Insurance question and answer method, device and equipment and storage medium

Abstract

The invention belongs to the field of artificial intelligence, and relates to an insurance-based question and answer method, which comprises the following steps: receiving an insurance question input by a user, and extracting insurance question characteristics of the insurance question; selecting a splitting strategy corresponding to the insurance problem characteristics according to the insurance problem characteristics, and splitting the insurance problem according to the splitting strategy to obtain a plurality of insurance sub-problems; based on each insurance sub-question, processing each insurance sub-question, generating an answer corresponding to each insurance sub-question, and obtaining a plurality of initial answers; and integrating all the initial answers to generate a target answer, and feeding back the target answer to the user. The invention further provides an insurance-based question and answer device, equipment and a medium. In addition, the invention also relates to a block chain technology, and insurance questions, target answers and the like can be stored in a block chain. The insurance question and answer accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to an insurance question-answering method. Background Art

[0002] With the rapid development of the insurance industry, there are more and more types of insurance and more and more customers who buy insurance. With the surge in users, the workload of the insurance service team is also increasing;

[0003] At present, the intelligent insurance question-answering system is used to automatically identify the insurance question texts raised by customers and give corresponding answers based on the insurance question texts. As a key tool to improve service quality and efficiency, the intelligent insurance question-answering system can not only answer the questions raised by users in a timely manner, but also help staff reduce work pressure;

[0004] However, the existing intelligent insurance question and answer system can only retrieve the answer corresponding to the insurance question from the insurance question and answer database based on the insurance question raised by the user and feedback it to the user. It is unable to handle the deeper content of the insurance question, resulting in low accuracy of the insurance question and answer. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose an insurance question and answer method, device, equipment and storage medium, the main purpose of which is to improve the accuracy of insurance question and answer.

[0006] In order to solve the above technical problems, the embodiment of the present application provides an insurance question-and-answer method, which adopts the following technical solution:

[0007] Receiving an insurance question input by a user, and extracting insurance question features of the insurance question;

[0008] According to the characteristics of the insurance problem, a splitting strategy corresponding to the characteristics of the insurance problem is selected, and the insurance problem is split according to the splitting strategy to obtain multiple insurance sub-problems;

[0009] Based on each of the insurance sub-problems, each of the insurance sub-problems is processed to generate an answer corresponding to each of the insurance sub-problems to obtain multiple initial answers;

[0010] All the initial answers are integrated to generate a target answer, and the target answer is fed back to the user.

[0011] Furthermore, the extracting of insurance problem features of the insurance problem includes:

[0012] Performing part-of-speech tagging on the insurance question to obtain the insurance question after part-of-speech tagging;

[0013] Using a pre-trained topic model to perform topic identification on the insurance questions after part-of-speech tagging, to obtain the topic of the insurance questions;

[0014] extracting insurance problem keywords from the insurance problem according to the subject of the insurance problem;

[0015] Feature extraction is performed based on the insurance issue keywords to obtain insurance issue features of the insurance issue.

[0016] Furthermore, after receiving the insurance question input by the user, the method further includes:

[0017] Identify the type of data of the insurance issue;

[0018] If the data type is identified as a non-text form, obtaining a text conversion method for the non-text form;

[0019] According to the text conversion method, the insurance question is converted into text form.

[0020] Furthermore, according to the characteristics of the insurance problem, a splitting strategy corresponding to the characteristics of the insurance problem is selected, and the insurance problem is split according to the splitting strategy to obtain multiple insurance sub-problems, including:

[0021] Processing the features of the insurance problem by a pre-trained classifier to obtain the problem type of the insurance problem;

[0022] According to the pre-built strategy mapping table and the problem type, select the splitting strategy corresponding to the insurance problem;

[0023] The insurance problem is split according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems.

[0024] Furthermore, the insurance problem is split according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems, including:

[0025] Semantically matching the insurance problem feature with each node in the insurance knowledge graph to obtain matching nodes;

[0026] Obtaining the hierarchical relationship of the matching node in the insurance knowledge graph;

[0027] If the splitting strategy is a process splitting strategy, the insurance problem is split according to the horizontal hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems;

[0028] If the splitting strategy is a conditional splitting strategy, the insurance problem is split according to the vertical hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems.

