Insurance information extraction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides an insured information extraction method and device, electronic equipment and a storage medium, belongs to the technical field of information extraction, and is suitable for the field of financial science and technology. The method comprises the steps of performing semantic recognition based on insured consultation sub-data of a target object to obtain semantic recognition data; generating consultation prompt data based on the semantic recognition data and a target insurance data model matched with the target insurance type; performing intention recognition on the consultation prompt data and the semantic recognition data based on an intention recognition sub-model of a preset verbal skill generation model to obtain an insured intention label of the target object; based on a preset verbal skill generation model, a verbal skill generation sub-model performs verbal skill generation on the insured intention label and the semantic recognition data to obtain a consultation response verbal skill; obtaining object response data of the target object; and performing insured information extraction on the object response data to obtain object insured key information. According to the embodiment of the invention, the accuracy of insured information extraction can be improved.

Description

Technical Field

[0001] The present application relates to the field of information extraction technology, is applicable to the field of financial technology, and in particular to a method and device for extracting insurance information, an electronic device, and a storage medium. Background Art

[0002] Insurance information extraction refers to the process of extracting insurance information under this type of insurance by using the descriptive information entered by the user in the insurance promotion scenario. Currently, the descriptive information entered by the user can be identified by intelligent robots, and since most intelligent robots use command-based settings, it is necessary to exhaustively enumerate the possible commands that the user may enter, or use natural language processing (NLP) to analyze the correlation between the descriptive information entered by the customer and the command, and then use a decision tree to decide what action to perform.

[0003] However, the development dimension of intelligent robots with command-based settings is highly complex and it is difficult to meet the complex and changeable scenarios in actual insurance promotion, which affects the accuracy of insurance information extraction. In addition, the user's description method varies greatly, and the semantic analysis method used by related technologies is prone to errors, which affects the accuracy of insurance information extraction. Therefore, how to improve the accuracy of insurance information extraction has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method and device for extracting insurance information, an electronic device, and a storage medium, aiming to improve the accuracy of extracting insurance information.

[0005] To achieve the above purpose, a first aspect of an embodiment of the present application proposes a method for extracting insurance information, the method comprising:

[0006] Obtaining insurance consultation data of the target object, wherein the insurance consultation data includes insurance consultation sub-data, and the format types of any two insurance consultation sub-data are different;

[0007] Perform semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data of the target insurance type;

[0008] Acquire a target insurance data model matching the target insurance type;

[0009] Generate prompt data based on the semantic recognition data and the target insurance data model to obtain consulting prompt data;

[0010] The intention recognition sub-model based on the preset speech generation model performs intention recognition on the consultation prompt data and the semantic recognition data to obtain the insurance participation intention label of the target object;

[0011] Based on the speech generation sub-model of the preset speech generation model, speech generation is performed on the insurance intention label and the semantic recognition data to obtain the consultation response speech;

[0012] Acquiring object response data fed back by the target object based on the consultation response script;

[0013] The insurance information of the object response data is extracted to obtain the key insurance information of the object, and the key insurance information of the object matches the target insurance type.

[0014] In some embodiments, the intent recognition sub-model includes an encoder, a context extraction unit, a position encoder, a self-attention unit, and a decoder;

[0015] The intention recognition sub-model based on the preset speech generation model performs intention recognition on the consultation prompt data and the semantic recognition data to obtain the insurance participation intention label of the target object, including:

[0016] Based on the encoder, the consultation prompt data is encoded to obtain a consultation prompt sequence;

[0017] Performing data encoding on the semantic recognition data based on the encoder to obtain a semantic recognition sequence;

[0018] Performing sequence fusion on the consulting prompt sequence and the semantic recognition sequence to obtain an initial fused sequence;

[0019] Performing feature extraction on the initial fusion sequence based on the context extraction unit to obtain an initial fusion feature;

[0020] Performing position encoding on the initial fusion sequence based on the position encoder to obtain a position encoding feature;

[0021] Based on the self-attention unit, the initial fusion feature and the position encoding feature are subjected to self-attention processing to obtain a target fusion feature;

[0022] The target fusion feature is decoded based on the decoder to obtain the insurance intention label.

[0023] In some embodiments, the speech generation sub-model based on the preset speech generation model generates speech for the insurance intention label and the semantic recognition data to obtain the consultation response speech, including:

[0024] Selecting the insurance participation rule data of the target insurance type from a preset insurance knowledge graph based on the insurance participation intention label, wherein the insurance participation rule data includes insurance participation rule sub-data;

[0025] Acquiring initial recognition sub-data from the semantic recognition data;

[0026] Performing data matching on the initial identification sub-data and the insurance participation rule sub-data to obtain a data matching result;

[0027] Determining the unuploaded sub-data from the insurance rule sub-data based on the data matching result;

[0028] Based on the speech generation sub-model, speech generation is performed on the initial recognition sub-data and the non-uploaded sub-data to obtain the consultation response speech.

[0029] In some embodiments, the generating speech for the initial recognition sub-data and the non-uploaded sub-data based on the speech generation sub-model to obtain the consultation response speech includes:

[0030] If the unuploaded sub-data is empty, generate a speech for the initial identification sub-data based on the speech generation sub-model to obtain the consultation response speech, and the consultation response speech is used to guide the target object to confirm the information of the initial identification sub-data;

[0031] If the non-uploaded sub-data is not empty, speech generation is performed for the non-uploaded sub-data based on the speech generation sub-model to obtain the consultation response speech, and the consultation response speech is used to guide the target object to supplement information of the non-uploaded sub-data.

[0032] In some embodiments, the performing semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data having a target insurance type includes:

[0033] Get the data format type of the insurance consultation sub-data;

[0034] Determine the target semantic detection model of the insurance consultation sub-data based on the data format type;

[0035] Performing semantic detection on the insurance consultation sub-data based on the target semantic detection model to obtain a detection text;

[0036] Performing text enhancement on the detected text to obtain enhanced text;

[0037] Perform consultation intention identification on the insurance consultation sub-data to obtain a consultation intention label and the target insurance type;

[0038] Generate a hypothetical document based on the consultation intention label and the enhanced text to obtain a target hypothetical document;

[0039] The semantic recognition data is obtained by performing data integration based on the enhanced text, the consulting intention label, the target insurance type and the target hypothetical document.

