Call-out system and method based on artificial intelligence

Abstract

The invention relates to the technical field of mobile data communication, and particularly discloses an outbound system and method based on artificial intelligence, and the system comprises a multi-source data collection module which is integrated in a smart phone terminal and is used for collecting a user face image, speaking voice, ambient light intensity, a background noise decibel value and terminal real-time positioning data; the edge processing module is used for carrying out standardized preprocessing on the collected data and uploading the preprocessed data to the cloud server; by integrating the multi-source data acquisition module, related information of a user can be comprehensively acquired, the edge processing module is utilized to preprocess the acquired data, the response speed and processing efficiency of the system are improved, the cloud server performs biological feature verification, behavior feature verification and comprehensive risk analysis on the preprocessed data, and the safety of the system is improved. And a personalized response strategy is generated according to an analysis result, so that the accuracy and safety of identity verification can be remarkably improved, and the transaction risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile data communication, and specifically to an outbound call system and method based on artificial intelligence. Background Art

[0002] In the current wave of digital transformation, the outbound call system, as the core carrier of intelligent communication, has been deeply integrated into various fields of the national economy. In the field of financial credit, it supports business such as daily loan reviews and overdue reminders of tens of millions. For example, a certain commercial bank uses the outbound call system to increase the credit card activation rate by 40%. In the marketing scenario, the precision marketing outbound call system helps an e-commerce platform gain over 20 million new customers annually, with the marketing cost reduced by 35%. In the field of public services, the government affairs hotline outbound call system processes over 500,000 social security consultations, vaccine notifications and other affairs daily.

[0003] Traditional outbound call systems mainly rely on fixed conversation templates and simple voice interactions, and can only complete basic information transmission and preliminary customer communication. Their core functions are usually limited to making calls, playing preset voice content, and simple button interactions, and cannot be flexibly adjusted according to the real-time status and environment of customers. As a result, traditional outbound call systems mainly rely on methods such as entering passwords and answering preset questions for identity verification. However, these methods are easily stolen or cracked by others, posing a significant security risk. For example, in financial credit operations, criminals may maliciously apply for loans by obtaining customers' account passwords, causing huge losses to financial institutions and customers. Summary of the Invention

[0004] The purpose of the present invention is to provide an outbound call system and method based on artificial intelligence, and solve the following technical problems:

[0005] How to achieve high security in identity verification, accurate risk identification, and intelligent response strategies during the outbound call process through multi-dimensional biometric verification, real-time behavior analysis, and dynamic risk assessment.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An outbound call system based on artificial intelligence, the system includes:

[0008] A multi-source data collection module, integrated in a smart phone terminal, for collecting user facial images, speaking voices, ambient light intensity, background noise decibel values, and terminal real-time positioning data;

[0009] An edge processing module, for performing standardized preprocessing on the collected data and uploading the preprocessed data to a cloud server;

[0010] A cloud server is used to sequentially perform biometric verification and behavioral feature verification on the received preprocessed data, and conduct comprehensive risk analysis by combining environmental feature parameters and user credit data, and generate corresponding response strategies according to the analysis results.

[0011] Further, the process of the edge processing module preprocessing the collected information includes:

[0012] Perform grayscale conversion and histogram equalization processing on the user's facial image in sequence. After locating the key facial feature points, normalize the pupil distance to a preset pixel value;

[0013] Perform pre-emphasis filtering, frame addition, and windowing processing on the spoken voice, extract the voice segments with a fundamental frequency exceeding a preset threshold, and remove the silent segments through endpoint detection;

[0014] Perform high-frequency noise desensitization processing on the environmental light intensity and background noise decibel value to remove random interference;

[0015] Convert the terminal positioning coordinates into a Geohash encoded string of 12 characters.

