Internet insurance claim settlement service processing method, system, equipment and medium

Abstract

The invention provides an internet insurance claim settlement service processing method, system and device and a medium, and a server receives a claim settlement material generation problem through a user side and prompts a user to retell and answer. A user terminal collects a user video and uploads the user video to a server, the server performs character recognition on the video to divide retelling and answer parts and extract corresponding features, determines a question and answer risk level according to the features, determines a material risk level by combining voice content and claim settlement materials, and comprehensively obtains a comprehensive risk level of a claim settlement case. And determining a target business processing end according to the comprehensive risk level and the processing end authority to generate a claim settlement application form and sending the claim settlement application form. And finally, the target business processing end completes claim settlement payment according to the comprehensive risk level, the application form and the insurance contract. According to the method, multi-dimensional risk assessment and grading processing are realized, the fraud recognition rate can be improved, hidden contradictions can be found, low-risk fast claim, high-risk fine review, resource allocation optimization and false claim hidden danger reduction are realized, and the claim settlement efficiency and user experience are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of financial management, and in particular to a method, system, device and medium for processing internet insurance claims. Background Art

[0002] With the in-depth application of internet technology in the financial and insurance sectors, the traditional insurance claims model is gradually transforming towards intelligent and digital approaches. Users are placing higher demands on the efficiency, convenience, and security of the claims process. However, existing technologies, which primarily rely on manual review or single-modal data (such as text materials) for claims processing, struggle to meet the risk management needs in complex scenarios. Especially in the context of increasingly diverse fraud risks, how to achieve accurate risk assessment through multi-dimensional data integration and how to dynamically adapt the processing process based on risk levels have become urgent technical challenges for the industry.

[0003] At present, most Internet insurance claims processing is based on traditional manual review or simple automated processes. In the manual review mode, claims personnel manually check the paper or electronic claims materials submitted by users (such as accident certificates, medical receipts, etc.), and judge the authenticity of the materials and the rationality of the claims based on experience. Some automated systems use OCR technology to extract text information from materials, perform entity alignment and logical verification according to preset rules, or adopt a unified fixed review process for all cases. At the same time, existing technologies mostly rely on single text data for risk assessment, and lack comprehensive analysis and utilization of multimodal data such as user voice and video.

[0004] However, research has found that existing internet insurance claims technology has significant shortcomings. First, data utilization is limited in dimension, focusing only on the text of claim materials and ignoring behavioral data such as voice and facial expressions generated by users during the interaction process. It is unable to identify potential fraud risks through behavioral characteristics such as changes in micro-expressions and abnormal speech speed, resulting in low fraud identification accuracy. Second, the risk assessment mechanism is imperfect and lacks multi-dimensional cross-validation, making it difficult to detect hidden contradictions between materials and user statements. Third, the process is inflexible, using the same processing process regardless of the risk level of the case. This leads to inefficient processing of low-risk cases and the risk of incorrect or missed claims in high-risk cases due to insufficient review depth. Fourth, resource scheduling is not intelligent, and tasks cannot be dynamically allocated according to the risk level of the case and the load on the processing end, resulting in a waste of resources and an imbalance in processing efficiency. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an Internet insurance claims processing method, system, equipment and medium to improve the fraud identification rate, discover hidden contradictions, achieve low-risk quick compensation and high-risk precise review, optimize resource allocation, reduce the risk of wrong compensation, and significantly improve claims efficiency and user experience.

[0006] In a first aspect, an embodiment of the present application provides an internet insurance claims processing method, which is applied to an internet insurance claims processing system. The system includes a user terminal, a server, and several business processing terminals, each of which has different claims processing permissions. The method includes:

[0007] The server receives the claim materials uploaded by the user for the claim case through the user terminal; generates a number of claim case questions based on the claim materials and sends them to the user terminal; prompts the user to repeat each claim case question and answer each claim case question through the user terminal;

[0008] When the user repeats each claim case question and answers each claim case question, the user terminal collects the user video and sends it to the server;

[0009] The server performs text recognition on the user video to obtain voice content; divides the user video into a retelling part and an answer part according to the voice content; and performs feature recognition on the retelling part and the answer part to obtain a retelling feature and an answer feature;

[0010] The server determines the question-answer risk level of the claim case based on the repetition feature and the answer feature; determines the material risk level of the claim case based on the voice content and the claim materials; and determines the comprehensive risk level of the claim case based on the question-answer risk level and the material risk level.

[0011] The server determines a target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal; generates a claim application based on the claim materials and the voice content and sends it to the target business processing terminal;

[0012] The target business processing end makes a claim payment to the user based on the comprehensive risk level, the claim application form and the insurance contract.

[0013] Optionally, dividing the user video into a retelling part and an answer part according to the voice content includes:

[0014] identifying guide words in the voice content based on a keyword detection algorithm, and marking timestamps of when the guide words appear;

[0015] Calculating respectively a first similarity between the speech segment between each two adjacent guide words and a preset retelling template, and a second similarity between the speech segment and a preset answer template;

[0016] If the first similarity exceeds the second similarity, the speech segment between the two adjacent guide words is confirmed as a repetition segment; if the second similarity exceeds the first similarity, the speech segment between the two adjacent guide words is confirmed as an answer segment;

[0017] The retelling segments are integrated to obtain the retelling part, and the answer segments are integrated to obtain the answer part.

[0018] Optionally, the retelling feature includes a retelling expression feature and a retelling language feature, and the answer feature includes an answer expression feature and an answer language feature; the retelling expression feature includes an occurrence rate of abnormal retelling expression, and the answer expression feature includes an occurrence rate of abnormal answer expression; the retelling language feature includes an occurrence rate of abnormal retelling language, and the answer language feature includes an occurrence rate of abnormal answer language;

[0019] The performing feature recognition on the retelling part and the answer part to obtain retelling features and answer features includes:

[0020] Performing face detection on each video frame of the retelling part and the answering part, and extracting the coordinates of key points of the face in each video frame;

[0021] Determine whether abnormal expressions appear in each video frame based on the coordinates of facial key points in each video frame, and count the number of video frames with abnormal expressions;

[0022] Determining the occurrence rate of abnormal expression in the retelling part and the occurrence rate of abnormal expression in the answer part according to the number of video frames in which abnormal expression appears;

[0023] At the same time, performing speech flow analysis on the retelling part and the answering part respectively to determine the average retelling speed value and the average retelling volume value of the retelling part, and the average answering speed value and the average answering volume value of the answering part;

[0024] The occurrence rate of abnormal language in the repetition part is determined according to the average repetition speed value and the average repetition volume value; the occurrence rate of abnormal language in the answer part is determined according to the average answer speed value and the average answer volume value.

[0025] Optionally, the answer risk level includes an expression risk level and a language risk level; and determining the question-answer risk level of the claim case based on the repetition feature and the answer feature includes:

[0026] Configure corresponding differential intervals for different risk levels;

[0027] Calculating a first difference between the occurrence rate of the abnormal expression in the answer and the occurrence rate of the abnormal expression in the repetition; and determining the expression risk level according to the difference interval into which the first difference falls;

[0028] Calculating a second difference between the occurrence rate of the abnormal language in the answer and the occurrence rate of the abnormal language in the repetition; and determining the risk level of the question and answer based on the difference interval into which the second difference falls;

[0029] Determining the material risk level of the claim case based on the voice content and the claim materials includes:

[0030] Verifying the consistency of the voice content using entity alignment technology based on the claim materials to obtain a consistency verification result;

[0031] Verifying the rationality of the claim case through knowledge graph reasoning based on the claim materials to obtain a rationality verification result;

[0032] Calling third-party system data to cross-verify the correctness of the claim materials to obtain correctness verification results;

[0033] The material risk level is determined according to the consistency verification result, the rationality verification result, and the correctness verification result.

[0034] Optionally, determining the comprehensive risk level of the claim case based on the question-and-answer risk level and the material risk level includes:

[0035] Normalizing the expression risk level, the language risk level, and the material risk level, mapping them to the interval [0, 1], and obtaining standardized values ​​S1, S2, and S3 of each risk level, respectively;

[0036] The weight vector W = [w1, w2, w3] is determined by the hierarchical analysis method, where w1 is the expression risk level weight corresponding to the expression feature, w2 is the language risk level weight corresponding to the language feature, and w3 is the material risk level weight corresponding to the text feature.