[0040] In some embodiments, the method further comprises:

[0041] Acquire an underwriting rule decision tree of the target insurance type, wherein the underwriting rule decision tree includes an underwriting rule node and a node information type of the underwriting rule node;

[0042] Perform node information matching on the key information of the insurance subject and the underwriting rule node based on the node information type, and determine a node label of the underwriting rule node, wherein the node label includes a node matching label;

[0043] Generate an insurance application form for the target object based on the node label and the object insurance key information.

[0044] In some embodiments, generating the insurance application form for the target object based on the node label and the object insurance key information includes:

[0045] Selecting valid extracted information that matches the underwriting rule node from the subject's insurance key information;

[0046] Classifying the effectively extracted information based on a preset information classification model to obtain an information category;

[0047] Performing information verification on the information category and the valid extracted information based on a preset information verification model to obtain a verification result;

[0048] If the verification result is that the verification is passed and the node labels of the underwriting rule nodes are all the node matching labels, the insurance application form for the target object is generated based on the key insurance information of the object.

[0049] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a device for extracting insurance information, the device comprising:

[0050] The first acquisition module is used to acquire the insurance consultation data of the target object, wherein the insurance consultation data includes insurance consultation sub-data, and the format types of any two insurance consultation sub-data are different;

[0051] A semantic recognition module, used for performing semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data of the target insurance type;

[0052] A second acquisition module, used to acquire a target insurance data model matching the target insurance type;

[0053] A generating module, used for generating prompt data based on the semantic recognition data and the target insurance data model to obtain consulting prompt data;

[0054] An intention recognition module, used to perform intention recognition on the consultation prompt data and the semantic recognition data based on the intention recognition sub-model of the preset speech generation model, and obtain the insurance intention label of the target object;

[0055] A speech generation module, used for generating speech for the insurance intention label and the semantic recognition data based on the speech generation sub-model of the preset speech generation model to obtain the consultation response speech;

[0056] A third acquisition module is used to acquire object response data fed back by the target object based on the consultation response script;

[0057] The extraction module is used to extract the insurance information of the object response data to obtain the key insurance information of the object, and the key insurance information of the object matches the target insurance type.

[0058] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0059] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0060] The present application proposes a method and device for extracting insurance information, an electronic device, and a storage medium. The method obtains the insurance consultation data of the target object, wherein the insurance consultation data includes insurance consultation sub-data, and the format types of any two insurance consultation sub-data are different; further, semantic recognition is performed based on the insurance consultation sub-data to obtain semantic recognition data having the target insurance type; a target insurance data model matching the target insurance type is obtained; prompt data is generated based on the semantic recognition data and the target insurance data model to obtain consultation prompt data; further, the intention recognition sub-model of the preset speech generation model is used to perform intent recognition on the consultation prompt data and the semantic recognition data to obtain the insurance intention label of the target object; the speech generation sub-model of the preset speech generation model is used to generate speech for the insurance intention label and the semantic recognition data to obtain the consultation response speech; further, the object response data fed back by the target object based on the consultation response speech is obtained; the insurance information is extracted from the object response data to obtain the object's insurance key information, and the object's insurance key information matches the target insurance type. Compared with the related technologies that need to exhaustively enumerate the possible instructions that users may input, and it is difficult to meet the complex and changeable scenarios in actual insurance promotion, this application can perform semantic recognition on the insurance consultation sub-data with different format types, and generate more accurate consultation prompt data according to the matching target insurance data model, so that more accurate consultation response words can be generated according to the consultation prompt data, which is convenient for information extraction in complex and changeable scenarios. Therefore, the embodiments of this application can improve the accuracy of insurance information extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is the first flow chart of the insurance information extraction method provided in the embodiment of the present application;

[0062] Figure 2 yes Figure 1 A flowchart of step S120 in FIG.

[0063] Figure 3 This is a schematic diagram of a framework flow of the method for extracting insurance information provided in an embodiment of the present application;

[0064] Figure 4 yes Figure 1 A flowchart of step S150 in FIG.

[0065] Figure 5 yes Figure 1 A flowchart of step S160 in FIG.

[0066] Figure 6 This is the second flow chart of the insurance information extraction method provided in the embodiment of the present application;

[0067] Figure 7 yes Figure 6A flowchart of step S630 in FIG.

[0068] Figure 8 This is a specific flow chart of generating an insurance application provided in an embodiment of the present application;

[0069] Fig. 9 It is a structural schematic diagram of the insurance information extraction device provided in the embodiment of the present application;

[0070] Fig.10 It is a hardware structure diagram of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0072] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0073] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0074] First, some nouns involved in this application are analyzed:

[0075] Artificial Intelligence (AI): It is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science. AI attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0076] Natural Language Processing (NLP): NLP uses computers to process, understand, and apply human languages ​​(such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is commonly used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis, and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0077] Intent recognition refers to the process of understanding the intentions expressed by users through natural language (such as text, voice, etc.) and classifying them into predefined intent categories. In the insurance consulting scenario of FinTech, by performing intent recognition on the conversation between users and intelligent robots, a response script that better matches the user's intention can be output, thereby responding to the user's intention based on the response script.

[0078] Knowledge Graph: An information system for storing and representing knowledge that can organize entities and their relationships in the form of a graph. Knowledge graphs are usually composed of nodes (representing entities, such as people, places, things, etc.) and edges (representing relationships between entities). This structure enables knowledge to be represented and queried in an intuitive way.

[0079] Insurance information extraction refers to the extraction of insurance information under this type by determining the type of insurance through the descriptive information entered by the user in the insurance promotion scenario. In the traditional insurance business development process, the collection and recording of customer information usually rely on manual operations, which are inefficient and prone to errors. In order to improve work efficiency and service quality, a more intelligent solution is needed. At present, the descriptive information entered by the user can be recognized by intelligent robots, and since intelligent robots mostly use command-based settings, it is necessary to exhaustively enumerate the instructions that the user may enter, or use natural language processing to analyze the correlation between the descriptive information entered by the customer and the instructions, and then use a decision tree to decide what action to perform.

