Cognition-enhanced multi-modal insurance proxy investment method, device, equipment and medium

Abstract

The application relates to the technical field of intelligent decision-making, and discloses a multi-modal insurance proxy application method and device based on cognitive enhancement, equipment and a medium. The method comprises the following steps: determining an intended product according to an insurance application intention; generating a dynamic proxy application authorization strategy and then generating encrypted proxy application authorization information according to the insurance application intention; generating personalized explanation text and performing a proxy application operation on the intended product according to multi-modal input information and clause text information. In the foregoing manner, the application automatically identifies an insurance application intention and matches a product through a user portrait and historical behavior, combines dynamic proxy application authorization and multi-modal identity verification, automatically generates structured insurance application data by using multi-modal input, outputs personalized spoken language explanation based on the user portrait, reduces the understanding threshold, and realizes intelligent proxy application. The application can be applied to the business fields of financial technology, medical health and old-age care, so that an insurance application system expresses insurance terms in a spoken language, thereby improving the proxy application efficiency.

Description

Technical Field

[0001] This application relates to the field of intelligent decision-making technology, and in particular to a multimodal insurance investment method, device, equipment and medium based on cognitive enhancement. Background Technology

[0002] With the rapid development of artificial intelligence, big data, and cloud computing technologies, the financial and insurance industry is accelerating its digital and intelligent transformation. Currently, while mainstream insurance application systems have generally implemented online self-service application or agent-assisted data entry, in actual business operations, they still heavily rely on users manually filling out application forms item by item. This is time-consuming and lacks automated information collection and intelligent data entry capabilities, making it difficult to support the business needs of immediate application. In the fintech field, in scenarios such as intelligent investment advisory, credit risk control, and compliance report generation, although data analysis and rule engine technologies are widely used, existing insurance application systems still generally suffer from problems such as limited modal processing capabilities, insufficient semantic understanding depth, and a lack of cross-modal information fusion mechanisms when dealing with unstructured multi-source data such as voice, images, and scanned documents. This results in poor intelligent interaction experience and limited business processing efficiency.

[0003] In the fields of healthcare and elderly care, with the advancement of electronic health record systems, medical insurance information platforms, and smart elderly care service platforms, the demand for data interaction between medical institutions, elderly care institutions, and insurance companies is becoming increasingly frequent. However, existing technological solutions mostly focus on single-point applications, such as extracting structured information from medical insurance cards or physical examination reports using optical character recognition technology, or achieving simple consultation interactions through voice recognition technology. They have not yet formed a unified semantic understanding and structured output capability covering multimodal inputs of voice, text, and images. Due to the varying digital literacy levels of users, the density of professional medical and insurance terminology, and the mixed formats of multi-source health data, existing insurance underwriting systems cannot effectively lower the cognitive threshold, resulting in serious obstacles for elderly users in understanding medical terminology, leading to misunderstandings, hesitation, and even service interruptions.

[0004] Therefore, in the fields of fintech and healthcare and elderly care, how to improve the efficiency of insurance application by using colloquial language to describe insurance terms has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a multimodal insurance application method, device, equipment, and medium based on cognitive enhancement to improve the efficiency of insurance application.

[0006] Firstly, this application provides a multimodal insurance purchase method based on cognitive enhancement, the method comprising: In response to a user's insurance application request, the system identifies the insurance application intent based on user profile data and historical behavior data, and determines the intended product from a pre-set product library based on the insurance application intent. A dynamic insurance authorization strategy matching the intended product is generated based on the stated insurance intention; User identity authentication is performed based on the dynamic proxy delivery authorization strategy. The target user's biometric data, behavioral data, and user ID data are collected and multimodal identity verification is performed to generate encrypted proxy delivery authorization information. The system acquires the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product, generates structured insurance data based on the multimodal input information, and generates personalized explanatory text that matches the user profile data based on the terms and conditions text information. The insurance company will perform an insurance application on behalf of the intended product based on the encrypted application authorization information, the structured insurance data, and the personalized explanatory text.

[0007] Secondly, this application also provides a multimodal insurance payment device based on cognitive enhancement, the device comprising: The intended product determination module is used to respond to a user's insurance application request, identify the insurance application intention based on user profile data and historical behavior data, and determine the intended product from a preset product library according to the insurance application intention. The dynamic insurance purchase authorization strategy generation module is used to generate a dynamic insurance purchase authorization strategy that matches the intended product based on the insurance purchase intention. The encrypted delivery authorization information generation module is used to authenticate user identity based on the dynamic delivery authorization strategy, collect the target user's biometric data, behavioral data and user certificate data and perform multimodal identity verification to generate encrypted delivery authorization information; The personalized explanation text generation module is used to obtain the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product, generate structured insurance data based on the multimodal input information, and generate personalized explanation text that matches the user profile data based on the terms and conditions text information. The proxy payment module is used to perform proxy payment operations on the intended product based on the encrypted proxy payment authorization information, the structured insurance data, and the personalized explanation text.

[0008] Thirdly, this application also provides a computer device, the computer device including a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the cognitive enhancement-based multimodal insurance purchase method as described above.

[0009] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the cognitive enhancement-based multimodal insurance payment method described above.

[0010] This application discloses a multimodal insurance purchase method, apparatus, device, and medium based on cognitive enhancement. The method includes responding to a user's insurance purchase request, identifying the purchase intention based on user profile data and historical behavior data, and determining the intended product from a preset product library according to the purchase intention; generating a dynamic purchase authorization strategy matching the intended product according to the purchase intention; performing user identity authentication based on the dynamic purchase authorization strategy, collecting the target user's biometric data, behavioral data, and user identification data and performing multimodal identity verification to generate encrypted purchase authorization information; obtaining the target user's multimodal input information and the terms and conditions text information corresponding to the intended product, generating structured insurance data based on the multimodal input information, and generating personalized interpretation text matching the user profile data based on the terms and conditions text information; and performing a purchase operation on the intended product based on the encrypted purchase authorization information, the structured insurance data, and the personalized interpretation text. Through the above methods, this application automatically identifies insurance intent and matches products based on user profiles and historical behavior. Combined with dynamic authorization and multimodal identity verification, it automatically generates structured insurance data using multimodal input and outputs personalized verbal explanations based on user profiles, lowering the understanding threshold and achieving intelligent processing from authorization to application. In the fields of fintech, healthcare, and elderly care, this improves the efficiency of application processing by using conversational language to express insurance terminology. Attached Figure Description

