Insurance double-recording real-time quality inspection method, device and equipment and storage medium

Abstract

The invention provides an insurance double-recording real-time quality inspection method, device and equipment and a storage medium, and the method comprises the steps: collecting real-time audio and video data generated in a double-recording process, and recognizing the questioning intention of a customer according to the real-time audio and video data; calling a knowledge base to generate a corresponding compliance voice reply based on the identified question intention, and broadcasting the corresponding compliance voice reply; performing real-time analysis on the real-time audio and video data according to the questioning intention and the broadcasted compliance voice reply to obtain a risk quantized value; when the risk quantized value meets a preset condition, generating a risk prompt based on the risk quantized value; and integrating the risk prompt and the associated historical violation data of the salesman, performing risk rating on the current double-recording video, and outputting the risk rating result as a quality inspection conclusion by adopting the method, so that the insurance double-recording and quality inspection efficiency can be remarkably improved, and the labor cost and compliance risk are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent insurance, in particular to an insurance dual-recording real-time quality inspection method, device, equipment and storage medium. BACKGROUND

[0002] Insurance sales behavior traceability management (referred to as "dual recording") is a core requirement of regulatory departments to standardize insurance sales processes and protect the legitimate rights and interests of policyholders. It needs to record product introduction, risk notification, customer confirmation and other key links through full recording of audio and video to ensure that the sales behavior is compliant and transparent. With the continuous expansion of insurance business, the amount of dual-recording audio and video data is growing explosively, and higher requirements are put forward for the efficiency and accuracy of quality inspection.

[0003] In the prior art, the long-term core mode of insurance dual-recording quality inspection is artificial quality inspection, that is, compliance auditors view the dual-recording audio and video one by one, compare the regulatory requirements and company compliance standards, manually check the compliance of policyholder data display, clause broadcast, customer question and answer, signature confirmation and other links, record violations and form quality inspection conclusions.

[0004] Such prior art has obvious defects: artificial quality inspection is extremely low in efficiency, which is difficult to meet the audit demand of massive dual-recording data, resulting in long quality inspection period and serious backlog; and artificial judgment is easily affected by subjective factors, which has the risk of missing or misjudging the violation situation, and the consistency and reliability of compliance control are insufficient, while the high labor cost also brings heavy operating burden to insurance companies. SUMMARY

[0005] Therefore, the purpose of the present application is to provide an insurance dual-recording real-time quality inspection method, device, equipment and storage medium, which can significantly improve the efficiency of insurance dual-recording and quality inspection, and reduce labor cost and compliance risk.

[0006] In a first aspect, the embodiments of the present application provide an insurance dual-recording real-time quality inspection method, which comprises: collecting real-time audio and video data generated in the dual-recording process, and identifying the questioning intention of the customer according to the real-time audio and video data; based on the identified questioning intention, calling a knowledge base to generate a corresponding compliant voice reply and broadcasting it; based on the questioning intention and the broadcasted compliant voice reply, performing real-time analysis on the real-time audio and video data to obtain a risk quantization value; when the risk quantization value meets a preset condition, generating a risk prompt based on the risk quantization value; comprehensively considering the risk prompt and the associated salesperson historical violation data, performing risk rating on the current dual-recording video, and outputting the risk rating result as a quality inspection conclusion.

[0007] Optionally, the identifying a question intention of the customer according to the real-time audio and video data comprises: converting a voice of the customer in the real-time audio and video data into text through automatic speech recognition technology; adopting an intention recognition algorithm to perform natural language processing on the text, analyze and classify the question intention, and output a corresponding intention recognition confidence.

[0008] Optionally, the generating a compliant voice reply corresponding to the question intention based on the identified question intention and playing the compliant voice reply comprises: querying the knowledge base to obtain a compliant reply text according to the question intention; synthesizing a voice with the voice tone of the salesperson through voice tone cloning technology based on a pre-collected voice sample of the salesperson; converting the compliant reply text into a voice stream using the synthesized voice tone and playing the voice stream to obtain the compliant voice reply.

[0009] Optionally, the obtaining a risk quantization value by analyzing the real-time audio and video data in real time according to the question intention and the played compliant voice reply comprises: adopting a multi-modal large model to perform synchronous analysis on the real-time audio and video data; calculating a semantic matching degree between the question intention and a standard reply intention corresponding to the compliant reply text obtained from the knowledge base; calculating the risk quantization value through a risk triggering algorithm based on the semantic matching degree and the intention recognition confidence.

[0010] Optionally, the generating a risk prompt based on the risk quantization value comprises: comparing the risk quantization value with a preset risk triggering threshold; when the risk quantization value reaches or exceeds the risk triggering threshold, generating prompt information containing a description of a violation situation and operation instructions as the risk prompt.

[0011] Optionally, the risk rating of the current dual-recording video is performed by comprehensively considering the risk prompt and associated historical violation data of the salesperson, comprising: calculating a real-time risk component according to a risk type corresponding to the risk prompt, a triggering frequency, and a preset weight; calculating a historical risk association component based on the historical violation data of the salesperson; fusing the real-time risk component and the historical risk association component to obtain a risk total score of the current dual-recording video; determining the risk rating according to a predefined score interval in which the risk total score is located.

[0012] Optionally, the method further comprises: screening error cases according to the quality inspection conclusion; based on the error cases, iteratively optimizing parameters involved in the risk trigger algorithm or risk rating calculation.