[0080] However, the development dimension of intelligent robots with command-based settings is highly complex and it is difficult to meet the complex and changeable scenarios in actual insurance promotion, which affects the accuracy of insurance information extraction. In addition, the user's description method varies greatly, and the semantic analysis method used by related technologies is prone to errors, which affects the accuracy of insurance information extraction. Therefore, how to improve the accuracy of insurance information extraction has become a technical problem that needs to be solved urgently.

[0081] Based on this, the embodiments of the present application provide a method and device for extracting insurance information, an electronic device, and a storage medium, aiming to improve the accuracy of extracting insurance information.

[0082] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0083] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0084] The insurance information extraction method provided in the embodiment of the present application relates to the field of artificial intelligence technology. The insurance information extraction method provided in the embodiment of the present application can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or it can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other basic cloud computing services; the software can be an application that implements the insurance information extraction method, etc., but is not limited to the above forms.

[0085] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0086] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on the subject's insurance data, the subject's voice data, and the subject's feedback data and other data related to the subject's identity or characteristics, the subject's permission or consent will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present application needs to obtain the sensitive personal information of the subject, the subject's separate permission or separate consent will be obtained through a pop-up window or jump to a confirmation page. After clearly obtaining the subject's separate permission or separate consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0087] See also Figure 1 , Figure 1 is an optional flow chart of the method for extracting insurance information provided in the embodiments of the present application. In some embodiments of the present application, Figure 1 The method may specifically include but is not limited to steps S110 to S180.

[0088] Step S110, obtaining insurance consultation data of the target object;

[0089] Step S120, performing semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data of the target insurance type;

[0090] Step S130, obtaining a target insurance data model matching the target insurance type;

[0091] Step S140, generating prompt data based on the semantic recognition data and the target insurance data model to obtain consulting prompt data;

[0092] Step S150, performing intent recognition on the consultation prompt data and the semantic recognition data based on the intent recognition sub-model of the preset speech generation model to obtain the insurance participation intention label of the target object;

[0093] Step S160, generating speech for the insurance intention label and semantic recognition data based on the speech generation sub-model of the preset speech generation model to obtain the consultation response speech;

[0094] Step S170, obtaining the object response data fed back by the target object based on the consultation response script;

[0095] Step S180, extracting the insurance information of the object's response data to obtain the key insurance information of the object.

[0096] In step S110 of some embodiments, the insurance consultation data refers to the information of the target object in the insurance system for insurance consultation. For example, in the insurance field of financial technology, the target object is a user who consults insurance products (such as social insurance, medical insurance, etc.), then the target object can exchange information with the intelligent robot in the insurance system, and the information replied by the target object is equivalent to the insurance consultation data. At this time, the insurance consultation data can include the target object's questions, needs, preferences, etc.

[0097] It should be noted that the insurance consultation data includes multiple insurance consultation sub-data, and each insurance consultation sub-data can represent a part or an aspect of the consultation data. For example, the insurance consultation sub-data can be a query about the type of insurance, premium calculation, claim process, insurance information, etc., without specific limitation. In addition, the insurance consultation sub-data of this application can contain different format types, that is, the format types of any two insurance consultation sub-data are different to meet the diversity of data. For example, the format type of the insurance consultation sub-data can be text format (such as questions sent by users through chat or email), voice format (such as voice records left by users through telephone consultation), image format (such as pictures of insurance documents uploaded by users), PDF or Word documents (such as insurance-related documents that users may upload), video format (such as questions that users may leave through video consultation), etc., without specific limitation.

[0098] It should be noted that due to the diversity of insurance consultation sub-data, this application can use different technologies and methods to parse and understand each format of data for subsequent semantic recognition and data processing, and the specific processing process will be described in detail later, so it will not be repeated here. In this way, the flexibility and adaptability of the system can be improved, enabling it to handle various forms of user input.

[0099] In step S120 of some embodiments, further, the present application may analyze the collected insurance consultation sub-data, identify and understand the intentions and needs of the target object, and identify semantic recognition data with the target insurance type, so as to analyze whether the target object is consulting about the relevant information of the target insurance type and whether it has the intention to take out insurance. The target insurance type is used to indicate different types of pre-set insurance, such as health insurance, car insurance, pension insurance, etc. By identifying and understanding the semantic recognition data of the target object with the target insurance type, the target object's consultation can be matched with a specific insurance product or service to provide more accurate and relevant information.

[0100] It should be noted that when performing semantic recognition based on the insurance consultation sub-data, the insurance consultation sub-data can be pre-processed such as cleaning and formatting to improve data quality. Furthermore, key features such as keywords, phrases, entities, etc. can be extracted from the pre-processed data, and the extracted features can be analyzed using machine learning models (such as classifiers) to identify the consulting intentions of the target objects. Furthermore, entity information such as specific insurance types, terms, conditions, etc. mentioned in the insurance consultation sub-data can be identified, and the consulting content of the target object can be understood through the results of comprehensive intent recognition and entity recognition, and matched with the target insurance type to generate semantic recognition data containing the target insurance type, providing a basis for subsequent data processing and response generation.

[0101] See also Figure 2 , Figure 2 It is a specific flow chart of step S120 provided in an embodiment of the present application. In some embodiments of the present application, step S120 may specifically include but is not limited to steps S210 to S270.

[0102] Step S210, obtaining the data format type of the insurance consultation sub-data;

[0103] Step S220, determining a target semantic detection model for the insurance consultation sub-data based on the data format type;

[0104] Step S230, performing semantic detection on the insurance consultation sub-data based on the target semantic detection model to obtain a detection text;

[0105] Step S240, performing text enhancement on the detected text to obtain enhanced text;

[0106] Step S250, identifying the consultation intention of the insurance consultation sub-data to obtain the consultation intention label and the target insurance type;

[0107] Step S260, generating a hypothetical document based on the consultation intention label and the enhanced text to obtain a target hypothetical document;

[0108] Step S270, performing data integration based on the enhanced text, the consultation intention label, the target insurance type and the target hypothetical document to obtain semantic recognition data.