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

[0012] Figure 1 This is a schematic diagram illustrating the application environment of a multimodal insurance purchase method based on cognitive enhancement provided by an embodiment of this application; Figure 2 This is a schematic flowchart of a multimodal insurance purchase method based on cognitive enhancement provided in the first embodiment of this application; Figure 3 This is a schematic flowchart of a multimodal insurance purchase method based on cognitive enhancement provided in the second embodiment of this application; Figure 4A schematic block diagram of a cognitively enhanced multimodal insurance payment device provided for embodiments of this application; Figure 5 A schematic block diagram of the structure of a computer device provided for an embodiment of this application. Detailed Implementation

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

[0014] The multimodal insurance purchase method based on cognitive enhancement provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server responds to a user's insurance application request by identifying the application intent based on user profile data and historical behavior data, determining the desired product from a pre-set product library according to the intent; generating a dynamic application authorization strategy matching the desired product based on the intent; authenticating the user's identity based on the dynamic application authorization strategy by collecting the target user's biometric data, behavioral data, and user identification data, performing multimodal identity verification, and generating encrypted application authorization information; acquiring the target user's multimodal input information and the terms and conditions text information corresponding to the desired product, generating structured insurance data based on the multimodal input information, and generating personalized interpretation text matching the user profile data based on the terms and conditions text information; and executing the application operation for the desired product based on the encrypted application authorization information, the structured insurance data, and the personalized interpretation text. Through the above methods, this application automatically identifies insurance intent and matches products based on user profiles and historical behavior. Combined with dynamic authorization and multimodal identity verification, it automatically generates structured insurance data using multimodal input and outputs personalized spoken explanations based on user profiles, lowering the understanding threshold and achieving intelligent processing from authorization to application. In the fields of fintech, healthcare, and elderly care, this improves the efficiency of insurance application by using colloquial language to express insurance terminology. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a dedicated server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0015] Please see Figure 2 , Figure 2This is a schematic flowchart illustrating a cognitive enhancement-based multimodal insurance application method provided in the first embodiment of this application. This cognitive enhancement-based multimodal insurance application method can be applied to insurance application systems to improve the efficiency of application processing.

[0016] like Figure 2 As shown, the cognitive enhancement-based multimodal insurance purchase method specifically includes steps S10 to S50.

[0017] Step S10: In response to the user's insurance application request, identify the insurance application intention based on user profile data and historical behavior data, and determine the intended product from the preset product library according to the insurance application intention; Specifically, after receiving an insurance application request initiated by a user through the client, the insurance application system retrieves the user's profile data and historical behavior data from the user's historical database and user profile database. User profile data may include the user's age, occupation, income range, health tags, and past insurance preferences. Historical behavior data may include browsing history, types of insurance products previously purchased, and interaction feedback. The insurance application system inputs this data into a pre-trained user intent encoder, which outputs the insurance intent category and its corresponding probability distribution through temporal modeling and feature fusion. Insurance intent categories include critical illness protection, medical expense reimbursement, and retirement savings, among others. Based on the insurance intent category, the insurance application system searches and ranks products in a pre-set product database according to their matching degree, identifying the candidate product with the highest matching degree as the intended product, and obtaining the product identifier and terms index of the intended product.

[0018] In some embodiments, this embodiment is illustrated using the fintech business sector as an example.

[0019] In fintech business scenarios such as robo-advisory, credit services, or insurance sales, users initiate insurance applications through bank applications, wealth management platforms, or smart terminals. The insurance application system first identifies the user's identity and collects static attribute features from the user profile data. In fintech scenarios, static attribute features include at least risk tolerance level, asset size range, credit score, years of investment experience, compliance qualification results, and protection needs preferences. At the same time, it extracts dynamic interaction features from the user's historical behavior data within a preset time window. Dynamic interaction features include at least historical insurance application records, financial transaction records, product browsing history, intelligent customer service consultation conversation content, and dwell time on product details pages. After fusing and encoding the static attribute features and dynamic interaction features, the system inputs them into a pre-trained intent classification model, outputs the intent probability distribution of each preset intent category, and takes the intent category with the highest probability as the insurance application intent.

[0020] The system retrieves a product index library corresponding to a pre-defined product database, which covers insurance products, wealth management products, and credit products. Each pre-defined product undergoes feature extraction to form a pre-defined product feature vector containing product type, risk level, coverage scope, return structure, underwriting conditions, and target customer group attributes. The underwriting intent is mapped to an underwriting intent feature vector, and the semantic correlation between this underwriting intent feature vector and the feature vectors of each pre-defined product is calculated. Finally, pre-defined products with a correlation greater than a pre-defined correlation threshold are identified as intended products. When multiple candidate products meet the threshold, they are sorted in descending order of correlation and compliance verification is performed in conjunction with suitability management rules in the fintech field to select intended products that both meet user intent and regulatory compliance requirements.

[0021] Step S20: Generate a dynamic insurance authorization strategy that matches the intended product based on the stated insurance intention; Specifically, based on the insurance application intention and the chosen product, the system retrieves the initial authorization scope corresponding to that product from a pre-defined permission template library. This initial authorization scope includes a whitelist of fields that can be filled in, the authorization validity period, and the single-operation permission level. Simultaneously, the insurance application system obtains the target user's historical authorization acceptance rate and risk preference tags. These risk preference tags are dynamically generated based on the user's past confirmation behavior regarding auto-filled categories.

[0022] The system invokes a large language model, using the initial authorization scope, historical authorization acceptance rate, and risk preference labels as input context. Through sample inference, it generates an adjusted set of authorization fields and differentiated timeliness configurations, ultimately outputting a dynamic proxy authorization strategy.

[0023] In some embodiments, this embodiment is described in conjunction with the medical, health and elderly care business field.

[0024] In the field of medical, health and elderly care business, when the intended product is elderly care insurance, commercial medical insurance or pension security plan, the product attribute information of the intended product is first analyzed, including product type, coverage, insurance conditions and list of required fields. At the same time, the application form is fully scanned and semantically recognized to extract a set of sensitive fields involving user health privacy, such as health declaration information, past medical history, abnormal physical examination indicators, nursing level assessment, medical insurance participation status, medication records and beneficiary information.