[0013] In a second aspect, the embodiments of the present application provide an insurance double recording real-time quality inspection device, which comprises: a question intention determination module, configured to collect real-time audio and video data generated in a double recording process, and identify a question intention of a customer according to the real-time audio and video data; a voice reply broadcast module, configured to call a knowledge base to generate a corresponding compliant voice reply and broadcast it based on the identified question intention; a risk quantification value determination module, configured to analyze the real-time audio and video data in real time according to the question intention and the broadcasted compliant voice reply, and obtain a risk quantification value; a risk prompt generation module, configured to generate a risk prompt based on the risk quantification value when the risk quantification value meets a preset condition; a quality inspection conclusion generation module, configured to comprehensively consider the risk prompt and associated salesperson historical violation data, perform risk rating on a current double recording video, and output the risk rating result as a quality inspection conclusion.

[0014] Optionally, the identifying a question intention of a customer according to the real-time audio and video data comprises: converting a customer voice in the real-time audio and video data into text through automatic speech recognition technology; performing natural language processing on the text by using an intention recognition algorithm, analyzing and classifying the question intention, and outputting a corresponding intention recognition confidence.

[0015] Optionally, the calling a knowledge base to generate a corresponding compliant voice reply and broadcasting it based on the identified question intention comprises: querying the knowledge base to obtain a compliant reply text according to the question intention; synthesizing a voice with a tone of a salesperson through tone cloning technology based on a pre-collected voice sample of the salesperson; converting the compliant reply text into a voice stream using the synthesized tone and broadcasting it to obtain the compliant voice reply.

[0016] Optionally, the analyzing the real-time audio and video data in real time according to the question intention and the broadcasted compliant voice reply to obtain a risk quantification value comprises: synchronously analyzing the real-time audio and video data by using a multi-modal large model; Calculate the semantic matching degree between the question intent and the standard response intent corresponding to the compliant response text obtained from the knowledge base; Based on the semantic matching degree and the intent recognition confidence degree, the risk quantification value is calculated through a risk triggering algorithm.

[0017] Optionally, generating a risk warning based on the risk quantification value includes: The risk quantification value is compared with a preset risk trigger threshold; When the risk quantification value reaches or exceeds the risk trigger threshold, a prompt message containing a description of the violation and operation instructions is generated as the risk warning.

[0018] Optionally, the risk rating of the current dual-recorded video is based on the combined risk warnings and related historical violation data of the salesperson, including: Calculate the real-time risk component based on the risk type, trigger count, and preset weight corresponding to the risk warning; Calculate the historical risk correlation component based on the salesperson's historical violation data; By fusing the real-time risk component with the historical risk correlation component, the total risk score of the current dual-recorded video is obtained; The risk rating is determined based on the predefined score range in which the total risk score falls.

[0019] Optionally, the device further includes a parameter optimization module for: Error cases were screened based on the quality inspection conclusions. Based on the aforementioned error cases, the parameters involved in the risk triggering algorithm or risk rating calculation are iteratively optimized.

[0020] Thirdly, embodiments of this application provide a computer device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the real-time quality inspection method for dual-recording insurance described in any of the optional embodiments of the first aspect are performed.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the real-time quality inspection method for dual-recording insurance described in any of the optional embodiments of the first aspect.

[0022] The technical solution provided in this application includes, but is not limited to, the following beneficial effects: The process involves collecting real-time audio and video data generated during the dual recording process and identifying the customer's questioning intent based on this data. This step can instantly capture key interactive information throughout the entire dual recording process, accurately pinpoint the customer's core needs and questions, avoid communication inefficiencies caused by information delays or misunderstandings, provide accurate demand guidance for subsequent compliant responses, and ensure that the dual recording interaction revolves around the customer's true intentions.

[0023] Based on the identified intent of the question, the knowledge base is invoked to generate a corresponding compliant voice response and broadcast it. This step, supported by the knowledge base, can quickly output response content that meets regulatory requirements and compliance standards, ensuring the accuracy and compliance of the response. At the same time, the voice broadcast format is intuitive and easy to understand, which not only improves the efficiency of customers receiving information, but also standardizes the response behavior of salespersons and avoids misleading statements or compliance omissions.

[0024] Based on the stated intent of the question and the compliant voice response, the real-time audio and video data is analyzed in real time to obtain a risk quantification value. This step enables a dynamic and objective assessment of the risk of dual recording. By replacing vague risk judgments with quantitative indicators, the degree of risk becomes measurable and traceable, thus moving away from the assessment model that relies on subjective experience and providing a scientific and accurate decision-making basis for risk management.

[0025] When the risk quantification value meets the preset conditions, a risk warning is generated based on the risk quantification value. This step moves risk control from "post-event verification" to "in-event intervention", which can provide an immediate warning when violations occur. The targeted warning content helps salespersons quickly locate the problem and complete the rectification, effectively avoid the accumulation of compliance risks, reduce the cost of re-recording due to violations, and improve the smoothness and compliance rate of the dual recording process.

[0026] Based on the aforementioned risk warnings and related historical violation data of sales personnel, a risk rating is performed on the current dual-recorded videos, and the risk rating result is output as the quality inspection conclusion. This step, by combining real-time risk with historical compliance performance, makes the risk rating more comprehensive and objective. Differentiated risk levels enable precise allocation of quality inspection resources, avoiding resource waste. At the same time, clear quality inspection conclusions provide a clear basis for subsequent business reviews and compliance assessments, improving the efficiency and pertinence of the overall quality inspection work.

[0027] In summary, the invention of this application, through a complete process of "intent identification - compliant response - real-time quantification - risk warning - graded rating," has realized the transformation of insurance dual recording quality inspection from passive response to proactive control. This not only ensures the compliance of sales behavior and the rights and interests of customers, but also significantly improves the efficiency of dual recording and quality inspection, reduces labor costs and compliance risks, and fully meets the needs of regulatory requirements and the scaled development of insurance business.