[0037] The comprehensive risk score S of the claim case is determined based on the standardized value of each risk level and the weight vector. i ×w i ), i = 1, 2, 3;

[0038] The comprehensive risk level is determined according to the comprehensive risk score.

[0039] Optionally, determining the target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal includes:

[0040] Define the comprehensive risk level range corresponding to different claims processing permissions. The higher the claims processing permission, the wider the corresponding comprehensive risk level range;

[0041] Filtering a set of business processing terminals with claims processing authority based on the range of the comprehensive risk level;

[0042] The business processing end with the least number of claims cases to be processed is selected from the business processing end set as the target business processing end.

[0043] Optionally, the making a claim payment to the user based on the comprehensive risk level, the claim application and the insurance contract includes:

[0044] Determine the claim amount of the claim case based on the claim application and the insurance contract;

[0045] Determining a target claims process for the claim case based on the comprehensive risk level of the claim case;

[0046] According to the target claim settlement process, a claim settlement is paid to the user in the claim settlement amount.

[0047] In a second aspect, an embodiment of the present application provides an Internet insurance claims processing system, which includes a user terminal, a server, and several business processing terminals, each of which has different claims processing permissions:

[0048] The server is configured to receive, through the user terminal, claim materials uploaded by the user regarding a claim case; generate a number of claim case questions based on the claim materials and send them to the user terminal; and prompt the user, through the user terminal, to repeat and answer each claim case question;

[0049] The user terminal is used to collect user videos and send them to the server when the user repeats and answers the claims case questions;

[0050] The server is configured to perform text recognition on the user video to obtain voice content; divide the user video into a retelling portion and an answer portion according to the voice content; and perform feature recognition on the retelling portion and the answer portion to obtain retelling features and answer features;

[0051] The server is configured to determine a question-answer risk level of the claim case based on the repetition feature and the answer feature; determine a material risk level of the claim case based on the voice content and the claim materials; and determine a comprehensive risk level of the claim case based on the question-answer risk level and the material risk level;

[0052] The server is configured to determine a target business processing terminal based on the comprehensive risk level and the claims processing authority of each business processing terminal; generate a claim application based on the claim materials and the voice content and send it to the target business processing terminal;

[0053] The target business processing end is used to make claim payments to the user based on the comprehensive risk level, the claim application form and the insurance contract.

[0054] Optionally, dividing the user video into a retelling part and an answer part according to the voice content includes:

[0055] identifying guide words in the voice content based on a keyword detection algorithm, and marking timestamps of when the guide words appear;

[0056] Calculating respectively a first similarity between the speech segment between each two adjacent guide words and a preset retelling template, and a second similarity between the speech segment and a preset answer template;

[0057] If the first similarity exceeds the second similarity, the speech segment between the two adjacent guide words is confirmed as a repetition segment; if the second similarity exceeds the first similarity, the speech segment between the two adjacent guide words is confirmed as an answer segment;

[0058] The retelling segments are integrated to obtain the retelling part, and the answer segments are integrated to obtain the answer part.

[0059] Optionally, the retelling feature includes a retelling expression feature and a retelling language feature, and the answer feature includes an answer expression feature and an answer language feature; the retelling expression feature includes an occurrence rate of abnormal retelling expression, and the answer expression feature includes an occurrence rate of abnormal answer expression; the retelling language feature includes an occurrence rate of abnormal retelling language, and the answer language feature includes an occurrence rate of abnormal answer language;

[0060] The performing feature recognition on the retelling part and the answer part to obtain retelling features and answer features includes:

[0061] Performing face detection on each video frame of the retelling part and the answering part, and extracting the coordinates of key points of the face in each video frame;

[0062] Determine whether abnormal expressions appear in each video frame based on the coordinates of facial key points in each video frame, and count the number of video frames with abnormal expressions;

[0063] Determining the occurrence rate of abnormal expression in the retelling part and the occurrence rate of abnormal expression in the answer part according to the number of video frames in which abnormal expression appears;

[0064] At the same time, performing speech flow analysis on the retelling part and the answering part respectively to determine the average retelling speed value and the average retelling volume value of the retelling part, and the average answering speed value and the average answering volume value of the answering part;

[0065] The occurrence rate of abnormal language in the repetition part is determined according to the average repetition speed value and the average repetition volume value; the occurrence rate of abnormal language in the answer part is determined according to the average answer speed value and the average answer volume value.

[0066] Optionally, the answer risk level includes an expression risk level and a language risk level; and determining the question-answer risk level of the claim case based on the repetition feature and the answer feature includes:

[0067] Configure corresponding differential intervals for different risk levels;

[0068] Calculating a first difference between the occurrence rate of the abnormal expression in the answer and the occurrence rate of the abnormal expression in the repetition; and determining the expression risk level according to the difference interval into which the first difference falls;

[0069] Calculating a second difference between the occurrence rate of the abnormal language in the answer and the occurrence rate of the abnormal language in the repetition; and determining the risk level of the question and answer based on the difference interval into which the second difference falls;

[0070] Determining the material risk level of the claim case based on the voice content and the claim materials includes:

[0071] Verifying the consistency of the voice content using entity alignment technology based on the claim materials to obtain a consistency verification result;

[0072] Verifying the rationality of the claim case through knowledge graph reasoning based on the claim materials to obtain a rationality verification result;

[0073] Calling third-party system data to cross-verify the correctness of the claim materials to obtain correctness verification results;

[0074] The material risk level is determined according to the consistency verification result, the rationality verification result, and the correctness verification result.

[0075] Optionally, determining the comprehensive risk level of the claim case based on the question-and-answer risk level and the material risk level includes:

[0076] Normalizing the expression risk level, the language risk level, and the material risk level, mapping them to the interval [0, 1], and obtaining standardized values ​​S1, S2, and S3 of each risk level, respectively;

[0077] The weight vector W = [w1, w2, w3] is determined by the hierarchical analysis method, where w1 is the expression risk level weight corresponding to the expression feature, w2 is the language risk level weight corresponding to the language feature, and w3 is the material risk level weight corresponding to the text feature.

[0078] The comprehensive risk score S of the claim case is determined based on the standardized value of each risk level and the weight vector. i ×w i ), i = 1, 2, 3;

[0079] The comprehensive risk level is determined according to the comprehensive risk score.

[0080] Optionally, determining the target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal includes:

[0081] Define the comprehensive risk level range corresponding to different claims processing permissions. The higher the claims processing permission, the wider the corresponding comprehensive risk level range;

[0082] Filtering a set of business processing terminals with claims processing authority based on the range of the comprehensive risk level;

[0083] The business processing end with the least number of claims cases to be processed is selected from the business processing end set as the target business processing end.

[0084] Optionally, the making a claim payment to the user based on the comprehensive risk level, the claim application and the insurance contract includes:

[0085] Determine the claim amount of the claim case based on the claim application and the insurance contract;

[0086] Determining a target claims process for the claim case based on the comprehensive risk level of the claim case;

[0087] According to the target claim settlement process, a claim settlement is paid to the user in the claim settlement amount.

[0088] In a third aspect, an embodiment of the present application provides a computer device comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the Internet insurance claims processing method described in any optional implementation method of the first aspect are performed.

[0089] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the Internet insurance claims processing method described in any optional implementation method of the first aspect above are executed.