[0025] The insurance application system matches pre-configured risk level tags for intended products with corresponding preset permission templates. It combines the urgency of the application intention with the preset permission templates for analysis to dynamically determine the read and write permissions for each field. For core sensitive fields such as health declaration information, past medical history, and nursing level assessment, it sets read-only or prohibits proxy filling permissions. For basic information such as height, weight, and contact information, it sets permissions to allow direct proxy filling. For fields such as medical insurance status and physical examination reports, it sets permissions to allow proxy filling but require secondary identity verification. Based on the sensitive field set and field read and write permissions, it assembles strategies to generate a dynamic proxy application authorization strategy, which includes a list of fields that can be filled by proxy, authorization validity period, permission scope, and operational constraints.

[0026] Step S30: Based on the dynamic proxy delivery authorization strategy, perform user identity authentication, collect the target user's biometric data, behavioral data and user document data and perform multimodal identity verification to generate encrypted proxy delivery authorization information; Specifically, the insurance underwriting system initiates an identity authentication process based on a dynamic authorization strategy, collecting the target user's biometric data, behavioral data, and user identification data. For example, biometric data can be a real-time captured facial video stream, behavioral data can be a trajectory sequence of electronic signature handwriting (including pen pressure, speed, and spatial path), and user identification data can be images of ID cards and bank cards.

[0027] Liveness detection and facial feature extraction are performed on the face video stream to generate a facial feature vector; the electronic signature handwriting is dynamically time-normalized and compared with a pre-stored signature template to output a signature matching score; key fields such as name, ID number, and card number are extracted from ID card and bank card images through optical character recognition to generate document feature vectors.

[0028] The facial feature vector, signature matching score, and document feature vector are input into the multimodal joint verification model. The three types of features are fused and judged through a comparative learning framework. When the face liveness detection is passed, the signature matching score exceeds the preset threshold, and the format of the key fields of the document is verified to be correct, encrypted delivery authorization information is generated. The encrypted delivery authorization information generally carries the user identifier, authorization scope, and validity period.

[0029] Step S40: Obtain the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product; generate structured insurance data based on the multimodal input information; and generate personalized explanatory text that matches the user profile data based on the terms and conditions text information. Specifically, the system acquires multimodal information input by target users via voice, text, or images, and simultaneously retrieves the corresponding terms and conditions text information for the intended products from the product terms and conditions database. For multimodal input, voice information is converted into text using a speech recognition engine. The converted text, along with the directly input text, is then processed by a natural language understanding module to extract entities, attribute values, and semantic relationships. Image information is processed using optical character recognition and computer vision models to extract key fields.

[0030] The multimodal extraction results are input into an attention-based multimodal semantic fusion engine. Through field alignment, conflict resolution, and missing value inference, structured insurance data in JSON format is output. Furthermore, personalized explanatory text is generated based on the policy terms, constructing an insurance domain knowledge graph. This knowledge graph includes terminology entities and their definitional relationships, synonym relationships, and common user question pairs. Key terms and phrases are extracted from the policy terms. Using age, education level, and cognitive ability tags from the user profile data as constraints, matching colloquial explanation templates are retrieved from the knowledge graph. A large language model is then invoked, using key terms, phrases, and colloquial explanation templates as input, combined with user profile constraints, to rewrite, simplify, and emotionally refine the text, ultimately generating personalized explanatory text that matches the user profile data.

[0031] Step S50: Perform the insurance application on behalf of the intended product based on the encrypted insurance application authorization information, the structured insurance application data, and the personalized explanation text.

[0032] Specifically, the encrypted application authorization information is parsed to extract the authorization scope and validity period, and it is verified whether the current operation time is within the validity period and whether the current data entry operation falls within the authorized scope. Within the authorized scope, the insurance application system automatically maps and fills the structured insurance data into the various fields of the application form corresponding to the intended product. At the same time, personalized explanatory text is displayed in a pop-up window associated with the corresponding field or clause item, and is simultaneously read aloud via a speech synthesis module. Users can click to confirm or ask questions for each explanatory text.

[0033] Collect user confirmation instructions for each explanatory text. Once all required fields are filled and the user has completed all confirmation instructions, generate a proxy insurance request message. The proxy insurance request message may contain encrypted proxy insurance authorization information, the hash value of the filled insurance data, and the user confirmation timestamp.

[0034] This embodiment discloses a multimodal insurance purchase method based on cognitive enhancement. The method includes responding to a user's insurance purchase request, identifying the purchase intention based on user profile data and historical behavior data, determining the intended product from a preset product library according to the purchase intention; generating a dynamic purchase authorization strategy matching the intended product according to the purchase intention; performing user identity authentication based on the dynamic purchase authorization strategy, collecting the target user's biometric data, behavioral data, and user identification data and performing multimodal identity verification to generate encrypted purchase authorization information; obtaining the target user's multimodal input information and the terms and conditions text information corresponding to the intended product, generating structured insurance data based on the multimodal input information, and generating personalized interpretation text matching the user profile data based on the terms and conditions text information; and performing a purchase operation on the intended product based on the encrypted purchase authorization information, the structured insurance data, and the personalized interpretation text. Through the above methods, this application automatically identifies insurance intent and matches products based on user profiles and historical behavior. Combined with dynamic authorization and multimodal identity verification, it automatically generates structured insurance data using multimodal input and outputs personalized verbal explanations based on user profiles, lowering the understanding threshold and achieving intelligent processing from authorization to application. In the fields of fintech, healthcare, and elderly care, this improves the efficiency of application processing by using conversational language to express insurance terminology.

[0035] Please see Figure 3 , Figure 3 This is a schematic flowchart illustrating a cognitive enhancement-based multimodal insurance application method provided in the second embodiment of this application. This cognitive enhancement-based multimodal insurance application method can be applied to insurance application systems to improve the efficiency of application processing.

[0036] like Figure 3 As shown, step S10 includes steps S101 to S105.

[0037] Step S101: Extract static attribute features from the user profile data and dynamic interaction features from the historical behavior data; Specifically, static attribute features of target users are extracted from the user profile database. These features may include user age, gender, occupation category, income range, location, types of previously purchased insurance products, and health declaration tags. Simultaneously, dynamic interaction features are extracted from user behavior logs based on historical behavioral data. These features may include the user's product browsing sequence, click frequency, insurance policy reading time, time spent on each form page, and past records of insurance purchases, cancellations, or rejections. After numerically encoding both static attribute features and dynamic interaction features, they are concatenated to form a comprehensive user feature vector.