[0016] Further, the biometric verification process includes:

[0017] Input the preprocessed user facial image and the user's ID card photo into the trained face recognition model to extract the user facial image feature vector Q s and the user ID card photo feature vector Q std ;

[0018] Through the formula Analyze and calculate to obtain the face similarity R face ;

[0019] If R face > R1, it is determined that the face recognition is passed;

[0020] Otherwise, it is determined that the face recognition is not passed and the transaction is terminated;

[0021] Among them, R1 is a preset face similarity threshold.

[0022] Further, the biometric verification process also includes:

[0023] Input the preprocessed voice and the pre-stored user voiceprint template into the trained voiceprint recognition model to extract the user voice feature vector W s and the user voiceprint template feature vector W std ;

[0024] Through the formula Analyze and calculate to obtain the voiceprint similarity R voice ;

[0025] If R voice >R2, determine that the voiceprint recognition has passed;

[0026] Otherwise, it is determined that the voiceprint recognition has failed and further analysis is required;

[0027] Among them, R2 is the preset voiceprint similarity threshold;

[0028] Further analysis process:

[0029] By formula R all =R face *ω1+R voice *ω2Analysis and calculation to obtain the biometric similarity R all ;

[0030] If R all >R3, determine that the biometric verification has passed;

[0031] Otherwise, if the biometric verification is judged to be unsuccessful, the transaction will be stopped;

[0032] Among them, R3 is the preset biometric similarity threshold.

[0033] Furthermore, the behavior feature verification process includes:

[0034]

[0035] The behavioral characteristic abnormality coefficient E is obtained by combining the above formulas for simultaneous analysis and calculation abn ;

[0036] Among them, A abn is the speech rate abnormality coefficient, P abn is the interaction anomaly coefficient, is the weight coefficient corresponding to the influencing factors of abnormal behavior, V s is the user's current speaking speed, V std is the preset speaking speed, N is the number of times the user clicks the mobile terminal, T is the duration of the outbound call session, θ is the deflection angle of the mobile terminal during the session, and a is the adjustment parameter;

[0037] If E abn >E std , if the behavior feature verification fails, the transaction is stopped;

[0038] Otherwise, it is determined that the behavior characteristics verification has passed;

[0039] Among them, E std It is the abnormal critical coefficient of preset behavioral characteristics.

[0040] Furthermore, the process of comprehensive risk analysis includes:

[0041] Drisk =[(1-R all )*τ1+E abn *τ2+C s *τ3+b]*H risk

[0042]

[0043] The comprehensive risk characteristic value D is obtained by combining the above formulas for analysis and calculation risk ;

[0044] Among them, C s is the environmental risk factor, H risk is the user's bad credit evaluation parameter, Z s is the actual positioning deviation value of the mobile terminal, b is the credit impact adjustment parameter, τ1, τ2, τ3 are the corresponding weight coefficients of the comprehensive risk influencing factors, τ1>τ2>τ3, Z std is the preset positioning deviation control value, K s is the actual noise background decibel value, K std is the preset noise background control decibel value, θ1 and θ2 are the corresponding weight coefficients of environmental risk influencing factors;

[0045] The comprehensive risk characteristic value D risk Compare with the preset risk response threshold range (d1, D2]);

[0046] If D risk ≤D1, the comprehensive risk analysis is normal and trading is allowed;

[0047] If D1 <D risk <D2,则综合风险分析待定,需转人工进行风险审核;

[0048] If D risk >D2, the comprehensive risk analysis is abnormal and trading is stopped.