[0109] In step S210 of some embodiments, since the insurance consultation sub-data may include multiple format types such as text, voice, image, etc., the data format type of the insurance consultation sub-data can be obtained first to clarify the specific type of each insurance consultation sub-data for subsequent processing.

[0110] In step S220 of some embodiments, after identifying the data format type of the insurance consultation sub-data, the present application selects a suitable target semantic detection model according to each data format type. Each target semantic detection model is equivalent to a model trained based on historical insurance data of different data format types, that is, different data formats can use different processing methods and models to extract semantic information, thereby improving the accuracy of semantic recognition of data, and further improving the accuracy of extracting insurance information.

[0111] Please note that Figure 3 , Figure 3 It is a framework flow chart of the insurance information extraction method provided in the embodiment of the present application. Among them, the insurance consultation sub-data of the present application may include text data input by text, image data input by pictures and other multimodal input data, that is, multiple insurance consultation sub-data with different data format types are obtained. Furthermore, a target semantic detection model based on the NLP model structure can be used to match text data, a target semantic detection model based on the optical character recognition (Optical Character Recognition, OCR) model structure can be used to match image data, and a target semantic detection model based on the computer vision (Computer Vision, CV) model structure can be used to match multimodal input data, and these models can also be flexibly replaced according to actual needs without limitation. Furthermore, semantic detection of the insurance consultation sub-data can be performed based on the corresponding target semantic detection model.

[0112] It should be noted that the target semantic detection model constructed based on OCR in this application can learn open source frameworks such as PaddleOCR and EasyOCR, develop a self-developed insurance OCR model, and use a large number of object cases for training and verification, which can accurately extract the target object’s valid information and insurance plan information. The target semantic detection model constructed based on CV in this application can be based on the SimpleCV machine vision free framework, access high-performance computer vision libraries such as openCV, and build a free insurance document and information recognition model. The target semantic detection model constructed based on NLP in this application can use open source library packages and related algorithms to build an NLP module by itself, assist in splitting the semantic components and parts of speech of the sentence input by the target object, and assist in judging the target object’s intention and emotion.

[0113] In step S230 of some embodiments, the insurance consultation sub-data may be further analyzed based on the determined target semantic detection model to extract and understand its semantic content, and finally generate a detection text. Among them, the semantic detection of the insurance consultation sub-data based on the target semantic detection model in this application may include semantic understanding, synonym rewriting, text hypothesis, etc., to obtain the detection text, without limitation.

[0114] In step S240 of some embodiments, after obtaining the detected text, the present application may perform text enhancement on the detected text to improve the quality and usability of the text to obtain enhanced text. Text enhancement may include removing noise, filling in missing information, correcting errors, etc. to obtain more accurate and complete enhanced text.

[0115] In step S250 of some embodiments, further, the present application can identify the consultation intention of the target object by analyzing the enhanced text, and classify and mark these intentions as consultation intention labels. At the same time, the specific insurance type consulted by the user is identified to prepare for the subsequent provision of accurate insurance information. Among them, by identifying the consultation intention of the insurance consultation sub-data, the specific content of the target object's consultation can be understood, including consultation questions, needs, concerns, etc. The result of the consultation intention identification is to assign one or more labels to the consultation of the target object, which represent the consultation intention of the target object. For example, if the target object asks about the coverage of medical insurance, the system may assign an intention label of "medical insurance coverage" to this consultation. In this way, in order to enable the system to respond to the consultation of the target object more effectively, by understanding the intention and needs of the target object, the system can provide more targeted answers and suggestions. For example, if the system identifies that a user is interested in health insurance, detailed information about health insurance can be provided, including the characteristics, terms and precautions of different products. Such personalized services can improve user satisfaction and increase conversion rates.

[0116] In step S260 of some embodiments, the present application may further utilize the consultation intent tags and enhanced text to generate a hypothetical document that simulates possible answers or related information to assist in subsequent data processing and answer generation.

[0117] In step S270 of some embodiments, the present application may further integrate the enhanced text, the consultation intent label, the target insurance type and the hypothetical document to form a comprehensive semantic recognition data, which will be used for subsequent decision support and response generation.

[0118] In step S130 of some embodiments, the present application may also simultaneously obtain a target insurance data model that matches the target insurance type. This target insurance data model is trained based on insurance product data related to the target insurance type and may include detailed information, terms, conditions, etc. of the insurance product for subsequent data processing and response generation.

[0119] In step S140 of some embodiments, further, as Figure 3 As shown, after the semantic recognition data and the target insurance data model are determined, prompt data can be generated based on the semantic recognition data and the target insurance data model to obtain consulting prompt data, that is, to obtain personalized customized prompt data. In this way, the generated consulting prompt data can include key information of insurance products, questions that the target object may need to know, etc., to guide the target object to further consult.

[0120] In step S150 of some embodiments, the present application may further use the intention recognition sub-model in the preset speech generation model to analyze the consultation prompt data and semantic recognition data to determine the insurance intention label of the target object. This insurance intention label can help the system understand the user's specific needs and purposes.

[0121] It should be noted that the preset speech generation model of this application can be based on the model structure of artificial intelligence generated content (AIGC), and use the quantized low-rank adapter (QLoRA) method to fine-tune the open source general large language model (such as the Qwen72B model, etc.) to form a special large model of this application in the insurance field. This model can understand insurance expertise and accurately extract information related to the insurance model. Among them, Figure 3As shown, this application can first use insurance professional knowledge training (such as insurance terms, insurance knowledge, classic cases, etc.) to extract professional data, mark proper nouns, etc., and vectorize the above data, and optimize the AIGC model in combination with QLoRA and other large language models to obtain a preset speech generation model. At the same time, in practice, the model algorithm can also be optimized to record the process of intelligent customer service as a historical record of the target object, and use the historical record as input data for the optimization algorithm to optimize the service of intelligent customer service.