[0038] Step S102: Generate an intent probability distribution based on the static attribute features and dynamic interaction features using a pre-trained intent classification model, and take the intent category with the highest probability as the insurance intent; Specifically, the pre-trained intent classification model is invoked. The input layer receives the user's comprehensive feature vector, which is then mapped layer by layer through multiple fully connected layers and non-linear activation functions. A normalized exponential function outputs an intent probability distribution, covering various preset insurance intent categories, including critical illness protection, medical expense reimbursement, accidental injury protection, retirement savings, and long-term care protection. The insurance system selects the intent category with the highest probability value from the intent probability distribution as the current user's insurance intent.

[0039] Step S103: Obtain the product index library corresponding to the preset product library, wherein the product index library contains the preset product feature vector of each preset product; Specifically, the pre-set product library contains basic information about pre-set products, including product name, product type, summary of coverage, rate table, and terms and conditions index. A product index library corresponding to the pre-set product library is obtained. Each pre-set product corresponds to a pre-set product feature vector, which is generated by encoding multi-dimensional attributes such as product coverage keywords, applicable population tags, rate range, waiting period days, and deductible settings.

[0040] Step S104: Calculate the correlation between the insurance intention feature vector of the insurance intention and each of the preset product feature vectors; Specifically, the insurance intention is transformed into an insurance intention feature vector, the dimension of which is consistent with the dimension of each preset product feature vector in the product index. Based on the semantic attributes associated with the insurance intention, such as coverage type, target population, and compensation method, these attributes are mapped to the corresponding feature dimensions and assigned appropriate weights. Then, the correlation degree between this insurance intention feature vector and each preset product feature vector in the product index is calculated one by one. The correlation degree can be calculated using methods such as cosine similarity or inner product similarity.

[0041] Step S105: Determine the preset products whose relevance is greater than the preset relevance threshold as the intended products.

[0042] Specifically, a relevance threshold is preset. This threshold can be derived from historical successful matching data or manually configured by operations personnel. Each relevance value is compared to the preset threshold, and all preset products with a relevance greater than the threshold are selected. If the filter results contain multiple products, they can be sorted from highest to lowest relevance, and one or more products with the highest relevance are identified as the intended products. If the filter results are empty, the relevance threshold is lowered or a fuzzy matching strategy is used to re-search until at least one intended product is identified.

[0043] based on Figure 2 In the illustrated embodiment, step S20 includes: The risk level label of the intended product is analyzed to identify the set of sensitive fields contained in the intended product, wherein the set of sensitive fields includes health disclosure information, beneficiary information and payment account information; Specifically, the system reads the pre-configured risk level tags from the metadata of the intended product, including three levels: low risk, medium risk, and high risk. These levels are set by the insurance institution based on the product type, complexity of coverage, probability of claims, and regulatory requirements. The insurance application system parses the application form structure of the intended product to identify the set of sensitive fields it contains. The set of sensitive fields includes at least health disclosure information (such as past medical history, family history of hereditary diseases, and current medication), beneficiary information (such as beneficiary's name, ID number, and relationship to the insured), and payment account information (such as bank card number and bank name).

[0044] A preset permission template is determined based on the risk level label, and field read / write permissions are determined according to the insurance intention and the preset permission template. Specifically, the system matches the corresponding preset permission template from a pre-defined permission template library. These templates predefine the range of operable fields and authorization constraints for different risk levels. For example, for low-risk products, the preset permission template allows the filling in of most information fields except for the payment account; for high-risk products, the preset permission template restricts the filling in of key medical history items in the health declaration and requires user confirmation for each item.

[0045] The dynamic proxy delivery authorization strategy is generated based on the set of sensitive fields and the read / write permissions of the fields.

[0046] Specifically, the sensitive field set is integrated with field read / write permissions to generate a structured, dynamic authorization policy for insurance applications. Each sensitive field in the authorization policy corresponds to a permission description object, which includes whether the field allows for proxy filling, whether real-time user confirmation is required, whether additional authentication is needed, and the maximum number of operations allowed within the authorization period. The insurance application system also automatically sets the validity period of the entire authorization policy based on risk level labels; for example, the authorization validity period for high-risk products does not exceed 24 hours, while that for low-risk products can be extended to seven calendar days.

[0047] based on Figure 2 In the illustrated embodiment, step S30 includes: In response to the triggering command of the dynamic proxy authorization strategy, the current facial image of the target user is collected as the biometric data, the electronic signature trajectory data of the target user is collected as the behavioral feature data, and the ID image of the target user is collected as the user ID data. Specifically, after the dynamic proxy authorization strategy is confirmed and triggered by the user, the current facial image of the target user is collected through the camera of the mobile terminal device as biometric data. The user faces the camera in natural light and completes live actions such as blinking and shaking their head. At the same time, the electronic signature trajectory data of the user's handwritten input is collected on the touch screen as behavioral feature data, recording the coordinates, timestamp, pen pressure value and movement speed of each sampling point during the signing process. The camera or album function is called to collect the target user's ID card image and bank card image as user ID data, ensuring that the text is clear and unobstructed.

[0048] The facial key point feature vectors in the biometric data are extracted using a preset convolutional neural network. Specifically, the current face image is input into a pre-trained convolutional neural network, which extracts discriminative local features from the face image, performs alignment and normalization processing on the face image, and generates a fixed-dimensional facial key point feature vector in the output layer.

[0049] The electronic signature feature vector is extracted from the behavioral feature data by a long short-term memory network, wherein the electronic signature feature vector includes a pen pressure feature vector, a rate feature vector, and an acceleration feature vector; Specifically, the system reads the collected electronic signature trajectory data (i.e., a series of sampling points sorted by time) frame by frame through a long short-term memory network. The system extracts the original information such as the coordinates, pen pressure, and time difference of each sampling point and extracts three independent feature vectors: the pen pressure feature vector, which reflects the pressure change pattern when the user signs; the rate feature vector, which reflects the speed and rhythm of the signature; and the acceleration feature vector, which reflects the smoothness and abruptness of the pen tip movement during the signing process.