[0028] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 A flowchart of a real-time quality inspection method for dual-recording insurance provided in Embodiment 1 of this application is shown; Figure 2 A flowchart of a method for determining question intent provided in Embodiment 1 of this application is shown; Figure 3 A flowchart of a voice response broadcasting method provided in Embodiment 1 of this application is shown; Figure 4 A flowchart of a risk quantification value determination method provided in Embodiment 1 of this application is shown; Figure 5 A flowchart of a risk warning method provided in Embodiment 1 of this application is shown; Figure 6 A flowchart of a risk rating method provided in Embodiment 1 of this application is shown; Figure 7 A flowchart of a parameter iterative optimization method provided in Embodiment 1 of this application is shown; Figure 8 This shows a schematic diagram of the structure of an insurance dual-recording real-time quality inspection device provided in Embodiment 2 of this application; Figure 9 This shows a schematic diagram of the structure of the second type of real-time quality inspection device for dual-recording insurance provided in Embodiment 2 of this application; Figure 10 A schematic diagram of the structure of a computer device provided in Embodiment 3 of this application is shown. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0032] Example 1 To facilitate understanding of this application, the following is combined with... Figure 1 The flowchart of the real-time quality inspection method for dual recording of insurance provided in Embodiment 1 of this application illustrates Embodiment 1 of this application in detail.

[0033] See Figure 1 As shown, Figure 1 A flowchart of a real-time quality inspection method for dual-recording insurance provided in Embodiment 1 of this application is shown, wherein the method includes steps S101 to S105: S101: Collect real-time audio and video data generated during the dual recording process, and identify the customer's questioning intent based on the real-time audio and video data.

[0034] Specifically, the collected real-time audio and video data must cover the entire insurance dual recording process, including all mandatory recording process nodes such as insurance application material display, terms and conditions broadcast, customer Q&A, and signature confirmation. This process serves the dual recording compliance needs of the life insurance operation APP development team (the project to which it belongs) and is designed under the leadership of the R&D Center - Mobile Application Development Department (the organizing department).

[0035] Identifying the intent behind a customer's question requires a real-time interactive semantic parsing module. First, ASR (Automatic Speech Recognition, used to convert human speech into text) is used to extract the customer's speech information. Then, the intent recognition confidence algorithm is used for processing. Combining speech recognition and facial expression recognition, the user's intent is comprehensively scored and evaluated from multiple dimensions.

[0036] S102: Based on the identified question intent, call the knowledge base to generate a corresponding compliant voice response and broadcast it.

[0037] Specifically, the knowledge base invoked is a multi-dimensional knowledge base system of "product-quality inspection-compliance", which includes three major categories: product knowledge base, quality inspection rule base, and compliance rule base. All knowledge base data is indexed according to insurance product type and regulatory region, supporting quick retrieval and access.

[0038] The product knowledge base includes basic terms and conditions of main and supplementary insurance, rules for legal combination, cash value calculation standards, claims conditions, frequently asked questions and compliance response templates; the quality inspection rules base includes standards for the duration and clarity of insurance application materials, a list of mandatory process nodes, detailed rules for standardized communication scripts, and requirements for recording the signing process; the compliance rules base includes the retrospective management methods for insurance sales behavior and supplementary regulations in various regions, as well as the company's internal red lines for sales behavior and compliance assessment standards.

[0039] Compliant voice responses must employ voice cloning technology, which involves sampling the salesperson's voice three times to synthesize a local model. This process can reproduce the salesperson's own voice, ensuring that the broadcast style is consistent with the salesperson's daily speech and improving customer acceptance.

[0040] S103: Based on the stated intent of the question and the broadcast compliant voice response, perform real-time analysis on the real-time audio and video data to obtain a risk quantification value.

[0041] Specifically, real-time analysis is performed by the real-time compliance quality inspection module. It uses a multimodal large model (which can process audio and video data simultaneously) to perform synchronous analysis on the real-time audio and video streams. The input is the raw audio and video streams, which are segmented into stages and then output as scores for each stage. The scores are then weighted and summed by combining the preset risk values ​​of each stage.

[0042] The risk quantification value is calculated using a real-time risk trigger threshold algorithm. This algorithm needs to integrate multiple parameters: first, the real-time compliance risk score, including the output of the intent recognition confidence algorithm, the compliance matching degree between the salesperson's response and the knowledge base, and the content of the salesperson's verbal response to customer inquiries; second, fixed coefficients, including the regulatory flexibility coefficient and the process node standardization weight coefficient (which are fixed values ​​after training, initially preset based on existing data, and determined after iterative regression optimization). Finally, the risk coefficient, i.e., the risk quantification value, is generated by weighted summation.

[0043] S104: When the risk quantification value meets the preset conditions, a risk warning is generated based on the risk quantification value.

[0044] Specifically, the preset condition is that the risk quantification value reaches or exceeds the preset risk trigger threshold. The risk trigger threshold needs to be set in combination with the insurance dual recording regulatory requirements and the company's compliance standards. Initially, it is calibrated based on historical compliance data, and subsequently adjusted continuously through feedback and iterative optimization modules.

[0045] Risk warnings need to be generated based on specific violations, which are divided into two categories: visual and auditory. Visual violations include incomplete display of application documents and certificates, failure to capture the signature process, insufficient time for document display after signing, blurry and illegible signature pages, more than 30% of the policyholder's face being obscured in the recording, participation of someone other than the policyholder in the recording, significant interference in the recording environment, and omission of mandatory recording steps. Auditory violations include salespersons failing to fully read standardized disclaimers, omitting core disclaimers, using prohibited or misleading statements, customers failing to provide clear responses or explicitly refusing to confirm key confirmation items, salespersons' responses to customer questions not matching the compliance requirements corresponding to the customer's intent tags, audio interruptions exceeding 10 seconds during dual recording without reasonable instructions, and salespersons speaking too fast, making key information unclear.

[0046] Risk warnings are pushed to the dual-recording operation terminal in the form of a semi-transparent floating window. The content must include both a description of the violation and operation instructions. The operation instructions must be written in strict accordance with the requirements of the quality inspection rule library and the compliance rule library to ensure that salespersons can quickly correct violations according to the prompts.