[0090] The technical solutions provided by this application include but are not limited to the following beneficial effects:

[0091] This application first receives claim documents through a server and generates interactive questions. Based on key information in the documents, it guides users to retell and answer questions, ensuring the accuracy and completeness of their case descriptions. It also proactively captures user voice and video data, laying the foundation for subsequent multimodal risk analysis and preventing audit bias caused by missing information. The user then captures and uploads interactive video, recording non-textual information such as facial expressions and voice intonation during the user's retelling and answering. This enables the server to capture behavioral data missing from traditional claims processing, providing a visual basis for uncovering potential fraud clues. The server then uses text recognition, segmentation, and feature extraction to convert the video data into structured facial and language features. Through phased (retelling / answering) comparative analysis, it accurately captures user behavioral differences across different interaction stages, providing fine-grained behavioral dimensionality for risk assessment. The behavioral features from the question-and-answer phase are then combined with consistency verification of the document text to construct a dual "behavior-document" risk assessment system: the question-and-answer risk level reflects the credibility of the user's statement, while the document risk level verifies the authenticity of the factual evidence. The two are integrated to form a three-dimensional comprehensive risk assessment, transcending the limitations of single-dimensional audits and enhancing the comprehensiveness of risk assessment. The system then dynamically matches processing authority based on the overall risk level, assigning high-risk cases to processing terminals with higher review authority. It also optimizes resource scheduling based on processing terminal load, achieving differentiated processing with "rapid transfer of low-risk cases and detailed review of high-risk cases." This not only improves overall claims processing efficiency but also ensures risk management. Finally, the target processing terminal implements a tiered payment process based on risk level, application, and contract: low-risk cases trigger quick payment, shortening user wait times; high-risk cases initiate in-depth review, strictly verifying the claim amount in accordance with contract terms. This ensures claims processing efficiency while effectively reducing the risk of erroneous and missed claims, thereby strengthening the compliance and security of insurance business.

[0092] The above steps build a full-link intelligent claims system covering "information collection-feature analysis-risk assessment-resource scheduling-payment execution" through multimodal data collection and analysis, dual risk verification mechanism and dynamic process adaptation. It can improve the fraud identification rate, discover hidden contradictions, achieve low-risk quick compensation and high-risk precise review, optimize resource allocation, reduce the risk of erroneous compensation, and significantly improve claims efficiency and user experience.

[0093] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0095] Figure 1 A flowchart of a method for processing an internet insurance claim settlement business provided by the first embodiment of the present invention is shown;

[0096] Figure 2 A flowchart of a user video segmentation method provided by the first embodiment of the present invention is shown;

[0097] Figure 3 A flow chart of a feature recognition method provided by the first embodiment of the present invention is shown;

[0098] Figure 4 A flowchart of a method for determining a question-and-answer risk level provided in the first embodiment of the present invention is shown;

[0099] Figure 5 A flow chart of a method for determining a material risk level provided in the first embodiment of the present invention is shown;

[0100] Figure 6 A flow chart of a comprehensive risk level determination method provided by the first embodiment of the present invention is shown;

[0101] Figure 7 A flowchart of a method for determining a target service processing terminal provided by the first embodiment of the present invention is shown;

[0102] Figure 8 A flowchart of a claim payment method provided in the first embodiment of the present invention is shown;

[0103] Figure 9 A schematic diagram showing the structure of an Internet insurance claims processing system provided by the second embodiment of the present invention is shown;

[0104] Figure 10 A schematic structural diagram of a computer device provided in the third embodiment of the present invention is shown. DETAILED DESCRIPTION

[0105] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0106] Example 1

[0107] To facilitate understanding of this application, Figure 1 The flowchart of an Internet insurance claims processing method provided by Example 1 of the present invention is shown to provide a detailed description of Example 1 of the present application.

[0108] See also Figure 1 As shown, Figure 1 A flowchart of an internet insurance claims processing method provided by a first embodiment of the present invention is shown. The method is applied to an internet insurance claims processing system. The system includes a user terminal, a server, and several business processing terminals, each of which has different claims processing permissions. The method includes steps S101 to S106:

[0109] S101: The server receives claim materials uploaded by the user for the claim case through the user terminal; generates a number of claim case questions based on the claim materials and sends them to the user terminal, and prompts the user to repeat each claim case question and answer each claim case question through the user terminal.

[0110] Specifically, the server receives claim materials uploaded by users (such as accident identification documents and medical invoices) through the user terminal, digitizes the paper documents using OCR technology, and verifies the integrity of the materials. Based on key information in the materials (such as the type of accident and the damaged items), the intelligent question generation model automatically generates several claim case questions (such as "Please repeat the specific time and process of the accident"), and prompts the user to repeat the question and answer it in a modal box through the user terminal. At the same time, a countdown mechanism is activated to ensure interaction efficiency.

[0111] S102: When the user repeats each claim case question and answers each claim case question, the user terminal collects the user video and sends it to the server.

[0112] Specifically, when the user begins to repeat and answer questions, the user's camera and microphone capture video and audio data, compress the video stream using H.264 encoding, and perform real-time noise reduction processing using WebRTC technology. The system synchronously collects the timestamps of video and audio frames to ensure data alignment and pre-processes the data locally in slices (e.g., 5 seconds per slice). Data is automatically cached in the event of network anomalies and transmitted to the server via WebSocket upon recovery. Ambient lighting and noise levels are also monitored in real time, prompting users to adjust shooting conditions.

[0113] S103: The server performs text recognition on the user video to obtain voice content; divides the user video into a retelling part and an answer part according to the voice content; and performs feature recognition on the retelling part and the answer part to obtain retelling features and answer features.

[0114] Specifically, the server performs speech and text recognition on the received user videos and uses the pre-trained Wav2Vec2 model to convert the audio into a text stream with timestamps. The position of the guide words is marked through the keyword detection algorithm (such as "retell" and "answer"), and the semantic similarity between the voice segments and the preset templates between adjacent guide words is calculated to divide the retelling part and the answer part. Face detection is performed on the video frames, and 68 facial key points are extracted to calculate micro-expression features (such as blinking frequency and lip shape changes). At the same time, the rhythmic features of the voice stream such as speech speed and volume are analyzed to generate the expression abnormality rate and language abnormality rate in the retelling and answering stages.

[0115] S104: The server determines the question-and-answer risk level of the claim case based on the repetition feature and the answer feature; determines the material risk level of the claim case based on the voice content and the claim materials; and determines the comprehensive risk level of the claim case based on the question-and-answer risk level and the material risk level.

[0116] Specifically, based on the feature data of the retelling and answering stages, the server calculates the difference in expression abnormality rate and language abnormality rate, matches the preset risk interval to determine the risk level of the question and answer (low / medium / high). At the same time, the consistency of the voice content and the claim materials is verified through entity alignment technology, and the rationality of the event is inferred using the knowledge graph (such as the correlation between "rainy day accident" and "brake marks"), and third-party data (such as traffic police records) is called to cross-verify the correctness of the materials to generate the material risk level. Finally, the two types of risks are integrated through the hierarchical analysis method, and the comprehensive risk score is calculated after normalization to divide the final risk level.

[0117] S105: The server determines a target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal; generates a claim application according to the claim materials and the voice content and sends it to the target business processing terminal.

[0118] Specifically, the server matches the business processing terminal's authority range based on the comprehensive risk level (e.g., the primary terminal handles cases with a score of ≤50), screens out the set of terminals with processing authority, and selects the target processing terminal with the fewest pending cases through a load balancing algorithm. Based on the claim materials and voice text, the server utilizes a template engine to automatically generate a claim application containing key information (such as the policy number and loss amount), integrates an electronic signature function, verifies required fields, and ultimately sends the application and risk assessment report to the target processing terminal.

[0119] S106: The target business processing end makes a claim payment to the user based on the comprehensive risk level, the claim application form and the insurance contract.

[0120] Specifically, the target business processing end determines the claims process based on the overall risk level: low-risk cases trigger automatic payment, with real-time transfers via blockchain smart contracts; medium-risk cases initiate manual review, with the payout amount calculated based on the insurance contract terms (deductibles and liability ratio). High-risk cases initiate on-site inspections and legal investigations, with payment executed after confirmation. Upon payment completion, an electronic claim notice is generated, including a payment receipt and a progress inquiry link, which is delivered to the user via SMS and app, completing the entire process.

[0121] In an alternative embodiment, see Figure 2 As shown, Figure 2 A flowchart of a user video segmentation method provided by the first embodiment of the present invention is shown, wherein the method of segmenting the user video into a retelling part and an answer part according to the voice content includes steps S201 to S204:

[0122] S201: Identify the guide words in the voice content based on a keyword detection algorithm, and mark the timestamp of the guide words appearing.