[0050] The structured field feature vectors in the user's identification data are extracted using an optical character recognition engine. Specifically, the captured document images are input into an optical character recognition (OCR) engine, which converts the characters in each text region into editable strings. For ID cards, the engine automatically extracts key fields such as name, gender, ethnicity, date of birth, ID number, and address; for bank card images, the engine extracts fields such as card number, expiration date, and issuing bank name. After recognition, the discrete string fields are structured according to predefined field names to generate structured field feature vectors.

[0051] The facial key point feature vector, the electronic signature feature vector, and the structured field feature vector are aligned and weighted and fused to perform a joint confidence assessment; Specifically, the facial key point feature vector, electronic signature feature vector, and structured field feature vector are jointly evaluated for confidence. The three feature vectors are dimensionally aligned to unify them into the same embedding space. Then, each feature vector is assigned a different weight according to a preset weighting coefficient. A multimodal fusion algorithm is used to weight and fuse these feature vectors to calculate a comprehensive confidence score. At the same time, the independent confidence of each modality is evaluated separately.

[0052] When the overall confidence level after fusion exceeds the preset high threshold, and the confidence levels of each independent modality are not lower than their corresponding low thresholds, the joint confidence level assessment is deemed to have passed; if the confidence level of any modality is too low, the data for the corresponding modality must be recollected.

[0053] When the facial key point feature vector, the electronic signature feature vector, and the structured field feature vector pass the joint confidence assessment, the encrypted proxy delivery authorization information is generated, wherein the encrypted proxy delivery authorization information includes user identifier, authorization scope identifier, authorization validity period, and operation log identifier.

[0054] Specifically, after the joint confidence assessment is passed, encrypted proxy application authorization information is generated. The insurance application system extracts the user identifier from the authenticated user's identity information. The user identifier is the user's unique account or ID number in the insurance application system. It extracts the authorization scope identifier from the dynamic proxy application authorization strategy. The authorization scope identifier is a predefined set of allowed fields to be filled in and the operation permission level. Simultaneously, it records the current timestamp plus a preset validity period as the authorization validity period. For example, the authorization validity period can be set to 24 hours or valid for a single session. The insurance application system assigns a globally unique operation log identifier to this authorization. The above four pieces of information are packaged according to a preset format, and the packaged information is digitally signed and encrypted to generate encrypted proxy application authorization information.

[0055] based on Figure 2 In the illustrated embodiment, step S40 includes: The voice input information is converted into speech recognition text information, the speech recognition text information and the text input information are semantically parsed to identify insurance-related entities and insurance semantic relationships, and key image fields in the image input information are extracted through a visual detection model. Specifically, the system acquires the user's voice input stream (i.e., voice input information) and feeds it into a speech recognition engine. Based on acoustic and language models, the continuous audio signal is decoded into a text sequence corresponding to the voice stream, generating speech-recognized text information. This speech-recognized text information is then merged with the user's directly input text information and fed into a natural language understanding module. Using techniques such as word segmentation, part-of-speech tagging, named entity recognition, and dependency parsing, the module extracts insurance-related entities (e.g., the insured's age, height, weight, social security status, and medical history) and the semantic relationships between these entities (e.g., "suffers from," "insured," and "excluded" relationships). Simultaneously, the system feeds user-uploaded image input information (e.g., medical examination reports and professional photos) into a visual detection model. Combining object detection and optical character recognition technologies, the model locates and extracts key text fields from the image input information, such as the name of the medical examination item, the numerical value of the examination result, the ID card number, and the insurance status, generating key image fields.

[0056] The insurance-related entities, the insurance semantic relationships, and the image key fields are encoded to generate insurance entity feature vectors, semantic relationship feature vectors, and image keyword body feature vectors, respectively. Specifically, the extracted insurance-related entities are encoded and converted, and each entity is mapped to a fixed-length numerical vector according to a preset entity encoding dictionary to generate an insurance entity feature vector; the extracted insurance semantic relationships are encoded, and each relationship type is represented as a semantic relationship feature vector; for key fields of the image, a combination of visual feature extraction and text feature fusion is used to encode the text content of each key field, as well as its position in the image, confidence level, and other information, into an image key field feature vector.

[0057] Map the insured entity feature vector, the semantic relationship feature vector, and the image keyword body feature vector to the same semantic vector space; Specifically, a target semantic vector space is established, with dimensions consistent with the three types of feature vectors mentioned above. Through linear or nonlinear projection transformation matrices, the insured entity feature vector, semantic relation feature vector, and image key field feature vector are projected and mapped, transforming these three types of vectors, originally distributed in different feature spaces, into the same semantic vector space. During the projection process, a pre-trained alignment loss function is used to ensure that semantically similar entities, relations, and image fields have similar vector representations in this space. For example, the entity "weight 75kg" and the "75kg" field vector identified in the image are close in distance in this space.

[0058] In the same semantic vector space, the insured entity feature vector, the semantic relationship feature vector, and the image keyword body feature vector are cross-modal correlated to generate a structured feature vector, and the structured feature vector is mapped to the structured insurance data.

[0059] Specifically, within the same semantic vector space, cross-modal association calculations are performed on the feature vectors of insured entities, semantic relationships, and key image fields. An attention mechanism or graph neural network is used to calculate the association weights between each entity and relevant image fields, as well as the enhancing effect of semantic relationships between entities on image fields. This fuses the three types of vectors into a comprehensive, multimodal, structured feature vector. In this structured feature vector, each dimension corresponds to a standardized insured field value (such as the insured's age, height, weight, social security status, or presence of a pre-existing condition). The structured feature vector is input into a field mapping network, which decodes each dimension into its corresponding specific field name and value, generating structured insurance data according to a predetermined format.

[0060] based on Figure 2 In the illustrated embodiment, step S40 includes: Extract education level tags, terminology dwell time information, and repetitive question frequency from the user profile data to construct a user cognitive profile; Specifically, various attributes of the user are read from the user profile database, and education level tags (such as doctoral, master's, bachelor's, high school and below) are extracted. Simultaneously, two dynamic indicators are collected from the user's behavior logs during historical insurance application interactions: first, the duration of time spent on terminology, i.e., the time the user spends reading or listening to explanations of insurance terms on each term page or audio segment; second, the frequency of repeated questions, i.e., the number of times the user raises secondary or multiple questions regarding the same technical term or clause content. After normalizing these static and dynamic features, they are integrated to construct a multi-dimensional user cognitive profile.