[0122] It should be noted that the intent recognition sub-model includes an encoder, a context extraction unit, a position encoder, a self-attention unit and a decoder.

[0123] See also Figure 4 , Figure 4 It is a specific flow chart of step S150 provided in an embodiment of the present application. In some embodiments of the present application, step S150 may specifically include but is not limited to steps S410 to S470.

[0124] Step S410, encoding the consultation prompt data based on the encoder to obtain a consultation prompt sequence;

[0125] Step S420, encoding the semantic recognition data based on the encoder to obtain a semantic recognition sequence;

[0126] Step S430, performing sequence fusion on the consultation prompt sequence and the semantic recognition sequence to obtain an initial fused sequence;

[0127] Step S440, extracting features from the initial fusion sequence based on the context extraction unit to obtain initial fusion features;

[0128] Step S450, position encoding the initial fusion sequence based on the position encoder to obtain a position encoding feature;

[0129] Step S460, performing self-attention processing on the initial fusion feature and the position encoding feature based on the self-attention unit to obtain the target fusion feature;

[0130] Step S470: decode the target fusion features based on the decoder to obtain the insurance intention label.

[0131] In step S410 and step S420 of some embodiments, when the present application performs intent recognition on the consultation prompt data and the semantic recognition data, the consultation prompt data and the semantic recognition data may be respectively encoded based on the encoder to obtain the consultation prompt sequence and the semantic recognition sequence. The encoder may be a neural network model, such as a recurrent neural network (RNN), a long short-term memory network (LSTM) or a transformer, etc., to convert the consultation prompt data and the semantic recognition data into a series of numerical representations, namely, the consultation prompt sequence and the semantic recognition sequence.

[0132] In step S430 of some embodiments, the consulting prompt sequence and the semantic recognition sequence may further be subjected to sequence fusion, such as sequence merging or sequence weighted merging, to create an initial fused sequence containing information of both.

[0133] In step S440 of some embodiments, further, a context extraction unit (such as a special neural network layer) can be used to extract key features from the initial fusion sequence, and these features can represent the context information in the sequence to obtain initial fusion features.

[0134] In step S450 of some embodiments, further, the position encoder can be used to add position information to each element in the sequence, which helps the model understand the order of elements in the sequence. In this way, the obtained position encoding feature is a feature with added position information.

[0135] In step S460 of some embodiments, further, the initial fusion feature and the position encoding feature are self-attention processed based on the self-attention unit to obtain the target fusion feature. The self-attention mechanism (such as the self-attention in the Transformer) is used to process the initial fusion feature and the position encoding feature to enhance the model's understanding of the importance of different parts in the sequence. In addition, the target fusion feature obtained after processing contains the context and position information of the sequence.

[0136] In step S470 of some embodiments, a decoder may be used to decode the target fusion feature to generate a final output, namely, a label of the insurance intention, which represents the recognition result of the model on the user's consultation intention.

[0137] In the above embodiments, the present application can construct an intent recognition sub-model based on the encoding-fusion-decoding process. In this way, the model can understand and process complex language information and generate accurate intent recognition results.

[0138] In step S160 of some embodiments, further, the present application can generate speech for the insurance intention label and semantic recognition data based on the speech generation sub-model of the preset speech generation model to obtain the consultation response speech. In other words, through the above continuously refined intention recognition and semantic recognition, the present application can deeply understand the conversation habits of the target object, thereby helping the system to generate business development speech that better simulates human conversation. And in the process of conversation, the needs of the target object are excavated, so that the required insurance products can be accurately recommended to the target object. For example, if the target object consults about a certain insurance product, the generated consultation response speech can also guide the target object to upload and complete the information of the relevant insurance product, thereby improving the efficiency of insurance information collection.

[0139] See also Figure 5 , Figure 5 It is a specific flow chart of step S160 provided in an embodiment of the present application. In some embodiments of the present application, step S160 may specifically include but is not limited to steps S510 to S550.

[0140] Step S510, selecting insurance rule data of the target insurance type from the preset insurance knowledge graph based on the insurance intention label;

[0141] Step S520, obtaining initial recognition sub-data from the semantic recognition data;

[0142] Step S530, performing data matching on the initial identification sub-data and the insurance rule sub-data to obtain a data matching result;

[0143] Step S540, determining the unuploaded sub-data from the insurance rule sub-data based on the data matching result;

[0144] Step S550, based on the speech generation sub-model, speech generation is performed on the initial recognition sub-data and the unuploaded sub-data to obtain the consultation response speech.

[0145] In step S510 of some embodiments, when generating the consultation response script, the present application may first retrieve the insurance rule data matching the user's consultation intention in the preset insurance knowledge graph according to the previously identified insurance intention label. These insurance rule data contain specific insurance rules and conditions, and the insurance rule data contain insurance rule sub-data, each of which may contain different rules or clauses.

[0146] In step S520 of some embodiments, the present application may further obtain initial recognition sub-data from the semantic recognition data. These initial recognition sub-data are the system's preliminary understanding of the consulting content of the target object, and may include key information such as the target object's problems and needs.

[0147] In step S530 of some embodiments, the present application may further perform data matching on the initial identification sub-data and the insurance rule sub-data to determine which rules are directly related to the target object's consultation and obtain data matching results. The data matching results can help the system understand the needs of the target object more accurately and provide a basis for generating consultation response scripts.

[0148] In step S540 of some embodiments, the present application can further identify which insurance rule sub-data have not been uploaded or mentioned by the target object based on the data matching results, and these unuploaded sub-data may contain information related to the target insurance type that the target object needs to know.

[0149] In step S550 of some embodiments, further, based on a speech generation sub-model (such as a machine learning model specifically used to generate natural language responses), the initial recognition sub-data and the unuploaded sub-data can be used to generate speech to obtain consultation response speech. In addition, these consultation response speech are intended to answer the consultation questions of the target object and provide additional information that may be required for the consultation response speech.