[0123] Specifically, an end-to-end speech recognition model (such as Wav2Vec2) is used to convert the audio stream of the user's video into a text stream, synchronously generating frame-level timestamps accurate to 50ms. A finite state machine (FSM) is used to construct a keyword recognizer, which performs a sliding window scan on the text stream to identify guide words such as "retell" and "answer," supporting fuzzy matching with an edit distance ≤ 1 (for example, automatically correcting "retell" to "retell"). To address the timing offset issue in speech recognition, audio energy detection (short-term energy > threshold) is used to relocate keyword boundaries, extending them forward by 0.3 seconds to include the complete semantic starting point. A language model (such as GPT-2) is then used to correct specialized terminology (for example, correcting "brain vibration" to "concussion"), ultimately generating a list of guide word locations with accurate timestamps.

[0124] S202: Calculating a first similarity between the speech segment between every two adjacent guide words and a preset retelling template, and a second similarity between the speech segment and a preset answer template.

[0125] Specifically, a dynamically updated insurance template library was constructed: retelling templates use a fill-in-the-blank structure (e.g., "The accident occurred at {X} and {Y}"), while answer templates are designed as free-form expressions (e.g., "An accident occurred at {time} and {location}, resulting in {amount} of loss"). These two template types automatically switch based on the case type (auto insurance / medical insurance). For each speech segment between two adjacent guide words, the BERT-base-Chinese model extracts semantic vectors, and calculates cosine similarity with the retelling and answer templates. Named Entity Recognition (NER) is also used to extract key entities such as time and amount, calculate the entity matching rate, and analyze the speech speed and stress distribution of the speech segment (e.g., a more uniform speech speed during the retelling phase), forming a multi-dimensional similarity scoring system.

[0126] S203: If the first similarity exceeds the second similarity, the speech segment between the two adjacent guide words is confirmed as a repetition segment; if the second similarity exceeds the first similarity, the speech segment between the two adjacent guide words is confirmed as an answer segment.

[0127] Specifically, a dual-threshold dynamic decision-making mechanism is employed: when a voice clip's first similarity with the retelling template is ≥0.85 and exceeds 15% of the similarity with the answer template, it is identified as a retelling clip; when its second similarity with the answer template is ≥0.75 and exceeds 10% of the similarity with the retelling template, it is identified as an answer clip. The threshold can be adaptively adjusted based on historical data (e.g., lowered to 0.8 in dialect regions). If the difference between the two similarities is less than 5%, visually assisted verification is initiated: the MTCNN algorithm detects facial key points in the video frame, analyzes the synchronization of lip movement features with speech (e.g., lip shape better matches the question voice during retelling), or resolves conflicts based on the temporal logic of the guide word (e.g., the clip immediately following "retell" is prioritized as the retelling content).

[0128] S204: Integrate the retelling segments to obtain the retelling part, and integrate the answer segments to obtain the answer part.

[0129] Specifically, the temporal and spatial continuity of similar segments is optimized: adjacent repetition segments with an interval of ≤1 second are merged, and those with an interval of >1 second are retained. Voice activity detection (VAD) is used to fill in missing segments caused by failed guide word detection to ensure video integrity. For multi-round interaction scenarios involving "question-repetition-answer," all segments are arranged chronologically, and a list of repetition intervals (such as [start1, end1] and [start2, end2]) and answer intervals with timestamps are generated, ultimately integrating them into a continuous video segment. To reduce bandwidth consumption, edge computing is used on the user side to perform keyword pre-detection and coarse segmentation. Only feature vectors are uploaded to the server, and a meta-learning algorithm is used to simultaneously optimize the template library to improve segment segmentation accuracy in small sample scenarios.

[0130] In an optional embodiment, the repetition features include repetition expression features and repetition language features, and the answer features include answer expression features and answer language features; the repetition expression features include the occurrence rate of abnormal repetition expression, and the answer expression features include the occurrence rate of abnormal answer expression; the repetition language features include the occurrence rate of abnormal repetition language, and the answer language features include the occurrence rate of abnormal answer language.

[0131] See also Figure 3 As shown, Figure 3 A flow chart of a feature recognition method provided by the first embodiment of the present invention is shown, wherein the feature recognition of the repetition part and the answer part to obtain the repetition feature and the answer feature includes steps S301 to S305:

[0132] S301: Performing face detection on each video frame of the retelling part and the answering part, and extracting the coordinates of key facial points in each video frame.

[0133] Specifically, the MTCNN (Multi-Task Cascaded Convolutional Neural Network) algorithm is used to detect facial regions in video frames during the retelling and answering phases. First, a three-layer cascade network (Proposal Network, Refine Network, Output Network) is used to quickly locate the facial bounding box. Then, for each detected face, 68 key point coordinates (including eyebrows, eyes, nose, mouth, jawline, and other key areas) are extracted. To improve detection accuracy under complex lighting conditions, the video frames are preprocessed using histogram equalization and noise filtering. A dynamic threshold adjustment strategy is also employed. When a face confidence score of less than 0.9 is detected, a secondary detection is automatically triggered to ensure that the error in key point coordinate extraction is ≤1 pixel.

[0134] S302: Determine whether abnormal expressions appear in each video frame based on the coordinates of facial key points in each video frame, and count the number of video frames in which abnormal expressions appear.

[0135] Specifically, facial dynamic features are calculated based on the coordinates of 68 key points, including lip aspect ratio (MAR), pupil dilation rate, and blinking frequency. For the lip aspect ratio, the ratio of vertical distance to horizontal distance is calculated through the mouth key points. MAR>0.3 is judged as abnormal mouth opening; for pupil dilation rate: the change in pupil area of ​​both eyes is calculated, and a single-frame change rate>15% is judged as abnormal; for blinking frequency: the duration of eye key point closure is calculated, and a frequency>15 times / minute is judged as abnormal. A sliding window (window size 5 frames) is used to analyze continuous expression changes. When a frame meets more than two abnormal features at the same time or a single feature exceeds the threshold by 1.5 times, it is marked as an abnormal expression frame. The LSTM network is used to learn the temporal pattern of normal expressions, and a secondary verification is performed on sudden abnormalities (such as eye closure not caused by blinking) to reduce misjudgment.

[0136] S303: Determine the occurrence rate of abnormal expression in the retelling part and the occurrence rate of abnormal expression in the answer part according to the number of video frames in which abnormal expression appears.

[0137] Specifically, the total number of video frames N and the number of abnormal expression frames M in the retelling and answering stages are counted respectively, and the calculation formula is: abnormal expression occurrence rate = (M / N) × 100%.

[0138] For the retelling phase, a baseline threshold is set: an abnormality rate >30% is considered high risk. During the answering phase, the threshold is adjusted to >40% due to natural expression differences. When abnormal expressions appear in three consecutive frames, the weight is automatically increased (for example, a single-frame abnormality is counted 1.5 times) to highlight persistent abnormal behavior. The distribution of abnormality types (such as the proportion of abnormal blinks and abnormal lip movements) is also recorded to provide fine-grained features for subsequent risk assessment.

[0139] S304: Simultaneously, performing speech flow analysis on the retelling part and the answering part respectively to determine an average retelling speed and an average retelling volume of the retelling part, and an average answering speed and an average answering volume of the answering part.

[0140] Specifically, the audio streams in the retelling and answering stages are framed (frame length 25ms, frame shift 10ms) and then speech rate calculation and volume analysis are performed. For speech rate analysis, after converting the speech to text, the average number of words per second is counted. The formula is: speech rate = (total number of words in the text / speech duration) × 60. The normal range for the retelling stage is set to 100-160 words / minute, and the answering stage is adjusted to 80-200 words / minute. For volume analysis: extract the short-time energy of each frame of audio, calculate the average volume value of the entire speech (unit: dB), and count the volume fluctuation amplitude (the difference between the maximum and minimum values).

[0141] S305: Determine the occurrence rate of abnormal language in the repetition part according to the average repetition speed value and the average repetition volume value; determine the occurrence rate of abnormal language in the answer part according to the average answer speed value and the average answer volume value.