[0049] Furthermore, the system further comprises:

[0050] The cloud server regularly analyzes the historical outbound call data of the manual review results within the preset period and adjusts the preset risk response threshold range;

[0051]

[0052] The stop transaction rate α in the manual review results within the preset period is obtained by analyzing and calculating the above formula;

[0053] Among them, Risk is the number of suspended transactions in the manual review results within the preset period, and All is the total number of manual reviews within the preset period;

[0054] Compare the stop trading rate α in the manual review result with the preset stop trading rate threshold range [α min , α max ;

[0055] If α < α min , adjust the preset risk response threshold range, D 1,new = D1 + (α min - α) * ρ, D 2,new = D2 + (α min - α) * ρ;

[0056] If α > α max , adjust the preset risk response threshold range, D 1,new = D1 - (α - α max ) * ρ, D 2,new = D2 - (α - α max ) * ρ;

[0057] If α ∈ [α min , α max , do not adjust the preset risk response threshold range, D 1,new = D1, D 2,new = D2;

[0058] Wherein, D 1,new is the new threshold lower limit in the preset risk response threshold range, and D 2,new is the new threshold upper limit in the preset risk response threshold range.

[0059] An outbound call method based on artificial intelligence, the method is used for an outbound call system based on artificial intelligence, and the method includes:

[0060] S1. The cloud server establishes a communication connection with the smart phone terminal, and collects the user's facial image, speech, ambient light intensity, background noise decibel value and terminal positioning data through the multi-source data collection module;

[0061] S2. The edge processing module preprocesses the collected data in sequence and uploads the preprocessed data to the cloud server;

[0062] S3. The cloud server sequentially performs biometric verification and behavior feature verification on the received preprocessed data and generates corresponding response strategies;

[0063] S4. The cloud server combines the verification data, environmental features and user credit data obtained in step S3 to perform comprehensive risk analysis and generate corresponding response strategies;

[0064] S5. Analyze the historical outbound call data of the manual review results within a preset period through the cloud server, and adjust the preset risk response threshold range.

[0065] Advantages of the present invention:

[0066] (1) By integrating a multi-source data acquisition module, the present invention can comprehensively collect information such as the user's facial image, speech, ambient light intensity, background noise decibel value, and terminal positioning data, providing richer and more accurate data support for subsequent identity verification and risk assessment. Using the edge processing module to preprocess the collected data can effectively reduce the data transmission volume and the computing burden of the cloud server, improve the system's response speed and processing efficiency. The cloud server conducts biometric verification, behavioral feature verification, and comprehensive risk analysis on the preprocessed data, and generates personalized response strategies according to the analysis results, which can significantly improve the accuracy and security of identity verification, reduce transaction risks, and provide users with higher-quality and more efficient services. At the same time, the system also has the function of dynamically adjusting the preset risk response threshold range, which can be adaptively adjusted according to the historical outbound call data of the manual review results, further improving the adaptability and flexibility of the system. Description of the Drawings

[0067] The present invention will be further described below with reference to the accompanying drawings.

[0068] Figure 1 is a schematic block diagram of an outbound call system based on artificial intelligence proposed by the present invention;

[0069] Figure 2 is a flowchart of the steps of an outbound call method based on artificial intelligence proposed by the present invention. Detailed Embodiments

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0071] Please refer to Figure 1 as shown. In one embodiment, an outbound call system based on artificial intelligence is provided, and the system includes:

[0072] The multi-source data acquisition module, integrated in the smart phone terminal, is used to collect the user's facial image, speech, ambient light intensity, background noise decibel value and the terminal's real-time positioning data. In practical applications, the smart phone terminal can use its built-in camera to collect the user's facial image, the microphone to collect the speech, the light sensor to obtain the ambient light intensity, the sound sensor to measure the background noise decibel value, and the GPS module to achieve the terminal's real-time positioning;

[0073] The edge processing module is used to perform standardized preprocessing on the collected data and upload the preprocessed data to the cloud server;

[0074] The cloud server is used to sequentially perform biometric verification and behavioral feature verification on the received preprocessed data, and perform comprehensive risk analysis by combining environmental feature parameters and user credit data, and generate corresponding response strategies according to the analysis results.