[0150] In the above embodiment, the present application generates personalized insurance consultation responses by combining knowledge graphs and natural language processing technology. In this way, the system can provide more accurate and comprehensive answers, thereby improving user satisfaction and consultation efficiency.

[0151] It should be noted that this application can analyze the conversation context of the target object through the intention recognition sub-model, explore the insurance demands of the target object, and use the consultation response speech generated by the speech generation sub-model to guide the target object to enter relevant insurance information. Furthermore, according to the requirements of the insurance terms, the target object's valid information and protection information can be extracted, and relevant speech can be generated for the target object to confirm or for the target object to modify or supplement the information.

[0152] It should be noted that the present application generates speech for the initial identification sub-data and the non-uploaded sub-data based on the speech generation sub-model to obtain the consultation response speech, which may specifically include the following situations:

[0153] If the uploaded sub-data is empty, the initial identification sub-data is used to generate a speech based on the speech generation sub-model to obtain the consultation response speech, which is used to guide the target object to confirm the information of the initial identification sub-data;

[0154] If the non-uploaded sub-data is not empty, speech generation is performed for the non-uploaded sub-data based on the speech generation sub-model to obtain the consultation response speech, which is used to guide the target object to supplement the information of the non-uploaded sub-data.

[0155] It should be noted that if the sub-data that has not been uploaded is empty, this means that the target object has not provided additional information or data, and the system cannot identify the content that needs to be supplemented. At this time, the initial identification sub-data can be used to generate speech based on the speech generation sub-model, that is, a preliminary understanding is made based on the initial identification sub-data of the target object, and consultation response speech is generated to guide the target object to confirm the information of the initial identification sub-data. If the sub-data that has not been uploaded is not empty, this means that the system has identified that the target object has missed some information or data in the consultation. At this time, the speech can be generated for the sub-data that has not been uploaded based on the speech generation sub-model to guide the target object to supplement the information of the sub-data that has not been uploaded, that is, to help the target object supplement the information missed in the consultation, so that the system can better understand the needs of the target object and provide accurate services.

[0156] In the above embodiment, the present application can adjust the response strategy according to the information status provided by the user (whether there is any sub-data that has not been uploaded) to ensure that the target object can confirm or supplement the necessary information. In other words, the present application can mine the insurance demands of the target object, guide the target object to enter the insurance information, and extract the valid information and protection information of the target object according to the insurance terms.

[0157] In step S170 of some embodiments, after the system provides the consultation response script, the target object can provide feedback based on the received consultation response script to obtain the object response data fed back by the target object. The object response data at this time can be questions replied by the target object, supplementary picture materials, document materials, etc., without limitation.

[0158] In step S180 of some embodiments, the present application may further analyze all the subject response data of the target subject and extract the key insurance information of the subject that matches the target insurance type. This information may include the personal information of the target subject, the selected insurance product, the insurance period, the subject information required for the insurance product, etc., which is used to complete the insurance consultation and purchase process.

[0159] See also Figure 6 , Figure 6 It is another optional flow chart of the method for extracting insurance information provided by the embodiment of the present application. In some embodiments of the present application, after step S180, the method for extracting insurance information provided by the present application may further include but not limited to steps S610 to S630.

[0160] Step S610, obtaining an underwriting rule decision tree for the target insurance type;

[0161] Step S620, matching the object insurance key information with the underwriting rule node based on the node information type, and determining the node label of the underwriting rule node;

[0162] Step S630: Generate an insurance application form for the target object based on the node label and the key insurance information of the object.

[0163] In step S610 of some embodiments, after the key information of all insured objects is extracted, the application can also intelligently initiate insurance, information review, document quality inspection, etc., and finally complete the insurance and send the insurance policy to the target object. Specifically, the underwriting rule decision tree of the target insurance type can be obtained first. The underwriting rule decision tree is a decision support tool for automating the underwriting process, consisting of multiple nodes, each node represents a decision point, that is, the underwriting rule decision tree includes underwriting rule nodes and node information types of underwriting rule nodes, and the node information type indicates the specific information or conditions required for each underwriting rule node.

[0164] In step S620 of some embodiments, the present application can further match the object's key information and the underwriting rule node based on the node information type to determine the node label of the underwriting rule node. Through this matching, the system can determine the node label of each underwriting rule node to determine whether the underwriting rule node matches and obtains the corresponding object information, that is, whether the information acquisition is complete, and how to make the next decision based on the information provided by the target object.

[0165] It should be noted that the node label includes a node matching label and a node non-matching label. The node matching label is used to indicate that the corresponding underwriting rule node has obtained the corresponding object information. The node non-matching label is used to indicate that the corresponding underwriting rule node has not obtained the corresponding object information.

[0166] In step S630 of some embodiments, further, an insurance application form for the target object can be generated based on the node label and the key information of the object. That is, in this last step, the system can generate an insurance application form using the determined node label and the key information of the object. This insurance application form is automatically generated according to the underwriting rule decision tree and can contain all the necessary information to complete the user's insurance application. This automated process can reduce manual input errors and improve efficiency and accuracy.

[0167] In the above embodiments, the present application provides an automated underwriting process, which guides underwriting decisions through a decision tree to ensure the accuracy and completeness of the insurance application.

[0168] See also Figure 7 , Figure 7 It is a specific flow chart of step S630 provided in an embodiment of the present application. In some embodiments of the present application, step S630 may specifically include but is not limited to steps S710 to S740.

[0169] Step S710, selecting valid extracted information that matches the underwriting rule node from the subject's insurance key information;

[0170] Step S720, classifying the effectively extracted information based on a preset information classification model to obtain an information category;

[0171] Step S730, performing information verification on the information category and the valid extracted information based on a preset information verification model to obtain a verification result;

[0172] Step S740: If the verification result is that the verification is passed and the node labels of the underwriting rule nodes are all node matching labels, an insurance application form for the target object is generated based on the object's insurance key information.

[0173] In step S710 of some embodiments, the present application may first select multiple valid extracted information that matches the underwriting rule node from the object's insurance key information. These valid extracted information are key data required in the underwriting process, such as health status, work status, etc.