[0142] Specifically, an abnormality determination model is established by combining speech speed and volume features. Retelling stage: speaking speed > 180 words / minute or < 80 words / minute is determined as abnormal speaking speed; volume fluctuation > 20dB or average volume < 40dB is determined as abnormal volume; abnormality rate = (number of frames with abnormal speaking speed + number of frames with abnormal volume) / total number of frames × 100%. Answering stage: semantic coherence detection is added (BERT similarity between text and question < 0.7), and each semantic deviation is converted into 10 frames of abnormality; abnormality rate = (number of frames with abnormal speaking speed + number of frames with abnormal volume + number of frames converted from semantic deviation) / total number of frames × 100%. The threshold is dynamically adjusted through the Gaussian mixture model (GMM) to adapt to the language habits of different users. For example, the speech speed threshold of dialect users can be increased by 20%.

[0143] In an optional embodiment, the answer risk level includes an expression risk level and a language risk level; see Figure 4 As shown, Figure 4 A flowchart of a method for determining a question-and-answer risk level provided in the first embodiment of the present invention is shown, wherein the question-and-answer risk level of the claim case is determined based on the repetition feature and the answer feature, including steps S401 to S403:

[0144] S401: Configure corresponding difference intervals for different risk levels.

[0145] Specifically, based on the characteristic distribution of normal and fraudulent cases in historical insurance claims data, the K-means clustering algorithm clusters the difference in the occurrence of abnormal expressions / language during the answering and retelling phases, creating three risk ranges. For example, after analyzing 100,000 historical cases, the low-risk range for expression risk was [-5%, 5%] (a close difference indicates emotional stability), the medium-risk range was (5%, 15%] or (-15%, -5%) (a small difference may indicate mild tension), and the high-risk range was >15% or <-15% (a significant difference may indicate deception). Due to the greater freedom of expression, the language risk ranges were set at [-8%, 8%], (8%, 20%] or (-20%, -8%), and >20% or <-20%, respectively. The thresholds for these ranges are dynamically adjusted monthly based on the latest case data. If new fraudulent tactics are discovered that lead to an overall increase in abnormal expression differences, the high-risk threshold will be automatically raised from 15% to 18%.

[0146] S402: Calculate a first difference between the occurrence rate of the abnormal expression in the answer and the occurrence rate of the abnormal expression in the repetition; and determine the expression risk level according to the difference interval into which the first difference falls.

[0147] Specifically, the difference between the rate of abnormal expressions in answers and the rate of abnormal expressions in retelling (the first difference) is calculated using the formula: first difference = abnormal rate in answers - abnormal rate in retelling. For example, if the abnormal rate in the answering stage is 38% and that in the retelling stage is 20%, then the first difference = 18%, which falls into the medium-risk range (5%, 15%) and is determined to be a medium-risk expression. The system adaptively adjusts the threshold based on the type of case: for car insurance cases, the high-risk threshold is raised to 20% because the on-site description is prone to causing normal tension; for medical insurance cases involving medical terminology, the medium-risk range is narrowed to (5%, 12%). For cases in which abnormal expressions appear for more than three consecutive frames, the difference is automatically multiplied by 1.2 times the weight to highlight the risk of persistent abnormal behavior.

[0148] S403: Calculate a second difference between the occurrence rate of the abnormal language in the answer and the occurrence rate of the abnormal language in the repetition; and determine the question and answer risk level according to the difference interval into which the second difference falls.

[0149] Specifically, the difference between the rate of abnormal language in answers and the rate of abnormal language in retelling (the second difference) is calculated, and the formula is: the second difference = abnormal rate in answers - abnormal rate in retelling. If the abnormal rate in the answering stage is 30% and that in the retelling stage is 10%, then the second difference = 20%, which is determined to be a medium language risk. Introduce an attention mechanism to dynamically adjust feature weights: the weight of language features in small cases accounts for 60% (focusing on efficiency), and the weight of facial expression features in large cases is increased to 70% (focusing on accuracy). When the difference between the facial expression and language risk levels is ≥2 levels (such as high risk for facial expression and low risk for language), the manual review process is initiated to eliminate non-fraud factors such as equipment failure or user nervousness through semantic coherence detection (such as BERT similarity <0.7) and third-party data cross-validation (such as retrieving traffic police records).

[0150] See also Figure 5 As shown, Figure 5 A flow chart of a method for determining a material risk level provided by a first embodiment of the present invention is shown, wherein determining the material risk level of the claim case based on the voice content and the claim materials includes steps S501 to S504:

[0151] S501: Verify the consistency of the voice content using entity alignment technology based on the claim materials to obtain a consistency verification result.

[0152] Specifically, named entity recognition (NER) technology extracts key entities (such as time, location, and amount) from claim documents (such as accident reports and medical invoices) and spoken content, constructing an entity alignment matrix. A dynamic time warping (DTW) algorithm is used to calculate the similarity between entities, with a threshold of 0.85 (for example, the similarity between "October 1, 2023" and "2023-10-01" is 0.95). For numerical entities (such as loss amount), a ±5% error margin is allowed. If the accident time recorded in the document is "2023 / 10 / 1" and the spoken description is "September 30, 2023", it is marked as a time conflict. The system assigns a weighted score to all entities (for example, 0.3 for time and amount, and 0.2 for location). A high-risk alert is triggered when the consistency score falls below 0.7.

[0153] S502: Verify the rationality of the claim case through knowledge graph reasoning based on the claim materials to obtain a rationality verification result.

[0154] Specifically, based on the insurance domain knowledge graph (e.g., "Rainy days → increased braking distance → increased rear-end collision probability"), rule inference is performed using the Neo4j graph database. A mapping relationship between accident type and loss characteristics is constructed (e.g., "vehicle collision → vehicle dent + broken glass"). When the voice description of the accident cause is "sudden braking and rear-end collision in heavy rain," but the document only records "vehicle scratches," a reasonableness query is triggered. A Bayesian network is used to calculate event probability: if the physical relationship between "brake mark length" and "described vehicle speed" does not conform to the kinematic formula (error > 15%), the claim is marked as unreasonable. For medical claims, the logic of the "symptom-diagnosis-treatment" chain is verified (e.g., a cold diagnosis leads to heart surgery), and paths that do not conform to medical common sense are classified as high-risk.

[0155] S503: Calling third-party system data to cross-verify the correctness of the claim materials to obtain a correctness verification result.

[0156] Specifically, the authenticity of the materials is verified through API connection to the traffic police system, hospital HIS system, meteorological bureau and other third-party data sources. For traffic accidents, the reporting time, accident location, responsibility determination and other information are compared. If the accident time in the material is 10 am on a weekday, but the third-party record shows that there is no accident alarm during this period, it is marked as suspicious. When verifying medical invoices, check the matching degree of the hospital's medical records, drug list and reimbursement amount. For example, if the invoice amount is 5,000 yuan, but the HIS system shows that the actual consumption is only 3,000 yuan, the correctness score drops to 0.6 (out of 1.0). Set the verification item weights: key items (such as accident identification documents) have a weight of 0.5, auxiliary items (such as meteorological certificates) have a weight of 0.2, and the final correctness score = Σ(verification item score × weight).

[0157] S504: Determine the material risk level according to the consistency verification result, the rationality verification result, and the correctness verification result.

[0158] Specifically, the risk score is calculated using the analytic hierarchy process (AHP) to assign weights to the consistency verification results, the rationality verification results, and the correctness verification results (for example, the weight assignment is: consistency score weight is 0.4, rationality score weight is 0.3, and correctness score weight is 0.3). The weighted sum is then taken to obtain the material risk score, and the score is finally mapped to the risk level. For example: material risk score = consistency score × 0.4 + rationality score × 0.3 + correctness score × 0.3. Low risk: score ≥ 0.85 (entities are highly consistent, logic is reasonable, and third-party verification is correct); medium risk: score ∈ [0.6, 0.85) (there are 1-2 minor anomalies); high risk: score < 0.6 (there are major contradictions or verification failed). When the following situations occur, it is directly judged as high risk: there are more than 3 inconsistencies in key entities such as time / amount, the knowledge graph reasoning probability is < 0.3, and the third-party verification error rate is > 20%. The system generates a risk detail report, marking specific contradictions (such as "the amount of medical invoice does not match the medical record"), providing clear guidance for subsequent investigations.

[0159] In an alternative embodiment, see Figure 6 As shown, Figure 6 A flowchart of a comprehensive risk level determination method provided by the first embodiment of the present invention is shown, wherein the comprehensive risk level of the claim case is determined based on the question-and-answer risk level and the material risk level, including steps S601 to S604:

[0160] S601: normalize the expression risk level, the language risk level, and the material risk level, map them to the interval [0, 1], and obtain standardized values ​​S1, S2, and S3 of each risk level, respectively.