[0029] Furthermore, each of the insurance sub-problems is processed to generate an answer corresponding to each of the insurance sub-problems, and multiple initial answers are obtained, including:

[0030] Based on each of the insurance sub-problems, converting the insurance sub-problems into vector representation through the input layer of the feature extraction model to obtain an insurance sub-problem vector;

[0031] Extracting local features of the insurance sub-problem vector through the convolutional layer of the feature extraction model;

[0032] Inputting the local features of the insurance sub-problem vector into the pooling layer of the feature extraction model to perform a dimensionality reduction operation to obtain a key vector of the insurance sub-problem;

[0033] Performing similarity matching between the key vector and each entry in the insurance database to obtain a matching result;

[0034] A search path is constructed according to the matching results, and the search path is executed to obtain an initial answer to the insurance sub-question.

[0035] Furthermore, after constructing a search path according to the matching results and executing the search path to obtain an initial answer to the insurance sub-question, the method further includes:

[0036] identifying the answer number of the initial answer;

[0037] When the number of answers is greater than one, obtaining basic information and behavior data of the user;

[0038] Answers to questions that interest the user are selected from the initial answers based on the basic information and behavior data of the user.

[0039] In order to solve the above technical problems, the embodiment of the present application further provides an insurance question-and-answer device, which adopts the following technical solution:

[0040] An extraction module, used to receive the insurance question input by the user and extract the insurance question features of the insurance question;

[0041] A splitting module, configured to select a splitting strategy corresponding to the characteristics of the insurance problem according to the characteristics of the insurance problem, and split the insurance problem according to the splitting strategy to obtain a plurality of insurance sub-problems;

[0042] A generating module, used for processing each of the insurance sub-problems based on each of the insurance sub-problems, generating an answer corresponding to each of the insurance sub-problems, and obtaining a plurality of initial answers;

[0043] The integration module is used to integrate all the initial answers, generate a target answer, and feed back the target answer to the user.

[0044] In order to solve the above technical problems, an embodiment of the present application also provides a computer device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the insurance question and answer method as described above.

[0045] In order to solve the above technical problems, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which implements the above-mentioned insurance question-and-answer method when executed by a processor.

[0046] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0047] By extracting the features of the question text and splitting it based on the features, the traditional insurance Q&A method can avoid the limitation of only answering based on the literal meaning, improve the flexibility of Q&A, and improve the ability to handle complex question texts, and improve the accuracy of the subsequent generation of target answers;

[0048] By selecting the splitting strategy corresponding to the characteristics of the insurance problem, the accuracy of the insurance sub-problem splitting is improved; by splitting the insurance problem into multiple insurance sub-problems, it is helpful to more accurately identify the comprehensiveness of the insurance problem. Based on the multiple insurance sub-problems after splitting, the comprehensiveness of the target answer generated subsequently is ensured, and the complex insurance problem is simplified, the difficulty of generating the target answer is reduced, and the efficiency of generating the subsequent target answer is improved;

[0049] By splitting insurance into multiple insurance sub-questions and generating initial answers to the insurance sub-questions based on the insurance sub-questions, the problem of being unable to generate in-depth answers in the existing technology can be avoided, thereby improving the accuracy of insurance question-answering.

[0050] By integrating all the initial answers and ultimately generating the target answers corresponding to the insurance questions, we can achieve in-depth answers to insurance questions, improve the accuracy and efficiency of answering insurance questions, and the integrated target answers are clear and concise, which further improves the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0053] Figure 2 A flowchart of an embodiment of the insurance question-and-answer method according to the present application;

[0054] Figure 3 is a structural schematic diagram of an embodiment of an insurance question-and-answer device according to the present application;

[0055] Figure 4 It is a structural schematic diagram of an embodiment of a device according to the present application. DETAILED DESCRIPTION

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0057] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0058] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0059] like Figure 1As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0060] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0061] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0062] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0063] It should be noted that the insurance question and answer method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the insurance question and answer device is generally set in the server / terminal device.

