results includes: Based on preset judgment rules and the intent data, determine whether the user is in a fraud scenario and generate a judgment result; Speech features are extracted from the speech data, and the speech features are analyzed to determine whether they are in an abnormal state in order to generate abnormal results. If the judgment result indicates a fraud scenario or the abnormal result is abnormal, then a fraud monitoring result indicating that the user has been defrauded is generated.

6. The method according to claim 5, characterized in that, The intent data includes explicit information intent and implicit physiological intent. The step of determining whether a user is in a fraud scenario based on preset judgment rules and the intent data and generating a judgment result further includes: Based on the displayed information intent and the preset intent determination rules, determine whether the user is in a fraud scenario and generate a first determination result; Based on the implicit physiological intent and preset outliers, determine whether the user is in a fraud scenario and generate a second judgment result; If the first judgment result is that the user is in a fraud scenario or the second judgment result is that the user is in a fraud scenario, then a judgment result indicating that the user is in a fraud scenario is generated.

7. The method according to claim 5, characterized in that, The step of analyzing the intent data and the voice data to generate fraud detection results further includes: The voice data is analyzed and extracted to generate keywords; Matching results are generated based on the keywords and a preset dynamic inducement word library; If the matching result is higher than the preset matching threshold, a fraud detection result is generated indicating that the user is in a fraud scenario.

8. A fraud detection device based on interaction and psychological perception, characterized in that, Includes a unit for performing the method as described in any one of claims 1-7.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the method as described in any one of claims 1-7.

Images