[0161] Specifically, in order to unify the measurement standards of risk levels in different dimensions, the expression risk level, language risk level and material risk level need to be mapped to the interval [0,1]. The general implementation method is to establish a linear or nonlinear mapping relationship based on the original division threshold of each risk level. For example, for the expression risk level, if the original low risk is defined as the difference in the occurrence rate of abnormal expressions ≤5%, then the interval is mapped to [0,0.2]; medium risk (difference ∈ (5%, 15%]) is mapped to (0.2, 0.5]; high risk (difference > 15%) is mapped to (0.5, 1]. Similar interval mapping strategies are also adopted for language and material risk levels. In order to highlight the impact of high-risk factors, nonlinear mapping functions (such as Sigmoid function) are often used to further amplify the values ​​of high-risk intervals to ensure that high-risk features occupy a greater weight in the comprehensive assessment. Finally, the risk levels of the three dimensions are converted into standardized values ​​S1, S2, and S3, respectively, to provide a unified data basis for subsequent weighted calculations.

[0162] S602: Determine the weight vector W = [w1, w2, w3] through the hierarchical analysis method, where w1 is the expression risk level weight corresponding to the expression feature, w2 is the language risk level weight corresponding to the language feature, and w3 is the material risk level weight corresponding to the text feature.

[0163] Specifically, first, an assessment team consisting of insurance industry experts and risk control personnel is organized. Based on their business experience and professional knowledge, a pairwise comparison is conducted on the three risk dimensions of expression, language, and materials to construct a judgment matrix. For example, in the judgment matrix, if the material risk is considered to be more important than the expression risk, a value greater than 1 (such as 3) is filled in the corresponding cell; otherwise, a value less than 1 is filled in. Next, the maximum eigenvalue of the judgment matrix and its corresponding eigenvector are calculated using the square root method or the eigenvector method, and normalized to obtain the weight vector W = [w1, w2, w3]. To ensure the rationality of the judgment matrix, a consistency test is required to calculate the consistency index CI and the random consistency index RI. When the consistency ratio CR = CI / RI < 0.1, the judgment matrix is ​​considered to have acceptable consistency and the weight vector is valid. In addition, the weight vector can be dynamically adjusted based on factors such as case type and business needs. For example, for small cases, the material risk weight can be appropriately reduced and the interactive feature weight can be increased.

[0164] S603: Determine the comprehensive risk score S of the claim case based on the standardized value of each risk level and the weight vector = ∑(S i ×w i ), i=1, 2, 3.

[0165] Specifically, based on the normalized risk level standardized values ​​S1, S2, S3 and the determined weight vector W = [w1, w2, w3], the weighted summation formula S = ∑(S i ×w i )(i=1,2,3) calculate the comprehensive risk score. The general implementation process is to multiply S1, S2, and S3 by the corresponding weights w1, w2, and w3 respectively, and then add the products. For example, if S1=0.6, w1=0.15; S2=0.7, w2=0.25; S3=0.8, w3=0.6, then the comprehensive risk score S=0.6×0.15+0.7×0.25+0.8×0.6=0.755. In order to strengthen the impact of high-risk factors, a risk amplifier mechanism can be introduced. When the risk of any dimension reaches a high level, the comprehensive score is multiplied by an amplification factor (such as 1.2) so that the final score can better reflect the actual risk level and provide an accurate basis for risk level determination.

[0166] S604: Determine the comprehensive risk level according to the comprehensive risk score.

[0167] Specifically, based on the calculated comprehensive risk score S, it is mapped to a pre-set risk level interval to determine the comprehensive risk level of the claim case. The general practice is to divide the risk into three intervals: low, medium, and high. For example, the low risk interval is set to S < 0.4, the medium risk interval is 0.4 ≤ S < 0.7, and the high risk interval is S ≥ 0.7. When the comprehensive score falls into different intervals, the corresponding risk level is assigned. For claims with high risk levels (S ≥ 0.7), the system automatically triggers the manual review process and generates a detailed risk assessment report, marking the risk scores of each dimension, key contradictions and recommended verification directions; applications with medium and low risk levels are processed according to the preset automated or semi-automated processes. At the same time, the system can regularly optimize the risk level interval division based on historical claims data and actual processing results to improve the accuracy of risk assessment and business adaptability.

[0168] In an alternative embodiment, see Figure 7 As shown, Figure 7 A flowchart of a method for determining a target business processing terminal provided by a first embodiment of the present invention is shown, wherein the method for determining the target business processing terminal based on the comprehensive risk level and the claims processing authority of each business processing terminal includes steps S701 to S703:

[0169] S701: Define the comprehensive risk level range corresponding to different claims processing permissions, where the higher the claims processing permission, the larger the corresponding comprehensive risk level range.

[0170] Specifically, in order to achieve hierarchical processing of claims business, the corresponding relationship between the authority and risk level of each business processing terminal needs to be set in advance. The general practice is to divide the comprehensive risk level into multiple intervals based on the complexity of the business and the needs of risk management, and assign corresponding processing scopes to business terminals with different processing permissions. For example, the primary business processing terminal has lower authority and only handles claims cases with a comprehensive risk score in the range of 0-0.4 (low risk); the intermediate processing terminal can handle cases with a score of 0.4-0.7 (medium risk); and the senior processing terminal is responsible for complex cases of 0.7 and above (high risk). Through such a setting, it is ensured that high-risk cases are reviewed by experienced and more authoritative processing terminals, thereby reducing the risk of claims, while allowing low-risk cases to pass through the primary processing terminal quickly, improving overall processing efficiency.

[0171] S702: Filter out a set of business processing terminals with claims processing authority based on the range of the comprehensive risk level.

[0172] Specifically, based on the authority-risk level correspondence established in step S701, the system automatically selects business processing terminals with the authority to handle the current claim case. In practice, the system compares the calculated comprehensive risk score with the risk level range corresponding to each processing terminal. If a processing terminal's risk level range includes that score, it is included in the set of business processing terminals that can handle it. For example, if a claim case has a comprehensive risk score of 0.55, which falls into the medium-risk range, the system will select all intermediate-level business processing terminals with authority to handle medium-risk cases, forming a set of candidate processing terminals, which will provide the basis for determining the final processing terminal in the next step.

[0173] S703: Select the business processing end with the least number of claims cases to be processed from the business processing end set as the target business processing end.

[0174] Specifically, in the set of selected business processing terminals, a load balancing strategy is used to determine the target business processing terminal. The system monitors the number of pending claims cases of each business processing terminal in real time, and selects the terminal with the least pending applications as the target business processing terminal for the final execution of this claims business. This process can obtain the task queue length of each processing terminal through database query, or use the task scheduling module in the distributed system for statistics. For example, there are three intermediate business processing terminals in the set, and their pending applications are 10, 8, and 5 respectively. The system will select the terminal with 5 pending applications as the target business processing terminal, thereby balancing the workload of each processing terminal, reducing case backlogs, and ensuring that the claims business is carried out efficiently and orderly.

[0175] In an alternative embodiment, see Figure 8 As shown, Figure 8A flowchart of a claim payment method provided in the first embodiment of the present invention is shown, wherein the claim payment to the user is made based on the comprehensive risk level, the claim application and the insurance contract, including steps S801 to S803:

[0176] S801: Determine the claim amount of the claim case based on the claim application and the insurance contract.

[0177] Specifically, the final compensation amount is calculated using a rules engine based on the loss items extracted from the claim application (such as vehicle repair costs and medical expenses) and the insurance contract terms. A common implementation involves first using template parsing technology to extract key data from the application (such as a loss amount of 10,000 yuan). The calculation is then based on the deductible clause in the contract (such as an absolute deductible of 500 yuan) and the compensation ratio (such as 80% compensation for full liability). For example, if the loss in a car accident is 10,000 yuan, and the contract stipulates a 500 yuan deductible and an 80% compensation ratio, the claim amount = (10,000 - 500 yuan) × 80% = 7,600 yuan. For complex cases (such as those involving multiple losses or liability divisions), a decision tree model is used to call different calculation formulas, while verifying the consistency of the calculated results with supporting documents such as invoices and damage assessments to ensure the accuracy of the amount.