[0064] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0065] Continue to refer Figure 2 , shows a flowchart of an embodiment of the insurance question-and-answer method according to the present application. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted. The insurance question-and-answer method provided in the embodiment of the present application can be applied to any scenario where insurance question-and-answer is required, and the insurance question-and-answer method can also be applied to products in these scenarios. The insurance question-and-answer method includes the following steps:

[0066] Step S201: receiving an insurance question input by a user, and extracting insurance question features of the insurance question.

[0067] In this embodiment, the insurance question-answering method is executed on an electronic device (eg Figure 1 The server / terminal device shown in the figure) can receive the query information input by the user through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future. In the business scenario of insurance questions and answers, insurance questions can be input through the user's handheld terminal or fixed terminal.

[0068] In this embodiment, the insurance question is an insurance question corresponding to a question answering request raised by the user based on his or her own needs, for example, "I would like to know what coverage items are included in your company's medical insurance?" or "How to apply for a claim?" The user raises insurance-related insurance questions through a terminal, and the above-mentioned terminal includes but is not limited to a handheld terminal (such as a mobile phone) and a fixed terminal (such as a computer), and insurance-related insurance questions raised through various channels (such as a website, APP (application), mini-program or customer service hotline, etc.) in the terminal.

[0069] An insurance question input by a user is received, the insurance question is preprocessed, and features are extracted from the preprocessed insurance question to obtain insurance question features of the insurance question.

[0070] Specifically, an insurance question input by a user is received, and it is detected whether there are spelling errors in the insurance question. If so, the errors are corrected; then a text cleaning operation is performed on the corrected insurance question text to remove irrelevant characters in the insurance question, such as HTML tags, special symbols, extra spaces, non-standard symbols, etc., to obtain a pre-processed insurance question, and a pre-trained topic model is used as an LDA (Latent Dirichlet Allocation) model to perform feature extraction on the pre-processed insurance question to obtain the insurance question features of the insurance question.

[0071] In this embodiment, by extracting the features of the question text and subsequently splitting it based on the features, the limitation of the traditional insurance question and answer method that can only answer based on the literal meaning is avoided, the flexibility of the question and answer is improved, the ability to process complex question texts is improved, and the accuracy of the subsequent generation of the target answer is improved.

[0072] Step S202: According to the characteristics of the insurance problem, a splitting strategy corresponding to the characteristics of the insurance problem is selected, and the insurance problem is split according to the splitting strategy to obtain multiple insurance sub-problems.

[0073] In this embodiment, a splitting strategy library is pre-constructed, and the splitting strategy library stores multiple splitting strategies, such as "conditional splitting strategy" and "process splitting strategy", etc.; and an insurance knowledge graph is pre-constructed. The above-mentioned insurance knowledge graph is a structured knowledge base that includes insurance-related terms, concepts, policies and rules, etc. The insurance knowledge graph represents the knowledge in the insurance field in the form of nodes and edges, wherein nodes represent various concepts or entities (such as "medical insurance", "auto insurance", etc.), and edges represent the relationship between these concepts or entities (such as "medical insurance" is associated with a variety of different types of medical insurance), such as "medical insurance", "auto insurance", "health insurance", etc., and "medical insurance" is associated with a variety of different medical insurances, etc., "auto insurance" is associated with "compulsory traffic insurance", "commercial insurance", etc., as well as the specific application process corresponding to each insurance.

[0074] In this embodiment, the insurance problem features are input into a pre-trained classifier (such as a decision tree, a neural network classifier, etc.), the problem type of the insurance problem is identified by the pre-trained classifier, and a strategy mapping table of problem types and splitting strategies is pre-constructed. The strategy mapping table includes a mapping relationship between all problem types and splitting strategies, such as a mapping relationship between process problem types and process splitting strategies; according to the strategy mapping table and the problem type, a splitting strategy corresponding to the insurance problem is selected from the splitting strategy library, and the insurance problem is split by the splitting strategy and the insurance knowledge graph to obtain multiple insurance sub-problems corresponding to the insurance problem.