[0047] S105: Based on the aforementioned risk warnings and related historical violation data of salespersons, conduct a risk rating for the current dual-recording video and output the risk rating result as the quality inspection conclusion.

[0048] Specifically, risk rating is performed by the risk grading and review module, based primarily on a comprehensive risk scoring algorithm that covers the entire process. The algorithm's expression is as follows: ; in, This refers to the comprehensive risk assessment of the entire dual-recording video process. For the first The result of the real-time risk trigger threshold, where m is the total number of real-time risk triggers. Weighted according to the severity of risk, This refers to the risk correction rate (i.e., the percentage of violations corrected by sales staff based on risk warnings). θ represents the historical risk correlation (calculated based on the salesperson's past dual-recording violation records and compliance assessment results), and is the historical risk weighting coefficient.

[0049] Risk ratings are divided into three levels: high, medium, and low. Different levels correspond to different audit resource allocations: high-risk videos are assigned to the priority human audit queue and reviewed by senior compliance personnel; medium-risk videos are assigned to the regular human audit queue; and low-risk videos can be automatically audited by AI, achieving precise allocation of quality inspection resources and improving audit efficiency.

[0050] The quality inspection conclusions need to be linked to the audit conclusions, risk warning data, salesperson's historical violation data, and key real-time audio and video clips to support subsequent risk tracing and bad-case screening. At the same time, the conclusions need to be synchronized to the life insurance operation APP for the business team to view and rectify.

[0051] This implementation plan, through real-time data collection covering the entire dual recording process, multi-dimensional knowledge base access, risk quantification, and risk rating, has transformed insurance dual recording from "post-event rectification" to "real-time intervention." It not only solves the problems of fragmented technical clauses and inconsistent compliance basis in the existing system, but also improves quality inspection efficiency through differentiated audit resource allocation. At the same time, relying on a complete data storage and traceability mechanism, it provides data support for subsequent system optimization, effectively reducing compliance risks and manual quality inspection costs.

[0052] In an optional implementation, see Figure 2 As shown, Figure 2 The flowchart of a method for determining questioning intent provided in Embodiment 1 of this application is shown, wherein the step of identifying the customer's questioning intent based on the real-time audio and video data includes steps S201-S202: S201: Convert the customer's voice in the real-time audio and video data into text using automatic speech recognition technology.

[0053] Specifically, Automatic Speech Recognition (ASR) technology is the core technology for collecting customer speech during dual recording and converting it into analyzable data. It can filter out environmental noise in speech, extract effective speech information, and convert it into standardized text, providing basic data for subsequent natural language processing.

[0054] The text processed by ASR technology needs to be standardized in format, including correcting speech recognition errors (such as homophones and colloquial expressions) and segmenting and annotating (dividing semantic units according to the pauses in the customer's question) to ensure that the text information accurately reflects the content of the customer's question.

[0055] S202: The text is processed by an intent recognition algorithm to parse and classify the question intent, and the corresponding intent recognition confidence score is output.

[0056] Specifically, Natural Language Processing (NLP) technology aims to enable computers to understand human language. Here, NLP technology is used to extract keywords (such as "claim conditions", "cash value", "exclusion clauses") and semantic logic (such as interrogative sentences and demand expressions) from the text. Combined with the classification of historical high-frequency consultation questions in the product knowledge base, the intention of the question is categorized (such as "term consultation", "claim questions", "confirmation of insurance conditions").

[0057] Intent recognition confidence is used to evaluate the reliability of intent classification. The value ranges from 0 to 1. The higher the confidence, the more accurate the intent recognition result. If the confidence is lower than the preset threshold (such as 0.6), secondary recognition needs to be triggered. Supplementary collection of customer facial expressions (such as doubt or confirmation expressions) is used to help judge the intent and improve recognition accuracy.

[0058] This implementation plan significantly improves the accuracy of customer inquiry intent recognition by combining the precise speech-to-text conversion of ASR technology with the deep semantic analysis of NLP technology, along with confidence assessment and secondary recognition mechanisms. It solves the problems of inaccurate intent recognition and inability to support compliant responses in existing technologies, laying the foundation for subsequent knowledge base calls to generate accurate responses. At the same time, the standardized text processing process also reduces the errors in subsequent semantic matching and risk assessment.

[0059] In an optional implementation, see Figure 3 As shown, Figure 3 The flowchart of a voice response broadcasting method provided in Embodiment 1 of this application is shown, wherein the step of generating a corresponding compliant voice response by calling a knowledge base and broadcasting it based on the identified question intent includes steps S301 to S303: S301: Based on the stated intent of the question, query the knowledge base to obtain a compliant response text.

[0060] Specifically, the query process relies on the classification index system of the knowledge base. First, the category is determined according to the intent of the question (such as the "Claims Conditions" module of the product knowledge base corresponding to "Claims Questions"). Then, the matching answer content is screened in combination with the type of insurance product to which the dual recording belongs (such as accident insurance and medical insurance) and the regulatory region (such as the supplementary compliance regulations of different regions) to ensure that the answer text complies with the regional regulatory requirements.

[0061] The compliant response text needs to be extracted from the preset content of the knowledge base: if it is a frequently asked question, the "compliant response template" in the product knowledge base can be directly called; if it is a special question, the response needs to be generated by combining the "main and supplementary insurance terms" in the product knowledge base and the "sales behavior norms" in the compliance rule base. After generation, the semantic compliance needs to be verified by the text reasoning model to avoid misleading statements.

[0062] S302: Based on pre-collected voice samples of salespersons, synthesize voices with the voice timbre of the salesperson using timbre cloning technology.