[0075] The process of the edge processing module preprocessing the collected information includes:

[0076] Perform grayscale conversion and histogram equalization processing on the collected user facial image in sequence. Grayscale conversion can convert a color image into a grayscale image, reducing the amount of data and facilitating subsequent processing; Histogram equalization processing can enhance the contrast of the image. After completing the above processing, locate the key facial feature points, such as the positions of the eyes, nose, mouth, etc., and then normalize the pupil distance to a preset pixel value, which can eliminate the influence of different user facial sizes on subsequent face recognition. For example, the preset pixel value can be set to 50 pixels;

[0077] Perform pre-emphasis filtering on the speech to enhance the high-frequency components of the speech signal; then perform frame segmentation and windowing processing, divide the speech signal into several frames, and add a window to reduce spectral leakage. After that, extract the speech segments with a fundamental frequency exceeding the preset threshold, and remove the silent segments through endpoint detection, only retaining the part with actual speech information;

[0078] Perform high-frequency noise desensitization processing on the ambient light intensity and background noise decibel value to remove random interference. A filtering algorithm, such as a low-pass filter, can be used to remove high-frequency noise;

[0079] Convert the terminal positioning coordinates into a 12-bit character Geohash encoded string. Geohash encoding can convert two-dimensional geographical coordinates into a one-dimensional string, which is convenient for storage and comparison.

[0080] Through the above technical solutions, this embodiment provides an outbound call system based on artificial intelligence. By integrating a multi-source data acquisition module, the system can comprehensively collect information such as the user's facial image, speech, ambient light intensity, background noise decibel value, and terminal positioning data, providing richer and more accurate data support for subsequent identity verification and risk assessment. Using the edge processing module to preprocess the collected data can effectively reduce the data transmission volume and the computing burden on the cloud server, improving the system's response speed and processing efficiency. The cloud server conducts biometric verification, behavioral feature verification, and comprehensive risk analysis on the preprocessed data, and generates a personalized response strategy based on the analysis results, which can significantly improve the accuracy and security of identity verification, reduce transaction risks, and provide users with better and more efficient services.

[0081] In one embodiment, the biometric verification process includes:

[0082] Input the preprocessed user facial image and the user's ID card photo into the trained face recognition model to extract the user facial image feature vector Q s and the user ID card photo feature vector Q std ;

[0083] Analyze and calculate through the formula to obtain the face similarity R face ;

[0084] If R face > R1, it is determined that the face recognition is passed;

[0085] Conversely, it is determined that the face recognition fails and the transaction is terminated;

[0086] wherein, R1 is a preset face similarity threshold, obtained by presetting according to experience, and generally ranges from 0.8 to 0.9, used to determine whether the face recognition is passed, ||Q s || is the modulus of the user facial image feature vector, ||Q std || is the modulus of the user ID card photo feature vector.

[0087] Through the above technical solution, this embodiment provides a face recognition method. The method inputs the preprocessed user facial image and ID card photo into the trained face recognition model, calculates the face similarity and compares it with a preset threshold. This method has the effect of accurately verifying the user's identity. Among them, the process of obtaining the trained face recognition model: first, collect a large number of face images of different ages, genders, races, expressions, illuminations, and postures, then perform preprocessing such as detection, cropping, and normalization on the images, select CNN architecture models such as FaceNet and ArcFace, define appropriate loss functions such as triplet loss, divide the data into training set, validation set, and test set, use the optimization algorithm to iteratively train the model, evaluate with the test set after training, if the performance is not good, adjust the architecture, data, or parameters to optimize, and finally save the qualified model.