[0075] For example, the insurance question is "I want to know what insurance items are included in your company's medical insurance?" The corresponding splitting strategy is "conditional splitting strategy". Through the conditional splitting strategy and the insurance knowledge graph, the above insurance question is split into multiple insurance sub-questions: Insurance sub-question 1: What medical insurance products does the company have; Insurance sub-question 2: What insurance items does each medical insurance product include? 3. What are the similarities and differences in the insurance items of different medical insurance products?

[0076] For example, the insurance question is "How to apply for a claim?" The corresponding splitting strategy is "Process Splitting Strategy". Through the process splitting strategy and the insurance knowledge graph, the above insurance question is split into multiple insurance sub-questions: Insurance sub-question 1: How can I confirm whether I have the conditions for claiming? Insurance sub-question 2: What materials do I need to prepare for claiming? Insurance sub-question 3: What are the steps for applying for a claim?

[0077] In this embodiment, by selecting a splitting strategy corresponding to the characteristics of the insurance problem, the accuracy of splitting the insurance sub-problem is improved; by splitting the insurance problem to obtain multiple insurance sub-problems, it is helpful to more accurately identify the comprehensiveness of the insurance problem, and based on the multiple insurance sub-problems after the split, the comprehensiveness of the target answer generated subsequently is ensured, and the complex insurance problem is simplified, the difficulty of generating the target answer is reduced, and the efficiency of generating the subsequent target answer is improved.

[0078] Step S203: Based on each of the insurance sub-problems, process each of the insurance sub-problems, generate an answer corresponding to each of the insurance sub-problems, and obtain multiple initial answers.

[0079] In this embodiment, based on each of the insurance sub-problems, a retrieval path is constructed in a pre-constructed insurance database. The constructing of the retrieval path refers to constructing a corresponding retrieval path in the insurance database according to the characteristics of the insurance sub-problem, such as "what medical insurance products does the company have", and the path can be constructed as medical insurance-type-quantity; specifically, for each insurance sub-problem, the key features of each insurance sub-problem are extracted through a feature extraction model, and the key features are matched in the insurance database to obtain a retrieval path, which is executed to finally obtain an initial answer corresponding to each insurance sub-problem, wherein the above-mentioned feature extraction model refers to a convolutional neural network, which includes an input layer, multi-layer convolutional layers, a pooling layer and an output layer, and the insurance sub-problem is input into the feature extraction model through the input layer, and feature extraction operations are performed on the insurance sub-problem through multi-layer convolutional layers and pooling layers to obtain the key features corresponding to the insurance sub-problem, and the key features corresponding to the insurance sub-problem are output through the output layer.

[0080] It is worth mentioning that the above feature extraction model can also be a recurrent neural network, a BERT model, etc.

[0081] In this embodiment, in order to ensure the accuracy of the initial answer corresponding to the insurance sub-question, it is necessary to obtain the insurance change policy in real time, and when the insurance change policy is updated, the insurance database is updated synchronously.

[0082] In this embodiment, by splitting insurance into multiple insurance sub-questions and generating initial answers corresponding to the insurance sub-questions based on the insurance sub-questions, the inability to generate in-depth answers in the prior art can be avoided, thereby improving the accuracy of insurance questions and answers.

[0083] Step S204: Integrate all the initial answers to generate a target answer, and feed back the target answer to the user.

[0084] In this embodiment, all the initial answers generated in step S203 are obtained, and a text comparison algorithm (such as cosine similarity, Jaccard similarity algorithm or hash value comparison method) is used to check whether there are duplicate contents in all the initial answers. If there are duplicate contents, duplicate removal is performed to obtain multiple initial answers after duplicate removal, and the initial answers after duplicate removal are integrated, wherein the integration includes logic combing, content integration, and format adjustment. The above-mentioned logic combing refers to sorting the initial answers after duplicate removal according to the logical order of the insurance sub-questions or the possible order of understanding of the user, and the above-mentioned content integration refers to combining all the sorted initial answers. And, get the initial target answer. The above-mentioned format adjustment refers to adjusting the initial target answer according to the preset format template to ensure that the format of the target answer is clear and easy to read, etc., and finally generate the target answer, and perform accuracy verification and completeness verification on the target answer. Among them, the accuracy verification is to ensure that the content of the target answer is correct, and the target answer is compared with the known information in the insurance knowledge graph. The target answer can also be automatically verified through natural language processing or neural networks, which is not restricted here; the completeness check is to check whether the target answer covers all questions raised by the user to ensure that there are no omissions, and compare the target answer with the insurance question.