[0063] Specifically, the voice sample collection of salespersons must meet the following standards: the collection environment must be noise-free, the collection content must include voice segments with different intonations (such as statements and questions) and speaking speeds, and a total of three samplings must be conducted to ensure that the samples cover the daily voice characteristics of salespersons, providing a basis for the authenticity of cloned voice timbre.

[0064] Voice cloning technology is essentially a speech synthesis technology. It extracts features (such as timbre, intonation, and speech rate) from collected speech samples using a local model to create a unique voice model for the salesperson. This model is then called when synthesizing speech to ensure that the similarity between the synthesized speech and the salesperson's real voice reaches a preset standard (such as more than 90%).

[0065] S303: Use the synthesized timbre to convert the compliant response text into a speech stream and broadcast it to obtain the compliant speech response.

[0066] Specifically, speech conversion requires control of speech rate and pauses: the speech rate must meet the requirements of the quality inspection rule base (such as 120-150 words per minute) to avoid making key information unclear due to excessive speed; pauses should be added before and after key clauses (such as disclaimers and claim conditions) to highlight key information and improve customer comprehension.

[0067] The broadcasting process requires simultaneous recording of voice stream data and corresponding text information, which are then stored in the dual-recording system database for subsequent quality inspection and risk assessment (such as verifying whether the salesperson broadcast the complete response content).

[0068] This implementation plan ensures the compliance and regional adaptability of the response text through a knowledge base query mechanism with categorized indexes, solving the problem that existing technologies with fixed scripts cannot handle personalized customer inquiries. At the same time, the application of voice cloning technology enhances the personalization and customer acceptance of dual-recording interaction. Combined with speech rate control and key pause design, it further ensures the customer's understanding of compliant responses and reduces compliance risks caused by interaction deviations.

[0069] In an optional implementation, see Figure 4 As shown, Figure 4 The flowchart illustrates a method for determining a risk quantification value provided in Embodiment 1 of this application. The step of obtaining a risk quantification value through real-time analysis of the real-time audio and video data based on the question's intent and the broadcast compliant voice response includes steps S401-S403: S401: The real-time audio and video data are synchronously analyzed using a multimodal large model.

[0070] Specifically, the multimodal large model uses a full-modal model, which can process audio and video data simultaneously: audio analysis focuses on the completeness of the salesperson's broadcast content, the clarity of the customer's response, and the duration of voice interruptions; video analysis focuses on the display status of the insurance application materials, the standardization of the signing process, the policyholder's facial recognition results, and interference factors in the recording environment.

[0071] Synchronous analysis needs to be conducted in segments according to the dual recording process (such as "terms broadcasting segment", "customer Q&A segment", "signature segment"). Each segment should have its own dedicated analysis dimensions and risk assessment criteria to avoid cross-segment data interference and improve the accuracy of risk identification.

[0072] S402: Calculate the semantic matching degree between the question intent and the standard response intent corresponding to the compliant response text obtained from the knowledge base.

[0073] Specifically, the standard response intent is the core semantics preset in the compliant response text in the knowledge base (e.g., the standard intent corresponding to the response text "Medical insurance claims require a diagnosis certificate and a list of expenses" is "clarify the materials required for medical insurance claims"). It is necessary to use a text reasoning model to semantically compare the question intent with the standard response intent.

[0074] The semantic matching degree calculation uses a similarity algorithm (such as cosine similarity), with a value range of 0-1. The higher the matching degree, the more the response text matches the customer's question requirements. If the matching degree is lower than the preset threshold (such as 0.7), it is judged as "response semantic deviation" and needs to be included in the real-time compliance risk score, affecting the subsequent risk quantification value calculation.

[0075] S403: Based on the semantic matching degree and the intent recognition confidence degree, the risk quantification value is calculated through a risk triggering algorithm.

[0076] Specifically, the risk triggering algorithm, also known as the real-time risk triggering threshold algorithm, requires the integration of three core parameters during calculation: first, semantic matching degree (weighted at, for example, 40%), second, intent recognition confidence degree (weighted at, for example, 30%), and third, real-time compliance supplementary score (such as the completeness of the salesperson's broadcast content and the clarity of the customer's response, weighted at, for example, 30%). The weights of each parameter are the process node specification weight coefficients, which are fixed after training.

[0077] The calculation logic for the risk quantification value is as follows: First, standardize various parameters (convert them into values ​​in the range of 0-1), then sum them according to preset weights, and finally multiply them by the regulatory flexibility coefficient (set according to the tightness of regulatory policies, such as 1.0 in normal times and 1.2 in strict regulatory periods) to obtain the final risk quantification value, which directly reflects the compliance risk level of the current dual recording process.

[0078] This implementation plan achieves synchronous analysis of audio and video data through a multimodal large model, covering violations in both visual and auditory dimensions, thus solving the problem of incomplete risk identification in existing single-dimensional technologies. At the same time, the weighted quantification algorithm combining semantic matching degree and confidence degree, along with the dynamic adjustment of the regulatory elasticity coefficient, enables the risk quantification value to accurately reflect the real-time compliance risk level, providing a scientific and reliable basis for triggering subsequent risk warnings and avoiding missed or misjudged risks.

[0079] In an optional implementation, see Figure 5 As shown, Figure 5 The flowchart of a risk warning method provided in Embodiment 1 of this application is shown, wherein the step of generating a risk warning based on the risk quantification value includes steps S501-S502: S501: Compare the risk quantification value with the preset risk trigger threshold.

[0080] Specifically, the preset risk trigger thresholds need to be set differently according to the stages of the dual recording process: for example, the threshold for the "Terms Broadcast Stage" is set lower (e.g., 0.3) because it involves core compliance requirements, and a prompt will be triggered even for minor violations; the threshold for the "Document Display Stage" is set higher (e.g., 0.5), and a prompt will be triggered only for more serious violations. The thresholds are initially calibrated based on historical violation case data and will be adjusted in subsequent iterations and optimizations based on feedback.