[0088] In one embodiment, the biometric verification process further includes:

[0089] Input the preprocessed voice and the pre-stored user voiceprint template into the trained voiceprint recognition model to extract the user voice feature vector W s and the user voiceprint template feature vector W std ;

[0090] Calculate and obtain the voiceprint similarity R through the formula ; voice ;

[0091] If R voice > R2, it is determined that the voiceprint recognition passes;

[0092] On the contrary, it is determined that the voiceprint recognition fails and further analysis is required;

[0093] Among them, R2 is the preset voiceprint similarity threshold, which is preset according to experience, and generally ranges from 0.7 to 0.8. ||W s || is the norm of the user voice feature vector, and ||W std || is the norm of the user voiceprint template feature vector;

[0094] The further analysis process:

[0095] Calculate and obtain the biometric similarity R through the formula R all = R face * ω1 + R voice * ω2; all ;

[0096] If R all > R3, it is determined that the biometric verification passes;

[0097] On the contrary, it is determined that the biometric verification fails and the transaction is stopped;

[0098] Among them, R3 is the preset biometric similarity threshold, which is obtained based on experience, and the general value range is 0.75-0.85. ω1 and ω2 are the weight coefficients corresponding to the factors affecting the similarity of voiceprint features, which are obtained based on experience, and the general value range of ω1 is 0.6-0.8, and the value range of ω2 is 0.2-0.4.

[0099] Through the above technical solution, this embodiment provides a method for voiceprint recognition and comprehensive biometric verification. The method inputs the preprocessed voice and the pre-stored voiceprint template into the trained voiceprint recognition model, calculates the voiceprint similarity and the biometric similarity and compares them with the preset threshold. This method has the effect of further verifying the user identity and improving the verification accuracy. The process of training the voiceprint recognition model includes: collecting voice data of different users and multiple speaking styles and performing data enhancement, framing the voice, extracting features such as MFCC, endpoint detection, selecting model architectures such as GMM-UBM and DNN, defining cross entropy loss, etc., iteratively training with an optimization algorithm after dividing the data set, evaluating with a test set, measuring with indicators such as error rate, optimizing the model when the performance is insufficient, and finally saving a qualified model.

[0100] In one embodiment, the behavior feature verification process includes:

[0101]

[0102]

[0103] The behavioral characteristic abnormality coefficient E is obtained by combining the above formulas for simultaneous analysis and calculation abn ;

[0104] Among them, A abn is the speech rate abnormality coefficient, P abn is the interaction anomaly coefficient, is the weight coefficient corresponding to the influencing factors of abnormal behavior, which is preset based on experience. Generally, The value range is 0.5-0.7, The value range is 0.3-0.5, V s is the user’s current speaking speed, which can be obtained by counting the number of words spoken by the user per unit time through speech recognition technology. std is the preset speaking speed, which is obtained by preset experience and generally ranges from 120 to 150 words per minute. N is the number of times the user clicks the mobile terminal, which is obtained through the event record statistics of the mobile terminal. T is the duration of the outbound call session, which is obtained by timing the system clock. θ is the deflection angle of the mobile terminal during the session, which is measured by the gyroscope sensor of the mobile terminal. a is an adjustment parameter, which is obtained by preset experience and generally ranges from 10 to 20.

[0105] If E abn >E std , if the behavior feature verification fails, the transaction is stopped;

[0106] Otherwise, it is determined that the behavior characteristics verification has passed;

[0107] Among them, E std It is the preset critical coefficient of abnormal behavior characteristics, which is obtained based on experience and generally ranges from 0.3 to 0.5.

[0108] Through the above technical scheme, this embodiment provides a method for verifying behavioral characteristics. The method calculates the speech rate abnormality coefficient and the interaction abnormality coefficient, and combines the weight coefficient to obtain the behavioral characteristic abnormality coefficient, and compares it with the preset threshold. This method has the effect of detecting whether the user behavior is abnormal.