[0085] In this embodiment, by integrating all the initial answers and finally generating the target answer corresponding to the insurance question, an in-depth answer to the insurance question is achieved, the accuracy and efficiency of answering the insurance question are improved, and the integrated target answer is clear and concise, which further improves the user experience.

[0086] In one embodiment, extracting the insurance problem features of the insurance problem includes:

[0087] Performing part-of-speech tagging on the insurance question to obtain the insurance question after part-of-speech tagging;

[0088] Using a pre-trained topic model to perform topic identification on the insurance questions after part-of-speech tagging, to obtain the topic of the insurance questions;

[0089] extracting insurance problem keywords from the insurance problem according to the subject of the insurance problem;

[0090] Feature extraction is performed based on the insurance issue keywords to obtain insurance issue features of the insurance issue.

[0091] In this embodiment, the above-mentioned part-of-speech tagging can use a preset part-of-speech tagging tool to tag the insurance questions, such as jieba (jieba segmentation), SnowNLP, PkuSeg, etc., by segmenting the insurance questions and then tagging each segmentation with part-of-speech, such as marking verbs (v), nouns (n), adjectives (adj), etc.; the above-mentioned pre-trained topic model is an LDA (Latent Dirichlet Allocation) model; the insurance questions are segmented and each segmentation is tagged with part-of-speech, and then all the segmentations with part-of-speech tags are summarized to obtain the insurance questions after part-of-speech tagging, and the insurance questions after part-of-speech tagging are used as input of the LDA model, and the pre-trained topic model is used to perform topic identification on the insurance questions after part-of-speech tagging to obtain the probability of each topic in insurance, select the topic with the highest probability as the topic of the insurance question, and output the topic of the insurance question, according to the topic of the insurance question, extract the insurance question keywords from the insurance question, and perform feature extraction on the insurance keywords, specifically, the keywords can be converted into feature vectors to finally obtain the insurance question features.

[0092] In this embodiment, a large amount of training data is also required to train the LDA model. The training data includes a large number of insurance questions and corresponding topic labels. The LDA is trained in a supervised manner. When the accuracy reaches the accuracy threshold, the training is completed to obtain a pre-trained LDA model.

[0093] It is worth mentioning that the above is not limited to using topic models to extract problematic insurance problem features, but other models (such as trained neural networks, etc.) can also be used.

[0094] In this embodiment, by tagging the insurance questions with parts of speech, the part of speech and grammatical function of each word can be clarified, which provides a basis for subsequent topic identification and keyword extraction, and improves the accuracy of insurance question features; by using the LDA model to identify the topics of the insurance questions after part of speech tagging, and extracting keywords from the topic, the core topic of the insurance question and the keywords closely related to the topic can be automatically and efficiently identified, and manual identification of keywords can be avoided, reducing manual intervention, thereby improving the accuracy and efficiency of extracting insurance features.

[0095] In one embodiment, after receiving the insurance question input by the user, the method further comprises:

[0096] Identify the type of data of the insurance issue;

[0097] If the data type is identified as a non-text form, obtaining a text conversion method for the non-text form;

[0098] According to the text conversion method, the insurance question is converted into text form.

[0099] In this embodiment, after receiving the insurance question input by the user, the specific query method can be directly judged. If the insurance question is a string type, it is judged that the insurance question is in text form and no conversion is performed; if the insurance question exists in the form of a file, the file extension is checked. When the file extension is .jpg, .png, etc., the insurance question is judged to be in image form. When the file extension is .wav, etc., the insurance question is judged to be an audio file; when the insurance question is in image form, OCR technology is used to extract the insurance question text to obtain the insurance question in text form. When the insurance question is in audio form, speech recognition technology (such as Speech-to-Text API, etc.) is used to convert the insurance question into text information.