[0178] S802: Determine a target claims process for the claims case based on the comprehensive risk level of the claims case.

[0179] Specifically, the preset differentiated process templates are matched according to the comprehensive risk level (low / medium / high), and the processing path is dynamically generated through the workflow engine. The general implementation logic is: low-risk cases (S < 0.4) trigger the automatic claims process without manual intervention; medium-risk cases (0.4 ≤ S < 0.7) start a semi-automatic process, such as system pre-audit + manual review of key items; high-risk cases (S ≥ 0.7) enter the full manual review process, which requires on-site inspection or third-party data verification. For example, low-risk cases with a comprehensive score of 0.3 directly enter the automatic payment link, while high-risk cases with a score of 0.8 must first be verified by the investigator to verify the authenticity of the accident, and then undergo multi-level approval. The process configuration supports customization on the business side, and the sequence of links and review nodes corresponding to each risk level can be adjusted through the visual editor.

[0180] S803: Pay the claim to the user in the claim amount according to the target claim process.

[0181] Specifically, based on the identified target process, the payment interface is called to complete the funds transfer and generate a receipt. Common implementations include: For low-risk processes, payments are automatically executed (e.g., to a user's bank account) via blockchain smart contracts, with on-chain evidence stored to generate an unalterable payment receipt. For medium-risk processes, payment is triggered through the core business system after manual verification and requires the user's electronic signature for confirmation. For high-risk processes, after all investigations and approvals are completed, financial personnel handle large transfers offline and simultaneously upload the bank receipt to the system. Once payment is completed, the system automatically sends a text message / app notification to the user, along with an electronic claim notice (including payment amount, time limit, and other information), and updates the claim status to "paid," creating a closed-loop business model.

[0182] Example 2

[0183] See also Figure 9 As shown, Figure 9 A schematic diagram of the structure of an Internet insurance claims processing system provided by Embodiment 2 of the present invention is shown, wherein the system includes a user terminal 901, a server 902, and several business processing terminals 903, each of which has different claims processing permissions:

[0184] The server is configured to receive, through the user terminal, claim materials uploaded by the user regarding a claim case; generate a number of claim case questions based on the claim materials and send them to the user terminal; and prompt the user, through the user terminal, to repeat and answer each claim case question;

[0185] The user terminal is used to collect user videos and send them to the server when the user repeats and answers the claims case questions;

[0186] The server is configured to perform text recognition on the user video to obtain voice content; divide the user video into a retelling portion and an answer portion according to the voice content; and perform feature recognition on the retelling portion and the answer portion to obtain retelling features and answer features;

[0187] The server is configured to determine a question-answer risk level of the claim case based on the repetition feature and the answer feature; determine a material risk level of the claim case based on the voice content and the claim materials; and determine a comprehensive risk level of the claim case based on the question-answer risk level and the material risk level;

[0188] The server is configured to determine a target business processing terminal based on the comprehensive risk level and the claims processing authority of each business processing terminal; generate a claim application based on the claim materials and the voice content and send it to the target business processing terminal;

[0189] The target business processing end is used to make claim payments to the user based on the comprehensive risk level, the claim application form and the insurance contract.

[0190] In an optional embodiment, dividing the user video into a retelling part and an answer part according to the voice content includes:

[0191] identifying guide words in the voice content based on a keyword detection algorithm, and marking timestamps of when the guide words appear;

[0192] Calculating respectively a first similarity between the speech segment between each two adjacent guide words and a preset retelling template, and a second similarity between the speech segment and a preset answer template;

[0193] If the first similarity exceeds the second similarity, the speech segment between the two adjacent guide words is confirmed as a repetition segment; if the second similarity exceeds the first similarity, the speech segment between the two adjacent guide words is confirmed as an answer segment;

[0194] The retelling segments are integrated to obtain the retelling part, and the answer segments are integrated to obtain the answer part.

[0195] In an optional embodiment, the repetition feature includes a repetition expression feature and a repetition language feature, and the answer feature includes an answer expression feature and an answer language feature; the repetition expression feature includes an occurrence rate of abnormal repetition expression, and the answer expression feature includes an occurrence rate of abnormal answer expression; the repetition language feature includes an occurrence rate of abnormal repetition language, and the answer language feature includes an occurrence rate of abnormal answer language;

[0196] The performing feature recognition on the retelling part and the answer part to obtain retelling features and answer features includes:

[0197] Performing face detection on each video frame of the retelling part and the answering part, and extracting the coordinates of key points of the face in each video frame;

[0198] Determine whether abnormal expressions appear in each video frame based on the coordinates of facial key points in each video frame, and count the number of video frames with abnormal expressions;

[0199] Determining the occurrence rate of abnormal expression in the retelling part and the occurrence rate of abnormal expression in the answer part according to the number of video frames in which abnormal expression appears;

[0200] At the same time, performing speech flow analysis on the retelling part and the answering part respectively to determine the average retelling speed value and the average retelling volume value of the retelling part, and the average answering speed value and the average answering volume value of the answering part;

[0201] The occurrence rate of abnormal language in the repetition part is determined according to the average repetition speed value and the average repetition volume value; the occurrence rate of abnormal language in the answer part is determined according to the average answer speed value and the average answer volume value.

[0202] In an optional embodiment, the answer risk level includes an expression risk level and a language risk level; and determining the question-answer risk level of the claim case based on the repetition characteristics and the answer characteristics includes:

[0203] Configure corresponding differential intervals for different risk levels;

[0204] Calculating a first difference between the occurrence rate of the abnormal expression in the answer and the occurrence rate of the abnormal expression in the repetition; and determining the expression risk level according to the difference interval into which the first difference falls;

[0205] Calculating a second difference between the occurrence rate of the abnormal language in the answer and the occurrence rate of the abnormal language in the repetition; and determining the risk level of the question and answer based on the difference interval into which the second difference falls;

[0206] Determining the material risk level of the claim case based on the voice content and the claim materials includes:

[0207] Verifying the consistency of the voice content using entity alignment technology based on the claim materials to obtain a consistency verification result;

[0208] Verifying the rationality of the claim case through knowledge graph reasoning based on the claim materials to obtain a rationality verification result;

[0209] Calling third-party system data to cross-verify the correctness of the claim materials to obtain correctness verification results;

[0210] The material risk level is determined according to the consistency verification result, the rationality verification result, and the correctness verification result.

[0211] In an optional embodiment, determining the comprehensive risk level of the claim case based on the question-and-answer risk level and the material risk level includes:

[0212] Normalizing the expression risk level, the language risk level, and the material risk level, mapping them to the interval [0, 1], and obtaining standardized values ​​S1, S2, and S3 of each risk level, respectively;

[0213] The weight vector W = [w1, w2, w3] is determined by the hierarchical analysis method, where w1 is the expression risk level weight corresponding to the expression feature, w2 is the language risk level weight corresponding to the language feature, and w3 is the material risk level weight corresponding to the text feature.

[0214] Determine the comprehensive risk score S=∑(Si×wi) of the claim case according to the standardized value of each risk level and the weight vector, where i=1, 2, 3;

[0215] The comprehensive risk level is determined according to the comprehensive risk score.

[0216] In an optional embodiment, determining the target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal includes:

[0217] Define the comprehensive risk level range corresponding to different claims processing permissions. The higher the claims processing permission, the wider the corresponding comprehensive risk level range;

[0218] Filtering a set of business processing terminals with claims processing authority based on the range of the comprehensive risk level;

[0219] The business processing end with the least number of claims cases to be processed is selected from the business processing end set as the target business processing end.

[0220] In an optional embodiment, the claim payment to the user based on the comprehensive risk level, the claim application and the insurance contract includes:

[0221] Determine the claim amount of the claim case based on the claim application and the insurance contract;

[0222] Determining a target claims process for the claim case based on the comprehensive risk level of the claim case;

[0223] According to the target claim settlement process, a claim settlement is paid to the user in the claim settlement amount.