[0061] Based on the user cognitive profile, the target users are divided into different levels of comprehension. Specifically, based on the generated user cognitive profile, target users are divided into different comprehension levels using preset grading rules or lightweight classifiers. For example, users with high education levels, short dwell time on terms, and extremely low frequency of repeated questions are categorized as "high comprehension level"; users with moderate education levels, moderate dwell time on terms, and occasional questions are categorized as "moderate comprehension level"; and users with low education levels, long dwell time on terms, or high frequency of repeated questions are categorized as "low comprehension level." Furthermore, more granular levels can be set, such as "elderly comprehension level" or "digital disadvantaged group level," with each level corresponding to different explanatory text generation strategies and parameter configurations.

[0062] The large language model is invoked to perform dimensionality reduction transformation on the technical terms in the clause text information, and the personalized explanatory text is generated by determining the differentiated vocabulary and differentiated sentence structure according to the comprehension level.

[0063] Specifically, the system reads the terms and conditions of the intended product, extracts the technical terms (such as waiting period, deductible, pre-existing conditions, guaranteed renewal, etc.), and uses a large language model to perform dimensionality reduction transformation, breaking down highly generalized and abstract professional expressions into easy-to-understand basic vocabulary and short sentences. During the transformation process, based on the target user's comprehension level, it dynamically selects and matches a differentiated vocabulary and sentence structure: for high comprehension levels, some technical terms are retained but accompanied by brief parenthetical explanations; for medium comprehension levels, common everyday analogies are used to replace technical terms; for low comprehension levels, extremely short sentences and high-frequency daily expressions are used, with examples and repetitive emphasis structures actively added. The large language model then generates personalized explanatory text for each item according to the selected vocabulary and sentence structure.

[0064] Based on any of the above embodiments, in this embodiment, step S50 includes: Decrypt the encrypted delivery authorization information to determine the scope and validity period of the authorization; Specifically, upon receiving the encrypted proxy submission authorization information from the front-end or middleware, the system uses a pre-set symmetric key or asymmetric private key to decrypt the information. Successful decryption yields the plaintext authorization data structure. From this structure, the authorization scope field is parsed, which may contain a list of product identifiers allowed to be filled in or a whitelist of fields that can be filled in on behalf of others. Simultaneously, the validity period field is parsed, containing the start and end timestamps of the authorization.

[0065] Obtain the current operation time, verify whether the current operation time is within the validity period, and verify whether the intended product is within the scope of authorization; Specifically, the system calls the operating system's time interface to obtain the current precise time, which is then used as the current operation time. Two independent checks are then performed: the first check compares the current operation time with the start and end times of the validity period parsed from the authorization information to determine if the current operation time falls within this closed interval; the second check compares whether the product identifier of the intended product to be operated on exists in the list of allowed product identifiers within the authorization scope, or whether the field to be filled is in the whitelist of fields within the authorization scope. Both checks are executed independently, and their results are recorded.

[0066] If the current operation time is within the valid period and the intended product is within the authorized scope, a verification pass identifier is generated; Specifically, a verification pass flag is generated only if the current operation time is within the authorized validity period and the intended product is within the authorized scope. The verification pass flag can be a token containing a timestamp and an authorized session ID. If either verification fails, no verification pass flag is generated; instead, an error code is returned and subsequent delivery operations are terminated.

[0067] Based on the verification pass identifier, the structured insurance data and the personalized explanation text are associated to generate insurance data to be purchased, and the application interface corresponding to the intended product is called. Specifically, after obtaining the verification pass indicator, the structured insurance application data is associated with the corresponding personalized explanatory text. Each field to be filled in the structured insurance application data is associated with its corresponding explanatory text fragment, forming a field-explanation pairing list. Simultaneously, the overall summary of the entire structured insurance application data and the collection of all explanatory texts are packaged into a single data object to be applied for. Based on the pre-registered interface configuration information for the intended product, the corresponding application processing interface for that product is dynamically invoked. This application processing interface is typically exposed in the form of an application programming interface (API).

[0068] The insurance application data and the intended product are encapsulated through the application agent interface to perform the application agent operation on the intended product.

[0069] Specifically, the data object to be insured and the unique identifier of the intended product are encapsulated, and a request message is constructed according to the request format required by the insurance agency interface. Encrypted agency authorization information (or a verification pass identifier) ​​is attached as identity credentials during the encapsulation process. After encapsulation, a request is sent to the insurance agency interface via a Hypertext Transfer Protocol client. Upon receiving the request, the insurance agency interface parses the data to be insured and writes each field into the insurance application form of the core insurance system. The interface returns a successful execution status code and an application number. Once the insurance application system receives the successful response, the agency operation for the intended product has been successfully executed.

[0070] Please see Figure 4 , Figure 4 This application provides a schematic block diagram of a cognitively enhanced multimodal insurance payment proxy device, which is used to execute the aforementioned cognitively enhanced multimodal insurance payment proxy method. The cognitively enhanced multimodal insurance payment proxy device can be configured on a server.

[0071] like Figure 4 As shown, the cognitive enhancement-based multimodal insurance payment device 400 includes: The intended product determination module 410 is used to respond to a user's insurance application request, identify the insurance application intention based on user profile data and historical behavior data, and determine the intended product from a preset product library according to the insurance application intention. The dynamic insurance purchase authorization strategy generation module 420 is used to generate a dynamic insurance purchase authorization strategy that matches the intended product based on the insurance purchase intention. The encrypted delivery authorization information generation module 430 is used to perform user identity authentication based on the dynamic delivery authorization strategy, collect the target user's biometric data, behavioral data and user certificate data and perform multimodal identity verification to generate encrypted delivery authorization information; The personalized explanation text generation module 440 is used to obtain the multimodal input information of the target user and the terms text information corresponding to the intended product, generate structured insurance data based on the multimodal input information, and generate personalized explanation text that matches the user profile data based on the terms text information. The proxy payment module 450 is used to perform proxy payment operations on the intended product based on the encrypted proxy payment authorization information, the structured insurance data, and the personalized explanation text.