[0081] The comparison process needs to be carried out in real time. At preset intervals (such as 1 second), the risk quantification value of the current stage is compared with the corresponding threshold to ensure that violations are detected in a timely manner and to avoid the accumulation of risks.

[0082] S502: When the risk quantification value reaches or exceeds the risk trigger threshold, a prompt message containing a description of the violation and operation instructions is generated as the risk prompt.

[0083] Specifically, the description of the violation must accurately correspond to the risk type: visual violations must clearly point out the specific problem (such as "the name page is not fully displayed on the application document" or "the policyholder's face is covered by more than 30%); auditory violations must indicate the content of the violation (such as "failure to announce the element of 'excluding war' in the disclaimer" or "speech speed exceeding 160 words per minute"), and avoid vague expressions.

[0084] The operational guidelines are executable: for example, regarding "incomplete document display", the guidelines state "please display the name page and validity period page of the insurance document in front of the camera for a duration of no less than 3 seconds"; regarding "speaking too fast", the guidelines state "please reduce your speaking speed to 120-150 words per minute and re-read the clause". The guidelines must be written strictly according to the quality inspection rule library to ensure that salespersons can directly implement them.

[0085] This implementation plan enables the immediate detection of violations through differentiated risk trigger thresholds and a real-time comparison mechanism, avoiding the risk accumulation problem of the traditional "post-event verification" model. At the same time, the accurate description of violations and the actionable instructions help sales staff quickly locate and correct violations, significantly improving the first-time pass rate of dual recording, reducing the cost of re-recording due to violations, and further strengthening the compliance control capabilities of the dual recording process.

[0086] In an optional implementation, see Figure 6 As shown, Figure 6 The flowchart of a risk rating method provided in Embodiment 1 of this application is shown, wherein the step of comprehensively considering the risk warning and the associated historical violation data of the salesperson to perform risk rating on the current dual-recorded video includes steps S601 to S604: S601: Calculate the real-time risk component based on the risk type, trigger count, and preset weight corresponding to the risk warning.

[0087] Specifically, the risk type weights are set according to the severity: for example, "using misleading statements" and "recording not by the insured person" are high-severity risks, with a weight of 1.0; "the presentation time of the information is less than 1 second" and "speaking at a slightly fast pace (151-160 words / minute)" are low-severity risks, with a weight of 0.3. The weight values ​​are determined with reference to the "sales behavior red line level" in the compliance rule library.

[0088] The real-time risk component calculation logic is as follows: First, calculate the score for each risk trigger (risk type weight × the quantified risk value for that instance). Then, sum all the individual scores to obtain the real-time risk component. This component reflects the real-time compliance risk level of the current dual-recorded video, corresponding to the comprehensive risk scoring algorithm for the entire process. part.

[0089] S602: Calculate the historical risk correlation component based on the salesperson's historical violation data.

[0090] Specifically, the salesperson's historical violation data comes from the company's compliance management system, including the number of violations in the past 12 months (such as the number of high-severity violations and the number of low-severity violations), the violation rectification rate (the percentage of violations that were not rectified), and the compliance assessment level (such as A / B / C / D level). The data must be anonymized before being accessed to protect the salesperson's privacy.

[0091] Specifically, the calculation logic for the historical risk correlation component is as follows: first, calculate the historical risk correlation degree based on historical violation data. (e.g., A-level salesperson) =0.1, D-level salesperson =0.8), then multiply by the historical risk weighting coefficient θ (e.g., set to 0.3) to obtain the historical risk correlation component. This component reflects the impact of the salesperson's past compliance performance on the current dual recording risk, corresponding to the risk assessment algorithm in the whole process. part.

[0092] S603: The real-time risk component and the historical risk correlation component are fused to obtain the total risk score of the current dual-recorded video.

[0093] Specifically, the integration process needs to incorporate a risk correction rate. (That is, the percentage of violations that a salesperson corrects based on risk warnings, such as correcting 2 out of 3 violations) =0.67), first multiply the real-time risk component by ( The corrected real-time risk component is obtained, and then added to the historical risk correlation component to finally obtain the total risk score. The formula corresponds to the full-process risk comprehensive scoring algorithm. ,in This is the total risk score.

[0094] The total risk score ranges from 0 to 100. When calculating the score, the real-time risk component and the historical risk correlation component need to be standardized (e.g., the original calculation results are converted into 0-70 points and 0-30 points) to ensure that the total score can intuitively reflect the degree of risk.

[0095] S604: Determine the risk rating based on the predefined score range in which the total risk score falls.

[0096] Specifically, the predefined score ranges are set according to the company's compliance requirements: for example, 0-30 points are low risk, 31-60 points are medium risk, and 61-100 points are high risk. The range division needs to be verified by historical data to ensure that high-risk ranges correspond to high violation rate cases and low-risk ranges correspond to low violation rate cases.

[0097] Risk rating results must be linked to the review queue: low-risk videos are automatically assigned to the AI ​​review queue, with a review time of no more than 1 minute, and are directly archived after passing the review; medium-risk videos are assigned to the regular human review queue, with a review time of no more than 3 minutes; high-risk videos are assigned to the key human review queue, which requires cross-review by 2 senior compliance personnel, with a review time of no more than 10 minutes to ensure review quality.

[0098] This implementation plan integrates real-time risk components with historical risk correlation components and dynamically adjusts the risk correction rate to ensure that the total risk score comprehensively and objectively reflects the compliance risk level of dual-recorded videos. Furthermore, the risk rating based on score ranges and the binding of differentiated review queues enable precise allocation of quality inspection resources. Humans focus on high-risk cases to ensure review quality, while AI processes low-risk cases to improve efficiency, thus solving the problems of inefficient allocation of quality inspection resources and high review costs associated with existing technologies.