[0109] In one embodiment, the comprehensive risk analysis process includes:

[0110] D risk =[(1-R all )*τ1+E abn *τ2+C s *τ3+b]*H risk

[0111]

[0112] The comprehensive risk characteristic value D is obtained by combining the above formulas for analysis and calculation risk ;

[0113] Among them, C s is the environmental risk factor, H risk Z is the user's bad credit evaluation parameter, which can be obtained from the credit reporting agency. The value range is 0-1. The larger the value, the worse the credit. s is the actual positioning deviation value of the mobile terminal, which is obtained by comparing the real-time positioning and historical positioning data of the mobile terminal. b is the credit impact adjustment parameter, which is obtained by experience and is used to balance the weight of credit data in the calculation of the comprehensive risk characteristic value, reflecting the influence of credit information on the final risk assessment result. τ1, τ2, τ3 are the corresponding weight coefficients of the comprehensive risk influencing factors, which are obtained by experience and preset. τ1>τ2>τ3, Z std K is the preset positioning deviation control value, which is obtained based on experience and generally ranges from 3km to 5km. s is the actual noise background decibel value, measured by the sound sensor of the mobile terminal, K stdIt is the preset noise background control decibel value, obtained by empirical preset, and the general value range is 50 - 70 decibels. θ1 and θ2 are the corresponding weight coefficients of environmental risk influencing factors, obtained by empirical preset. Generally, the value range of θ1 is 0.5 - 0.7, and the value range of θ2 is 0.3 - 0.5;

[0114] Compare the comprehensive risk eigenvalue D risk with the preset risk response threshold range (D1, D2], and the preset risk response threshold range (D1, D2] is used to judge the result of the comprehensive risk analysis;

[0115] If D risk ≤ D1, the comprehensive risk analysis is normal and trading is allowed;

[0116] If D1 < D risk < D2, the comprehensive risk analysis is pending and manual risk review is required;

[0117] If D risk > D2, the comprehensive risk analysis is abnormal and trading is stopped.

[0118] Through the above technical solution, this embodiment provides a method for comprehensive risk analysis. The method calculates the comprehensive risk eigenvalue by comprehensively considering the environmental risk coefficient, the user's bad credit evaluation parameter, and the actual positioning deviation value of the mobile terminal, and compares it with the preset threshold range. This method has the effect of comprehensively evaluating the trading risk.

[0119] In one embodiment, the system further includes:

[0120] The cloud server regularly analyzes the historical outbound call data of the manual review results within a preset period to adjust the preset risk response threshold range;

[0121]

[0122] Calculate the stop trading rate α in the manual review results within a preset period through the above formula analysis;

[0123] Among them, Risk is the number of stop trades in the manual review results within a preset period, statistically obtained from the manual review records, and All is the total number of manual reviews within a preset period, statistically obtained from the manual review records. The preset period can be obtained by empirical preset, and the general value range is 7 days - 30 days. The frequency of regular analysis is generally synchronized with the preset period. For example, if the preset period is 7 days, the frequency of regular analysis is generally set to analyze once every 7 days;

[0124] Compare the stop trading rate α in the manual review results with the preset stop trading rate threshold range [α min , α maxCompare, the preset stop trading rate threshold range [α min , α max can be preset and obtained according to industry big data analysis;

[0125] If α < α min , it indicates that the current risk control intensity is too strict, and it is necessary to increase the left endpoint value of the preset risk response threshold and adjust the preset risk response threshold range, D 1,new = D1 + (α min - α) * ρ, D 2,new = D2 + (α min - α) * ρ;

[0126] If α > α max , it indicates that the current risk control intensity is too loose, and it is necessary to decrease the left endpoint value of the preset risk response threshold and adjust the preset risk response threshold range, D 1,new = D1 - (α - α max ) * ρ, D 2,new = D2 - (α - α max ) * ρ;

[0127] If α ∈ [α min , α max , it indicates that the current risk control intensity is reasonable and no adjustment is made to the preset risk response threshold range, D 1,new = D1, D 2,new = D2;

[0128] Among them, D 1,new is the new threshold lower limit in the preset risk response threshold range, D 2,new is the new threshold upper limit in the preset risk response threshold range, and the ρ is the risk response threshold range adjustment coefficient, which can be preset and obtained according to experience.

[0129] Through the above technical solution, this embodiment provides a method for adjusting the preset risk response threshold range. The method adjusts the preset risk response threshold range by analyzing the stop trading rate in the manual review results within a preset period and comparing it with the preset threshold range. This method has the effect of adaptively optimizing the risk assessment standard.