[0174] In step S720 of some embodiments, the valid extracted information may be further classified based on a preset information classification model to identify the information category of each valid extracted information, for example, health status may be classified as "personal basic information", work status may be classified as "occupational information", etc.

[0175] In step S730 of some embodiments, further, the present application may perform information verification on the information category and the valid extracted information based on a preset information verification model to verify whether the classified information is accurate, complete and meets the requirements of the underwriting rules, and obtain a verification result. Among them, information verification may include checking the format, logical consistency, and matching degree of the information with the underwriting rules. The verification result is used to indicate whether the valid extracted information has passed the verification. If all valid extracted information has passed the verification, the verification result is verification passed. If there is valid extracted information that has not passed the verification, the verification result is verification failed.

[0176] In step S740 of some embodiments, if the verification result is that the verification is passed and the node labels of the underwriting rule nodes are all node matching labels, the present application can generate an insurance application form for the target object based on these verified key information of the object participating in the insurance. This insurance application form will contain all the necessary information to complete the insurance application process for the target object.

[0177] It should be noted that if the verification result includes verification failure, the effective extracted information is reclassified based on the preset information classification model to determine whether accurate information is not extracted. Alternatively, corresponding consultation response scripts can be generated to guide the target object to re-supplement relevant information.

[0178] It should be noted that if the node label of the underwriting rule node contains a node unmatched label, it is necessary to re-match the node information of the object insurance key information and the underwriting rule node based on the node information type until the verification result is passed and the node labels of the underwriting rule nodes are all node matching labels.

[0179] In the above embodiments, the present application provides an automated information processing and verification system, which aims to ensure that the information submitted by the target object is accurate and meets the preset underwriting requirements, thereby improving underwriting efficiency and accuracy.

[0180] See also Figure 8 , Figure 8 It is a specific flow chart of generating an insurance application provided by an embodiment of the present application. Specifically, after obtaining the key information of the object's insurance participation, the present application can obtain valid extraction information (i.e., the object's insurance information) from the key information of the object's insurance participation, match the node information of the object's insurance participation key information and the underwriting rule node based on the node information type of the underwriting rule node in the underwriting rule decision tree, and determine whether the node labels of the underwriting rule nodes are all node matching labels. If so, it means that the match is passed; if not, it means that the match is not passed. At the same time, the present application can classify the valid extracted information based on the preset information classification model to obtain the information category. Further, based on the preset information verification model, it is determined whether the verification result corresponding to the valid extraction information is a verification pass (i.e., all valid extraction information has passed the verification). If the verification result is a verification pass and the node labels of the underwriting rule nodes are all the node matching labels, an insurance application for the target object is generated based on the key information of the object's insurance participation. Further, the target object can confirm whether it agrees with the insurance application and complete the payment to receive the insurance business.

[0181] It should be noted that the non-Company software tools or components appearing in the embodiments of the present application are merely examples and do not represent actual use.

[0182] The method for extracting insurance information provided in the embodiment of the present application can automatically generate business development words according to the conversation habits of the object, completely simulate human conversation, and explore the needs of customers during the conversation, and accurately recommend the insurance products needed by the object. According to the conversation context, the effective information of the object is extracted, and the insurance information is automatically extracted and entered, which improves work efficiency and reduces the time and errors of manual operation. The object information is recorded in real time to facilitate subsequent analysis and management, which can help enterprises improve service quality and customer satisfaction. Compared with the related technology, it is necessary to exhaust the instructions that the user may enter, and it is difficult to meet the complex and changeable scenarios in actual insurance promotion. This application can perform semantic recognition on insurance consultation sub-data with different format types, and generate more accurate consultation prompt data according to the matching target insurance data model, so that more accurate consultation response words can be generated according to the consultation prompt data, which is convenient for information extraction in complex and changeable scenarios. Therefore, the embodiment of the present application can improve the accuracy of insurance information extraction.

[0183] See also Fig. 9 The embodiment of the present application also provides a device for extracting insurance information, which can implement the above-mentioned method for extracting insurance information, and the device includes:

[0184] The first acquisition module 910 is used to acquire the insurance consultation data of the target object, the insurance consultation data includes insurance consultation sub-data, and the format types of any two insurance consultation sub-data are different;

[0185] A semantic recognition module 920 is used to perform semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data with a target insurance type;

[0186] The second acquisition module 930 is used to acquire a target insurance data model matching the target insurance type;

[0187] A generating module 940 is used to generate prompt data based on the semantic recognition data and the target insurance data model to obtain consulting prompt data;

[0188] The intention recognition module 950 is used to perform intention recognition on the consultation prompt data and the semantic recognition data based on the intention recognition sub-model of the preset speech generation model to obtain the insurance intention label of the target object;

[0189] A speech generation module 960 is used to generate speech based on the speech generation sub-model of the preset speech generation model for the insurance intention label and the semantic recognition data to obtain the consultation response speech;

[0190] The third acquisition module 970 is used to acquire the object response data fed back by the target object based on the consultation response words;

[0191] The extraction module 980 is used to extract the insurance information of the object response data to obtain the key insurance information of the object, and the key insurance information of the object matches the target insurance type.

[0192] The specific implementation of the insurance information extraction device in the embodiment of the present application is basically the same as the specific implementation of the above-mentioned insurance information extraction method, and will not be repeated here.

[0193] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned insurance information extraction method when executing the computer program. The electronic device can be any smart terminal including a tablet computer, a car computer, etc.

[0194] See also Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0195] The processor 1010 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0196] The memory 1020 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1020 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1020, and the processor 1010 calls and executes the insurance information extraction method of the embodiment of this application;

[0197] Input / output interface 1030, used to implement information input and output;

[0198] The communication interface 1040 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0199] Bus 1050 , which transmits information between various components of the device (e.g., processor 1010 , memory 1020 , input / output interface 1030 , and communication interface 1040 );

[0200] The processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 are connected to each other in communication within the device via a bus 1050 .

[0201] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned insurance information extraction method is implemented.