[0099] In an optional implementation, see Figure 7 As shown, Figure 7 The flowchart of a parameter iterative optimization method provided in Embodiment 1 of this application is shown, wherein the method further includes steps S701-S702: S701: Screen erroneous cases based on the quality inspection conclusions.

[0100] Specifically, bad cases are selected based on the criteria of "inconsistency between real-time quality inspection results and manual review results," including three types of situations: first, semantic misjudgment (e.g., real-time identification of intent as "terms inquiry," but manual judgment as "claims question"); second, risk assessment deviation (e.g., real-time assessment as low risk, but manual judgment as high risk); and third, scoring deviation (e.g., real-time risk score of 40 points, adjusted to 65 points after manual review).

[0101] The screening process needs to be automatically executed by the feedback iteration optimization module. The quality inspection conclusion data of the previous day should be extracted periodically (e.g. daily), and the key results (intent tags, risk level, total risk score) of real-time quality inspection and manual review should be compared. Inconsistent cases should be marked and stored in the bad-case database, while recording the reasons for inconsistency (e.g., "unreasonable semantic matching algorithm parameters" or "historical risk weight coefficient is too high").

[0102] S702: Based on the aforementioned error cases, iteratively optimize the parameters involved in the risk triggering algorithm or risk rating calculation.

[0103] Specifically, the optimization process employs a dynamic iterative optimization algorithm with weights, the expression of which is: ,in This refers to the new weight parameters after optimization of the preceding algorithm (such as the semantic matching weight of the risk triggering algorithm, and the historical risk weight coefficient θ for risk rating calculation). Refers to the original weight parameters before optimization by the preceding algorithm. This refers to the bad-case impact factor (set based on the number and severity of bad cases; for example, when high-severity bad cases account for 30%). =0.2), This refers to the credibility factor of the review results (the credibility of manual review results is set to 1.0). This refers to the system operating state adaptation coefficient (set to 1.0 when the system is running stably, and to 0.8 when the load is too high).

[0104] Specifically, the optimized parameters fall into two categories: first, risk trigger algorithm parameters, such as semantic matching degree weight, regulatory flexibility coefficient, and risk trigger threshold; second, risk rating calculation parameters, such as risk severity weight and historical risk weight coefficient θ. After optimization, the effect needs to be verified through a test set (e.g., the bad-case incidence rate is reduced to below 5%). After successful verification, the system algorithm configuration is updated, and the knowledge base content is updated simultaneously (e.g., new compliance response templates are added, and the semantic parsing rules of the clauses are adjusted), forming a closed-loop mechanism of "screening-optimization-verification-update" to continuously improve the system's adaptability to complex scenarios.

[0105] This implementation plan achieves continuous updates to system parameters and knowledge base through precise bad-case screening and dynamic iterative optimization algorithms, enabling the system to continuously adapt to complex scenarios in the dual recording process (such as new types of customer inquiries and regulatory policy adjustments). This solves the problems of existing technologies lacking optimization mechanisms and having limited system adaptability. At the same time, the closed-loop optimization mechanism ensures that the quality inspection quality continuously improves with system use, can meet the dynamic changes in regulatory and company compliance requirements in the long term, and extends the system's life cycle and application value.

[0106] Example 2 See Figure 8 As shown, Figure 8 A schematic diagram of a real-time quality inspection device for dual-recording insurance provided in Embodiment 2 of this application is shown, wherein the device includes: The questioning intent determination module 801 is used to collect real-time audio and video data generated during the dual recording process and identify the customer's questioning intent based on the real-time audio and video data. The voice response broadcast module 802 is used to generate a corresponding compliant voice response based on the identified question intent and broadcast it. The risk quantification value determination module 803 is used to perform real-time analysis on the real-time audio and video data based on the questioning intent and the broadcast compliant voice response to obtain the risk quantification value; The risk warning generation module 804 is used to generate a risk warning based on the risk quantification value when the risk quantification value meets the preset conditions. The quality inspection conclusion generation module 805 is used to comprehensively analyze the risk warnings and related historical violation data of salespersons, perform risk rating on the current dual-recorded video, and output the risk rating result as the quality inspection conclusion.

[0107] In an optional implementation, identifying the customer's questioning intent based on the real-time audio and video data includes: The customer's voice in the real-time audio and video data is converted into text using automatic speech recognition technology; The text is processed by an intent recognition algorithm to parse and classify the question intent, and the corresponding intent recognition confidence score is output.

[0108] In an optional implementation, the step of generating a corresponding compliant voice response based on the identified question intent and broadcasting it by invoking a knowledge base includes: Based on the stated intent of the question, the knowledge base is queried to obtain a compliant response text; Based on pre-collected voice samples of salespersons, voices with the salesperson's timbre are synthesized using timbre cloning technology; The synthesized timbre is used to convert the compliant response text into a speech stream and broadcast it to obtain the compliant speech response.

[0109] In an optional implementation, the step of performing real-time analysis of the real-time audio and video data to obtain a risk quantification value based on the question's intent and the broadcast compliant voice response includes: A multimodal large model is used to synchronously analyze the real-time audio and video data; Calculate the semantic matching degree between the question intent and the standard response intent corresponding to the compliant response text obtained from the knowledge base; Based on the semantic matching degree and the intent recognition confidence degree, the risk quantification value is calculated through a risk triggering algorithm.

[0110] In an optional implementation, generating a risk warning based on the risk quantification value includes: The risk quantification value is compared with a preset risk trigger threshold; When the risk quantification value reaches or exceeds the risk trigger threshold, a prompt message containing a description of the violation and operation instructions is generated as the risk warning.