[0224] Example 3

[0225] Based on the same application concept, see Figure 10 As shown, Figure 10 FIG. 1 shows a schematic diagram of the structure of a computer device provided by the third embodiment of the present invention, wherein Figure 10 As shown, a computer device 1000 provided in the third embodiment of the present application includes:

[0226] Processor 1001, memory 1002 and 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 through the bus 1003, and the machine-readable instructions are executed by the processor 1001 to execute the steps of the Internet insurance claims business processing method shown in the above embodiment 1.

[0227] Example 4

[0228] Based on the same application concept, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of the Internet insurance claims processing method described in any one of the above embodiments are executed.

[0229] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems and devices can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0230] The computer program product for processing Internet insurance claims business provided by the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be found in the method embodiment and will not be repeated here.

[0231] The Internet insurance claims processing system provided by the embodiment of the present invention can be specific hardware on the device or software or firmware installed on the device. The system provided by the embodiment of the present invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference can be made to the corresponding content in the aforementioned method embodiment. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0232] In the embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some communication interface, device or unit, which may be electrical, mechanical or other forms.

[0233] The units described as separate components may or may not be physically separate, and 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 these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0234] In addition, each functional unit in the embodiment provided by the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0235] If the functions are implemented in the form of 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 the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0236] It should be noted that similar numbers and letters represent similar items in the following figures. 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 only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0237] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. However, such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for processing internet insurance claims, characterized in that: Applied to an internet insurance claims processing system, the system includes a user terminal, a server, and several business processing terminals, each of which has different claims processing permissions. The method includes: The server receives the claim materials uploaded by the user for the claim case through the user terminal; generates a number of claim case questions based on the claim materials and sends them to the user terminal; prompts the user to repeat each claim case question and answer each claim case question through the user terminal; When the user repeats each claim case question and answers each claim case question, the user terminal collects the user video and sends it to the server; The server performs text recognition on the user video to obtain voice content; divides the user video into a retelling part and an answer part according to the voice content; and performs feature recognition on the retelling part and the answer part to obtain a retelling feature and an answer feature; The server determines the question-answer risk level of the claim case based on the repetition feature and the answer feature; determines the material risk level of the claim case based on the voice content and the claim materials; and determines the comprehensive risk level of the claim case based on the question-answer risk level and the material risk level. The server determines a target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal; generates a claim application based on the claim materials and the voice content and sends it to the target business processing terminal; The target business processing end makes a claim payment to the user based on the comprehensive risk level, the claim application form and the insurance contract.

2. The method according to claim 1, characterized in that The dividing the user video into a retelling part and an answer part according to the voice content includes: identifying guide words in the voice content based on a keyword detection algorithm, and marking timestamps of when the guide words appear; Calculating respectively a first similarity between the speech segment between each two adjacent guide words and a preset retelling template, and a second similarity between the speech segment and a preset answer template; If the first similarity exceeds the second similarity, the speech segment between the two adjacent guide words is confirmed as a repetition segment; if the second similarity exceeds the first similarity, the speech segment between the two adjacent guide words is confirmed as an answer segment; The retelling segments are integrated to obtain the retelling part, and the answer segments are integrated to obtain the answer part.

3. The method according to claim 1, characterized in that The repetition features include repetition expression features and repetition language features, and the answer features include answer expression features and answer language features; the repetition expression features include the occurrence rate of abnormal repetition expression, and the answer expression features include the occurrence rate of abnormal answer expression; the repetition language features include the occurrence rate of abnormal repetition language, and the answer language features include the occurrence rate of abnormal answer language; The performing feature recognition on the retelling part and the answer part to obtain retelling features and answer features includes: Performing face detection on each video frame of the retelling part and the answering part, and extracting the coordinates of key points of the face in each video frame; Determine whether abnormal expressions appear in each video frame based on the coordinates of facial key points in each video frame, and count the number of video frames with abnormal expressions; Determining the occurrence rate of abnormal expression in the retelling part and the occurrence rate of abnormal expression in the answer part according to the number of video frames in which abnormal expression appears; At the same time, performing speech flow analysis on the retelling part and the answering part respectively to determine the average retelling speed value and the average retelling volume value of the retelling part, and the average answering speed value and the average answering volume value of the answering part; The occurrence rate of abnormal language in the repetition part is determined according to the average repetition speed value and the average repetition volume value; the occurrence rate of abnormal language in the answer part is determined according to the average answer speed value and the average answer volume value.

4. The method according to claim 3, characterized in that The answer risk level includes an expression risk level and a language risk level; the question-answer risk level of the claim case is determined based on the repetition characteristics and the answer characteristics, including: Configure corresponding differential intervals for different risk levels; Calculating a first difference between the occurrence rate of the abnormal expression in the answer and the occurrence rate of the abnormal expression in the repetition; and determining the expression risk level according to the difference interval into which the first difference falls; Calculating a second difference between the occurrence rate of the abnormal language in the answer and the occurrence rate of the abnormal language in the repetition; and determining the risk level of the question and answer based on the difference interval into which the second difference falls; Determining the material risk level of the claim case based on the voice content and the claim materials includes: Verifying the consistency of the voice content using entity alignment technology based on the claim materials to obtain a consistency verification result; Verifying the rationality of the claim case through knowledge graph reasoning based on the claim materials to obtain a rationality verification result; Calling third-party system data to cross-verify the correctness of the claim materials to obtain correctness verification results; The material risk level is determined according to the consistency verification result, the rationality verification result, and the correctness verification result.

5. The method according to claim 4, characterized in that The comprehensive risk level of the claim case is determined based on the question-and-answer risk level and the material risk level, including: Normalizing the expression risk level, the language risk level, and the material risk level, mapping them to the interval [0, 1], and obtaining standardized values ​​S1, S2, and S3 of each risk level, respectively; The weight vector W = [w1, w2, w3] is determined by the hierarchical analysis method, where w1 is the expression risk level weight corresponding to the expression feature, w2 is the language risk level weight corresponding to the language feature, and w3 is the material risk level weight corresponding to the text feature. The comprehensive risk score S of the claim case is determined based on the standardized value of each risk level and the weight vector. i ×w i ), i = 1, 2, 3; The comprehensive risk level is determined according to the comprehensive risk score.

6. The method according to claim 1, characterized in that Determining the target business processing terminal according to the comprehensive risk level and the claims processing authority of each business processing terminal includes: Define the comprehensive risk level range corresponding to different claims processing permissions. The higher the claims processing permission, the wider the corresponding comprehensive risk level range. Filtering a set of business processing terminals with claims processing authority based on the range of the comprehensive risk level; The business processing end with the least number of claims cases to be processed is selected from the business processing end set as the target business processing end.

7. The method according to claim 1, characterized in that The claim payment to the user based on the comprehensive risk level, the claim application and the insurance contract includes: Determine the claim amount of the claim case based on the claim application and the insurance contract; Determining a target claims process for the claim case based on the comprehensive risk level of the claim case; According to the target claim settlement process, a claim settlement is paid to the user in the claim settlement amount.

8. An Internet insurance claims processing system, characterized in that: The system includes a user terminal, a server, and several business processing terminals, each of which has different claims processing authority; The server is configured to receive, through the user terminal, claim materials uploaded by the user regarding a claim case; generate a number of claim case questions based on the claim materials and send them to the user terminal; and prompt the user, through the user terminal, to repeat and answer each claim case question; The user terminal is used to collect user videos and send them to the server when the user repeats and answers the claims case questions; The server is configured to perform text recognition on the user video to obtain voice content; Dividing the user video into a retelling part and an answer part according to the voice content; performing feature recognition on the retelling part and the answer part to obtain a retelling feature and an answer feature; The server is configured to determine a question-answer risk level of the claim case based on the repetition feature and the answer feature; and determine a material risk level of the claim case based on the voice content and the claim materials; Determine the comprehensive risk level of the claim case based on the question-and-answer risk level and the material risk level; The server is configured to determine a target business processing terminal based on the comprehensive risk level and the claims processing authority of each business processing terminal; Generate a claim application based on the claim materials and the voice content and send it to the target business processing terminal; The target business processing end is used to make claim payments to the user based on the comprehensive risk level, the claim application form and the insurance contract.

9. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the Internet insurance claims processing method as described in any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the Internet insurance claims processing method as described in any one of claims 1 to 7.

Images