[0130] Please refer to Figure 2 As shown, in one embodiment, an outbound call method based on artificial intelligence is provided. The method is characterized in that the method is used for an outbound call system based on artificial intelligence, and the method includes:

[0131] S1. The cloud server establishes a communication connection with the smart phone terminal through a security protocol, and collects the user's facial image, speech, ambient light intensity, background noise decibel value, and terminal location data through a multi-source data collection module;

[0132] S2. Sequentially preprocess the collected data through an edge processing module, such as image denoising, voice activity detection, etc., and upload the preprocessed data to the cloud server;

[0133] S3. Sequentially conduct biometric verification and behavior feature verification on the received preprocessed data through the cloud server, and generate corresponding response strategies;

[0134] S4. Through the cloud server, conduct comprehensive risk analysis by combining the verification data, environmental features, and user credit data obtained in step S3, and generate corresponding response strategies;

[0135] S5. Regularly analyze the historical outbound call data of the manual review results within a preset period through the cloud server, and adjust the preset risk response threshold range.

[0136] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equal changes and improvements made according to the scope of the present invention application shall still fall within the scope covered by the patent of the present invention.

Claims

1. An outbound call system based on artificial intelligence, characterized in that, The system includes: A multi-source data acquisition module, integrated in the smart phone terminal, for collecting user facial images, speech, ambient light intensity, background noise decibel values, and terminal real-time positioning data; An edge processing module, for performing standardized preprocessing on the collected data and uploading the preprocessed data to the cloud server; A cloud server, for successively performing biometric verification and behavioral feature verification on the received preprocessed data, and conducting comprehensive risk analysis by combining environmental feature parameters and user credit data, and generating corresponding response strategies according to the analysis results.

2. The outbound call system based on artificial intelligence according to claim 1, characterized in that, The process of the edge processing module preprocessing the collected information includes: Successively performing grayscale processing and histogram equalization on the user facial image, and after positioning the key facial feature points, normalizing the pupil distance to a preset pixel value; Performing pre-emphasis filtering and frame windowing processing on the speech, extracting the speech segments with the fundamental frequency exceeding a preset threshold, and removing the silent segments through endpoint detection; Performing high-frequency noise desensitization processing on the ambient light intensity and background noise decibel values to remove random interference; Converting the terminal positioning coordinates into a Geohash encoded string of 12 characters.

3. The outbound call system based on artificial intelligence according to claim 2, characterized in that, The biometric verification process includes: Input the preprocessed user facial image and the user ID card photo into the trained face recognition model to extract the feature vector Q of the user facial image s and the feature vector Q of the user ID card photo std ; Obtained through the formula Analyze and calculate to obtain the face similarity R face ; If R face > R1, it is determined that the face recognition is passed; Conversely, determine that the face recognition fails and terminate the transaction; Wherein, R1 is a preset face similarity threshold.

4. The outbound call system based on artificial intelligence according to claim 3, wherein The biometric verification process also includes: Input the preprocessed speech and the pre-stored user voiceprint template into the trained voiceprint recognition model to extract the user speech feature vector W s and the user voiceprint template feature vector W std ; Obtained through the formula Analyze and calculate to obtain the voiceprint similarity R voice ; If R voice > R2, it is determined that the voiceprint recognition is passed; Conversely, determine that the voiceprint recognition fails and further analysis is required; Wherein, R2 is a preset voiceprint similarity threshold; The further analysis process: Through the formula R all = R face * ω1 + R voice * ω2, the biometric similarity R is obtained by analysis and calculation all ; If R all > R3, it is determined that the biometric verification is passed; Conversely, determine that the biometric verification fails and stop the transaction; Wherein, R3 is a preset biometric similarity threshold.