[0072] Furthermore, the intended product determination module 410 includes: The feature extraction unit is used to extract static attribute features from the user profile data and dynamic interaction features from the historical behavior data. The insurance intent determination unit is used to generate an intent probability distribution based on the static attribute features and dynamic interaction features using a pre-trained intent classification model, and to take the intent category with the highest probability as the insurance intent. The product index library acquisition unit is used to acquire the product index library corresponding to the preset product library, wherein the product index library contains the preset product feature vector of each preset product. The correlation calculation unit is used to calculate the correlation between the insurance intention feature vector of the insurance intention and each of the preset product feature vectors; The intended product determination unit is used to determine the preset products whose relevance is greater than a preset relevance threshold as the intended products.

[0073] Furthermore, the dynamic proxy authorization strategy generation module 420 includes: A sensitive field set identification unit is used to parse the risk level label of the intended product and identify the sensitive field set contained in the intended product, wherein the sensitive field set includes health disclosure information, beneficiary information and payment account information; The field read / write permission determination unit is used to determine a preset permission template based on the risk level label, and to determine field read / write permissions according to the insurance intention and the preset permission template. The dynamic proxy authorization strategy generation unit is used to generate the dynamic proxy authorization strategy based on the set of sensitive fields and the read / write permissions of the fields.

[0074] Furthermore, the encrypted proxy authorization information generation module 430 includes: A multimodal data acquisition unit is used to, in response to the triggering command of the dynamic proxy authorization strategy, acquire the current face image of the target user as the biometric data, acquire the electronic signature trajectory data of the target user as the behavioral feature data, and acquire the ID image of the target user as the user ID data; A facial key point feature vector extraction unit is used to extract facial key point feature vectors from the biometric data through a preset convolutional neural network; The electronic signature feature vector extraction unit is used to extract the electronic signature feature vector from the behavioral feature data through a long short-term memory network, wherein the electronic signature feature vector includes a pen pressure feature vector, a rate feature vector, and an acceleration feature vector; The structured field feature vector extraction unit is used to extract the structured field feature vectors from the user's document data through the optical character recognition engine; The joint confidence assessment unit is used to align and weightedly fuse the facial key point feature vector, the electronic signature feature vector, and the structured field feature vector to perform a joint confidence assessment. The encrypted proxy authorization information generation unit is used to generate the encrypted proxy authorization information when the facial key point feature vector, the electronic signature feature vector, and the structured field feature vector pass the joint confidence assessment. The encrypted proxy authorization information includes a user identifier, an authorization scope identifier, an authorization validity period, and an operation log identifier.

[0075] Furthermore, the personalized explanation text generation module 440 includes: The key information recognition unit is used to convert the voice input information into voice recognition text information, perform semantic parsing on the voice recognition text information and the text input information, identify insurance-related entities and insurance semantic relationships, and extract key image fields from the image input information through a visual detection model. The encoding unit is used to encode the insurance-related entities, the insurance semantic relationships, and the image key fields, respectively generating insurance entity feature vectors, semantic relationship feature vectors, and image keyword body feature vectors; A vector mapping unit is used to map the insured entity feature vector, the semantic relationship feature vector, and the image keyword body feature vector to the same semantic vector space; The structured insurance data generation unit is used to perform cross-modal association of the insurance entity feature vector, the semantic relationship feature vector, and the image keyword body feature vector in the same semantic vector space to generate a structured feature vector, and to map the structured feature vector to the structured insurance data.

[0076] Furthermore, the personalized explanation text generation module 440 includes: The user cognitive profile building unit is used to extract education level tags, term dwell time information and repetitive question frequency from the user profile data to build a user cognitive profile. The comprehension level determination unit is used to divide the target user into different comprehension levels based on the user cognitive profile. The personalized explanatory text generation unit is used to call a large language model to perform dimensionality reduction transformation on the technical terms in the clause text information, and to generate the personalized explanatory text based on the comprehension level by determining a differentiated vocabulary and differentiated sentence structure.

[0077] Furthermore, the delivery operation module 450 includes: An encrypted delivery authorization information decryption unit is used to decrypt the encrypted delivery authorization information and determine the authorization scope and validity period. The verification unit is used to obtain the current operation time, verify whether the current operation time is within the validity period, and verify whether the intended product is within the scope of authorization. The verification pass identifier generation unit is used to generate a verification pass identifier when the current operation time is within the valid period and the intended product is within the authorized scope; The proxy application interface calling unit is used to associate the structured insurance application data and the personalized explanation text based on the verification pass identifier to generate insurance application data to be applied for, and to call the proxy application interface corresponding to the intended product. The proxy payment operation unit is used to encapsulate the insurance data to be purchased and the intended product through the proxy payment interface to perform the proxy payment operation on the intended product.

[0078] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and modules can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0079] The aforementioned device can be implemented as a computer program, which can be used in, for example... Figure 5 It runs on the computer device shown.

[0080] Please see Figure 5 , Figure 5 This is a schematic block diagram illustrating the structure of a computer device according to an embodiment of this application. The computer device may be a server.

[0081] See Figure 5 The computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0082] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any cognitively enhanced multimodal insurance delivery method.

[0083] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0084] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any cognitive enhancement-based multimodal insurance substitution method.

[0085] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0086] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0087] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: In response to a user's insurance application request, the system identifies the insurance application intent based on user profile data and historical behavior data, and determines the intended product from a pre-set product library based on the insurance application intent. A dynamic insurance authorization strategy matching the intended product is generated based on the stated insurance intention; User identity authentication is performed based on the dynamic proxy delivery authorization strategy. The target user's biometric data, behavioral data, and user ID data are collected and multimodal identity verification is performed to generate encrypted proxy delivery authorization information. The system acquires the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product, generates structured insurance data based on the multimodal input information, and generates personalized explanatory text that matches the user profile data based on the terms and conditions text information. The insurance company will perform an insurance application on behalf of the intended product based on the encrypted application authorization information, the structured insurance data, and the personalized explanatory text.

[0088] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement any of the cognitive enhancement-based multimodal insurance purchase methods provided in the embodiments of this application.

[0089] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0090] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multimodal insurance purchase method based on cognitive enhancement, characterized in that, include: In response to a user's insurance application request, the system identifies the insurance application intent based on user profile data and historical behavior data, and determines the intended product from a pre-set product library based on the insurance application intent. A dynamic insurance authorization strategy matching the intended product is generated based on the stated insurance intention; User identity authentication is performed based on the dynamic proxy delivery authorization strategy. The target user's biometric data, behavioral data, and user ID data are collected and multimodal identity verification is performed to generate encrypted proxy delivery authorization information. The system acquires the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product, generates structured insurance data based on the multimodal input information, and generates personalized explanatory text that matches the user profile data based on the terms and conditions text information. The insurance company will perform an insurance application on behalf of the intended product based on the encrypted application authorization information, the structured insurance data, and the personalized explanatory text.