[0111] In an optional implementation, the risk rating of the current dual-recorded video, based on the combined risk warnings and associated historical violation data of the salesperson, includes: Calculate the real-time risk component based on the risk type, trigger count, and preset weight corresponding to the risk warning; Calculate the historical risk correlation component based on the salesperson's historical violation data; By fusing the real-time risk component with the historical risk correlation component, the total risk score of the current dual-recorded video is obtained; The risk rating is determined based on the predefined score range in which the total risk score falls.

[0112] In an optional implementation, see Figure 9 As shown, Figure 9 The diagram illustrates the structure of the second type of real-time quality inspection device for dual-recording insurance provided in Embodiment 2 of this application. The device further includes a parameter optimization module 901, used for: Error cases were screened based on the quality inspection conclusions. Based on the aforementioned error cases, the parameters involved in the risk triggering algorithm or risk rating calculation are iteratively optimized.

[0113] Example 3 Based on the same application concept, see [link / reference] Figure 10 As shown, Figure 10This illustration shows a structural schematic diagram of a computer device provided in Embodiment 3 of this application, wherein, as shown... Figure 10 As shown, the computer device 1000 provided in Embodiment 3 of this application includes: The system includes a processor 1001, a memory 1002, and a bus 1003. The memory 1002 stores machine-readable instructions executable by the processor 1001. When the computer device 1000 is running, the processor 1001 and the memory 1002 communicate via the bus 1003. When the machine-readable instructions are executed by the processor 1001, they perform the steps of the real-time quality inspection method for dual-recording insurance shown in Embodiment 1.

[0114] Example 4 Based on the same concept, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the real-time quality inspection method for dual-recording insurance described in any of the above embodiments.

[0115] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0116] The computer program product for real-time quality inspection of dual recording in insurance provided in this application includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0117] The real-time quality inspection device for dual-recording insurance provided in this application embodiment can be specific hardware on the device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in this application embodiment are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0118] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] In addition, the functional units in the embodiments provided in this application 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.

[0121] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0122] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0123] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A real-time quality inspection method for dual-recording insurance data, characterized in that, The method includes: Collect real-time audio and video data generated during the dual recording process, and identify the customer's questioning intent based on the real-time audio and video data; Based on the identified question intent, the knowledge base is invoked to generate a corresponding compliant voice response and broadcast it. Based on the stated intent of the question and the broadcast compliant voice response, the real-time audio and video data is analyzed in real time to obtain a risk quantification value; When the risk quantification value meets the preset conditions, a risk warning is generated based on the risk quantification value; Based on the aforementioned risk warnings and related historical violation data of salespersons, a risk rating is assigned to the current dual-recorded video, and the risk rating result is output as the quality inspection conclusion.

2. The method according to claim 1, characterized in that, The step of identifying the customer's questioning intent based on the real-time audio and video data includes: The customer's voice in the real-time audio and video data is converted into text using automatic speech recognition technology; The text is processed by an intent recognition algorithm to parse and classify the question intent, and the corresponding intent recognition confidence score is output.

3. The method according to claim 2, characterized in that, The step of generating a corresponding compliant voice response based on the identified question intent and broadcasting it includes: Based on the stated intent of the question, the knowledge base is queried to obtain a compliant response text; Based on pre-collected voice samples of salespersons, voices with the salesperson's timbre are synthesized using timbre cloning technology; The synthesized timbre is used to convert the compliant response text into a speech stream and broadcast it to obtain the compliant speech response.

4. The method according to claim 3, characterized in that, The step of performing real-time analysis of the real-time audio and video data to obtain a risk quantification value based on the question's intent and the broadcast compliant voice response includes: A multimodal large model is used to synchronously analyze the real-time audio and video data; Calculate the semantic matching degree between the question intent and the standard response intent corresponding to the compliant response text obtained from the knowledge base; Based on the semantic matching degree and the intent recognition confidence degree, the risk quantification value is calculated through a risk triggering algorithm.

5. The method according to claim 1, characterized in that, The generation of risk alerts based on this risk quantification value includes: The risk quantification value is compared with a preset risk trigger threshold; When the risk quantification value reaches or exceeds the risk trigger threshold, a prompt message containing a description of the violation and operation instructions is generated as the risk warning.

6. The method according to claim 1, characterized in that, Based on the aforementioned risk warnings and related historical violation data of salespersons, a risk rating is assigned to the current dual-recording videos, including: Calculate the real-time risk component based on the risk type, trigger count, and preset weight corresponding to the risk warning; Calculate the historical risk correlation component based on the salesperson's historical violation data; By fusing the real-time risk component with the historical risk correlation component, the total risk score of the current dual-recorded video is obtained; The risk rating is determined based on the predefined score range in which the total risk score falls.

7. The method according to claim 1, characterized in that, The method further includes: Error cases were screened based on the quality inspection conclusions. Based on the aforementioned error cases, the parameters involved in the risk triggering algorithm or risk rating calculation are iteratively optimized.

8. A real-time quality inspection device for dual-recording insurance, characterized in that, The device includes: The questioning intent determination module is used to collect real-time audio and video data generated during the dual recording process and identify the customer's questioning intent based on the real-time audio and video data. The voice response broadcast module is used to generate a corresponding compliant voice response based on the identified question intent and broadcast it. The risk quantification value determination module is used to perform real-time analysis on the real-time audio and video data based on the questioning intent and the broadcast compliant voice response to obtain the risk quantification value; The risk warning generation module is used to generate a risk warning based on the risk quantification value when the risk quantification value meets the preset conditions. The quality inspection conclusion generation module is used to comprehensively analyze the risk warnings and related historical violation data of salespersons, perform risk rating on the current dual-recorded video, and output the risk rating result as the quality inspection conclusion.

9. A computer device, characterized in that, include: The system includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the real-time quality inspection method for dual-recording insurance as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the real-time quality inspection method for dual-recording insurance as described in any one of claims 1 to 7.

Images