5. The outbound call system based on artificial intelligence according to claim 4, characterized in that, The behavioral feature verification process includes: The behavioral characteristic abnormality coefficient E is obtained by combining the above formulas for simultaneous analysis and calculation abn ; Among them, A abn is the speech rate anomaly coefficient, P abn is the interaction anomaly coefficient, is the weight coefficient corresponding to the behavior anomaly influencing factor, V s is the current speech rate of the user, V std is the preset reference speech rate, N is the number of times the user clicks on the mobile terminal, T is the outbound call session duration, θ is the deflection angle of the mobile terminal during the session, and a is the adjustment parameter; If E abn >E std , it is determined that the verification of the behavioral characteristics fails, and the transaction is stopped; Conversely, determine that the behavioral feature verification passes; Among them, E std is a preset critical coefficient for abnormal behavioral characteristics.

6. The outbound call system based on artificial intelligence according to claim 5, characterized in that, The process of the comprehensive risk analysis includes: D risk = [(1 - R all ) * τ1 + E abn * τ2 + C s * τ3 + b] * H risk The comprehensive risk characteristic value D is obtained by combining the above formulas for analysis and calculation risk ; Among them, C s is the environmental risk coefficient, H risk is the user's bad credit evaluation parameter, Z s is the actual positioning deviation value of the mobile terminal, b is the credit influence adjustment parameter, τ1, τ2, τ3 are the weight coefficients corresponding to the comprehensive risk influence factors, τ1>τ2>τ3, Z std is the preset positioning deviation reference value, K s is the actual noise background decibel value, K std is the preset noise background reference decibel value, is the weight coefficient corresponding to the environmental risk influence factor; Compare the comprehensive risk eigenvalue D risk with the preset risk response threshold range (D1, D2]; If D risk ≤ D1, the comprehensive risk analysis is normal and trading is allowed; If D1 < D risk < D2, the comprehensive risk analysis is to be determined and manual risk review is required; If D risk > D2, the comprehensive risk analysis is abnormal and the transaction is stopped.

7. An outbound call system based on artificial intelligence according to claim 6, characterized in that, The system further includes: The cloud server regularly analyzes the historical outbound call data of the manual review results within a preset period and adjusts the preset risk response threshold range; Calculating and obtaining the stop transaction rate α in the manual review results within a preset period through the above formula analysis; Wherein, Risk is the number of stop transactions in the manual review results within a preset period, and All is the total number of manual reviews within a preset period; Compare the stop trading rate α in the manual review result with the preset stop trading rate threshold range [α min , α max ; If α < α min , adjust the preset risk response threshold range, D 1,new = D1 + (α min - α) * ρ, D 2,new = D2 + (α min - α) * ρ; If α > α max , adjust the preset risk response threshold range, D 1,new = D1 - (α - α max ) * ρ, D 2,new = D2 - (α - α max ) * ρ; If α ∈ [α min , α max , no adjustment is made to the preset risk response threshold range, D 1,new = D1, D 2,new = D2; Among them, D 1,new is the new lower threshold limit in the preset risk response threshold range, and D 2,new is the new upper threshold limit in the preset risk response threshold range.

8. An outbound call method based on artificial intelligence, characterized in that, The method is used for an outbound call system based on artificial intelligence described in claim 7, and the method includes: S1. The cloud server establishes a communication connection with the smart phone terminal, and collects user facial images, speech, ambient light intensity, background noise decibel values, and terminal positioning data through the multi-source data acquisition module; S2. Successively preprocess the collected data through the edge processing module and upload the preprocessed data to the cloud server; S3. Successively perform biometric verification and behavioral feature verification on the received preprocessed data through the cloud server and generate corresponding response strategies; S4. The cloud server conducts comprehensive risk analysis by combining the verification data, environmental features, and user credit data obtained in step S3 and generates corresponding response strategies; S5. The cloud server analyzes the historical outbound call data of the manual review results within a preset period and adjusts the preset risk response threshold range.

Images