2. The multimodal insurance purchase method based on cognitive enhancement according to claim 1, characterized in that, The step of responding to a user's insurance application request by identifying the insurance intent based on user profile data and historical behavior data, and determining the intended product from a pre-set product library according to the insurance intent, includes: Extract static attribute features from the user profile data and dynamic interaction features from the historical behavior data; The pre-trained intent classification model generates an intent probability distribution based on the static attribute features and dynamic interaction features, and takes the intent category with the highest probability as the insurance intent. Obtain the product index library corresponding to the preset product library, wherein the product index library contains the preset product feature vector of each preset product; Calculate the correlation between the insurance intention feature vector and each of the preset product feature vectors; The preset products whose relevance is greater than a preset relevance threshold are identified as the intended products.

3. The multimodal insurance purchase method based on cognitive enhancement according to claim 1, characterized in that, The process of generating a dynamic insurance authorization strategy that matches the intended product based on the insurance intention includes: The risk level label of the intended product is analyzed to identify the set of sensitive fields contained in the intended product, wherein the set of sensitive fields includes health disclosure information, beneficiary information and payment account information; A preset permission template is determined based on the risk level label, and field read / write permissions are determined according to the insurance intention and the preset permission template. The dynamic proxy delivery authorization strategy is generated based on the set of sensitive fields and the read / write permissions of the fields.

4. The multimodal insurance purchase method based on cognitive enhancement according to claim 1, characterized in that, The process of user authentication based on the dynamic proxy delivery authorization strategy involves collecting the target user's biometric data, behavioral data, and user identification data, performing multimodal identity verification, and generating encrypted proxy delivery authorization information, including: In response to the triggering command of the dynamic proxy authorization strategy, the current facial image of the target user is collected as the biometric data, the electronic signature trajectory data of the target user is collected as the behavioral feature data, and the ID image of the target user is collected as the user ID data. The facial key point feature vectors in the biometric data are extracted using a preset convolutional neural network. The electronic signature feature vector is extracted from the behavioral feature data by a long short-term memory network, wherein the electronic signature feature vector includes a pen pressure feature vector, a rate feature vector, and an acceleration feature vector; The structured field feature vectors in the user's identification data are extracted using an optical character recognition engine. The facial key point feature vector, the electronic signature feature vector, and the structured field feature vector are aligned and weighted and fused to perform a joint confidence assessment; When the facial key point feature vector, the electronic signature feature vector, and the structured field feature vector pass the joint confidence assessment, the encrypted proxy delivery authorization information is generated, wherein the encrypted proxy delivery authorization information includes user identifier, authorization scope identifier, authorization validity period, and operation log identifier.

5. The multimodal insurance purchase method based on cognitive enhancement according to claim 1, characterized in that, The multimodal input information includes voice input information, text input information, and image input information. The process of acquiring the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product, and generating structured insurance data based on the multimodal input information, includes: The voice input information is converted into speech recognition text information, the speech recognition text information and the text input information are semantically parsed to identify insurance-related entities and insurance semantic relationships, and key image fields in the image input information are extracted through a visual detection model. The insurance-related entities, the insurance semantic relationships, and the image key fields are encoded to generate insurance entity feature vectors, semantic relationship feature vectors, and image keyword body feature vectors, respectively. Map the insured entity feature vector, the semantic relationship feature vector, and the image keyword body feature vector to the same semantic vector space; In the same semantic vector space, the insured entity feature vector, the semantic relationship feature vector, and the image keyword body feature vector are cross-modal correlated to generate a structured feature vector, and the structured feature vector is mapped to the structured insurance data.

6. The multimodal insurance purchase method based on cognitive enhancement according to claim 5, characterized in that, The step of generating personalized explanatory text that matches the user profile data based on the terms and conditions text information includes: Extract education level tags, terminology dwell time information, and repetitive question frequency from the user profile data to construct a user cognitive profile; Based on the user cognitive profile, the target users are divided into different levels of comprehension. The large language model is invoked to perform dimensionality reduction transformation on the technical terms in the clause text information, and the personalized explanatory text is generated by determining the differentiated vocabulary and differentiated sentence structure according to the comprehension level.

7. The multimodal insurance purchase method based on cognitive enhancement according to any one of claims 1 to 6, characterized in that, The step of performing the application on behalf of the intended product based on the encrypted application authorization information, the structured insurance data, and the personalized explanatory text includes: Decrypt the encrypted delivery authorization information to determine the scope and validity period of the authorization; Obtain the current operation time, verify whether the current operation time is within the validity period, and verify whether the intended product is within the scope of authorization; If the current operation time is within the valid period and the intended product is within the authorized scope, a verification pass identifier is generated; Based on the verification pass identifier, the structured insurance data and the personalized explanation text are associated to generate insurance data to be purchased, and the application interface corresponding to the intended product is called. The insurance application data and the intended product are encapsulated through the application agent interface to perform the application agent operation on the intended product.

8. A multimodal insurance payment device based on cognitive enhancement, characterized in that, include: The intended product determination module is used to respond to a user's insurance application request, identify the insurance application intention based on user profile data and historical behavior data, and determine the intended product from a preset product library according to the insurance application intention. The dynamic insurance purchase authorization strategy generation module is used to generate a dynamic insurance purchase authorization strategy that matches the intended product based on the insurance purchase intention. The encrypted delivery authorization information generation module is used to authenticate user identity based on the dynamic delivery authorization strategy, collect the target user's biometric data, behavioral data and user certificate data and perform multimodal identity verification to generate encrypted delivery authorization information; The personalized explanation text generation module is used to obtain the multimodal input information of the target user and the terms and conditions text information corresponding to the intended product, generate structured insurance data based on the multimodal input information, and generate personalized explanation text that matches the user profile data based on the terms and conditions text information. The proxy payment module is used to perform proxy payment operations on the intended product based on the encrypted proxy payment authorization information, the structured insurance data, and the personalized explanation text.

9. A computer device, characterized in that, The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the cognitive enhancement-based multimodal insurance payment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the cognitive enhancement-based multimodal insurance payment method as described in any one of claims 1 to 7.

Images