[0202] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0203] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0204] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0205] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0206] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0207] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0208] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0209] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0210] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0211] In addition, each functional unit in each embodiment of the present application 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. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0212] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0213] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for extracting insurance information, characterized in that: The method comprises: Obtaining insurance consultation data of the target object, wherein the insurance consultation data includes insurance consultation sub-data, and the format types of any two insurance consultation sub-data are different; Perform semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data of the target insurance type; Acquire a target insurance data model matching the target insurance type; Generate prompt data based on the semantic recognition data and the target insurance data model to obtain consulting prompt data; The intention recognition sub-model based on the preset speech generation model performs intention recognition on the consultation prompt data and the semantic recognition data to obtain the insurance participation intention label of the target object; Based on the speech generation sub-model of the preset speech generation model, speech generation is performed on the insurance intention label and the semantic recognition data to obtain the consultation response speech; Acquiring object response data fed back by the target object based on the consultation response script; The insurance information of the object response data is extracted to obtain the key insurance information of the object, and the key insurance information of the object matches the target insurance type.

2. The method according to claim 1, characterized in that The intent recognition sub-model includes an encoder, a context extraction unit, a position encoder, a self-attention unit and a decoder; The intention recognition sub-model based on the preset speech generation model performs intention recognition on the consultation prompt data and the semantic recognition data to obtain the insurance participation intention label of the target object, including: Based on the encoder, the consultation prompt data is encoded to obtain a consultation prompt sequence; Performing data encoding on the semantic recognition data based on the encoder to obtain a semantic recognition sequence; Performing sequence fusion on the consulting prompt sequence and the semantic recognition sequence to obtain an initial fused sequence; Performing feature extraction on the initial fusion sequence based on the context extraction unit to obtain an initial fusion feature; Performing position encoding on the initial fusion sequence based on the position encoder to obtain a position encoding feature; Based on the self-attention unit, the initial fusion feature and the position encoding feature are subjected to self-attention processing to obtain a target fusion feature; The target fusion feature is decoded based on the decoder to obtain the insurance intention label.

3. The method according to claim 1, characterized in that The speech generation sub-model based on the preset speech generation model generates speech for the insurance intention label and the semantic recognition data to obtain the consultation response speech, including: Selecting the insurance participation rule data of the target insurance type from a preset insurance knowledge graph based on the insurance participation intention label, wherein the insurance participation rule data includes insurance participation rule sub-data; Acquiring initial recognition sub-data from the semantic recognition data; Performing data matching on the initial identification sub-data and the insurance participation rule sub-data to obtain a data matching result; Determining the unuploaded sub-data from the insurance rule sub-data based on the data matching result; Based on the speech generation sub-model, speech generation is performed on the initial recognition sub-data and the non-uploaded sub-data to obtain the consultation response speech.

4. The method according to claim 3, characterized in that The generating speech for the initial recognition sub-data and the non-uploaded sub-data based on the speech generation sub-model to obtain the consultation response speech includes: If the unuploaded sub-data is empty, generate a speech for the initial identification sub-data based on the speech generation sub-model to obtain the consultation response speech, and the consultation response speech is used to guide the target object to confirm the information of the initial identification sub-data; If the non-uploaded sub-data is not empty, speech generation is performed for the non-uploaded sub-data based on the speech generation sub-model to obtain the consultation response speech, and the consultation response speech is used to guide the target object to supplement information of the non-uploaded sub-data.

5. The method according to any one of claims 1 to 4, characterized in that: The semantic recognition is performed based on the insurance consultation sub-data to obtain semantic recognition data with the target insurance type, including: Get the data format type of the insurance consultation sub-data; Determine the target semantic detection model of the insurance consultation sub-data based on the data format type; Performing semantic detection on the insurance consultation sub-data based on the target semantic detection model to obtain a detection text; Performing text enhancement on the detected text to obtain enhanced text; Perform consultation intention identification on the insurance consultation sub-data to obtain a consultation intention label and the target insurance type; Generate a hypothetical document based on the consultation intention label and the enhanced text to obtain a target hypothetical document; The semantic recognition data is obtained by performing data integration based on the enhanced text, the consulting intention label, the target insurance type and the target hypothetical document.

6. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquire an underwriting rule decision tree of the target insurance type, wherein the underwriting rule decision tree includes an underwriting rule node and a node information type of the underwriting rule node; Perform node information matching on the key information of the insurance subject and the underwriting rule node based on the node information type, and determine a node label of the underwriting rule node, wherein the node label includes a node matching label; Generate an insurance application form for the target object based on the node label and the object insurance key information.

7. The method according to claim 6, characterized in that The generating the insurance application form of the target object based on the node label and the object insurance key information includes: Selecting valid extracted information that matches the underwriting rule node from the subject's insurance key information; Classifying the effectively extracted information based on a preset information classification model to obtain an information category; Performing information verification on the information category and the valid extracted information based on a preset information verification model to obtain a verification result; If the verification result is that the verification is passed and the node labels of the underwriting rule nodes are all the node matching labels, the insurance application form for the target object is generated based on the key insurance information of the object.

8. A device for extracting insurance information, characterized in that: The device comprises: The first acquisition module is used to acquire the insurance consultation data of the target object, wherein the insurance consultation data includes insurance consultation sub-data, and the format types of any two insurance consultation sub-data are different; A semantic recognition module, used for performing semantic recognition based on the insurance consultation sub-data to obtain semantic recognition data of the target insurance type; A second acquisition module, used to acquire a target insurance data model matching the target insurance type; A generating module, used for generating prompt data based on the semantic recognition data and the target insurance data model to obtain consulting prompt data; An intention recognition module, used to perform intention recognition on the consultation prompt data and the semantic recognition data based on the intention recognition sub-model of the preset speech generation model, and obtain the insurance intention label of the target object; A speech generation module, used for generating speech for the insurance intention label and the semantic recognition data based on the speech generation sub-model of the preset speech generation model to obtain the consultation response speech; A third acquisition module is used to acquire object response data fed back by the target object based on the consultation response script; The extraction module is used to extract the insurance information of the object response data to obtain the key insurance information of the object, and the key insurance information of the object matches the target insurance type.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images