[0100] In this embodiment, by supporting insurance questions input in various forms, the user experience can be improved, the flexibility of the query is increased, and the query scope is expanded, so as to improve the accuracy of query statement generation in the subsequent process.

[0101] In one embodiment, the step S202, according to the characteristics of the insurance problem, selects a splitting strategy corresponding to the characteristics of the insurance problem, splits the insurance problem according to the splitting strategy, and obtains multiple insurance sub-problems, including:

[0102] Processing the features of the insurance problem by a pre-trained classifier to obtain the problem type of the insurance problem;

[0103] According to the pre-built strategy mapping table and the problem type, select the splitting strategy corresponding to the insurance problem;

[0104] The insurance problem is split according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems.

[0105] In this embodiment, before the insurance question features are processed by a pre-trained classifier, the classifier needs to be trained. Specifically, a large number of insurance questions are obtained from a training database, which refers to data obtained from channels such as the customer service center and online platform of an insurance company as training data, and stored in a database to obtain a training database. The collected data is cleaned, deduplicated, and invalid or noisy to obtain pre-processed training data. The pre-processed training data is segmented, and each segmented word is tagged with a part of speech to obtain training data with a part of speech tag. A label is assigned to each training data to indicate the type of question to which it belongs. The classifier is trained with the labeled training data to obtain a pre-trained classifier.

[0106] In this embodiment, the insurance problem features are processed by a pre-trained classifier, and the classifier outputs the problem type corresponding to the insurance problem features, that is, the problem type of the insurance problem. According to the pre-built strategy mapping table mentioned in the above step S202 and the problem type output by the classifier, the corresponding splitting strategy is selected, and the insurance problem is split according to the splitting strategy to obtain multiple insurance sub-problems.

[0107] In this embodiment, the insurance problem features are quickly processed by a pre-trained classifier, and the problem type can be accurately identified, thereby avoiding the errors and time-consuming manual classification. For different problem types, corresponding splitting strategies are selected for splitting, which can more accurately locate the problem, achieve in-depth processing of insurance problems, and improve the accuracy and efficiency of problem processing.

[0108] In another embodiment, the insurance problem is split according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems, including:

[0109] Semantically matching the insurance problem feature with each node in the insurance knowledge graph to obtain matching nodes;

[0110] Obtaining the hierarchical relationship of the matching node in the insurance knowledge graph;

[0111] If the splitting strategy is a process splitting strategy, the insurance problem is split according to the horizontal hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems;

[0112] If the splitting strategy is a conditional splitting strategy, the insurance problem is split according to the vertical hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems.

[0113] In this embodiment, the above-mentioned horizontal hierarchical relationship refers to sub-nodes in the same hierarchical relationship. For example, in the claims process, the claim conditions, materials required for claims, and claims steps all belong to the same level, and their relationship is defined as a horizontal hierarchical relationship; the above-mentioned vertical hierarchical relationship refers to hierarchical relationships at different levels. For example, what protection items are included in medical insurance, the sub-node is medical insurance, the next level is the type of medical insurance, and the next level is the protection items included in different types of medical insurance.

[0114] Traverse the semantic features of each node in the insurance knowledge graph, calculate the first semantic similarity value between the insurance question feature and the semantic feature of each node, which can be calculated by a calculation method such as cosine similarity, and select the node with the highest similarity value as the matching node;

[0115] Obtain a splitting strategy corresponding to the insurance problem feature, determine the type of the splitting strategy, if the splitting strategy is a process splitting strategy, then according to the horizontal hierarchical relationship of the hierarchical relationship, obtain all sub-node contents from the same-level relationship, construct an initial insurance sub-problem according to the content of each sub-node, obtain a plurality of the initial insurance sub-problems, extract the semantic features of each initial insurance sub-problem, calculate the second semantic similarity value between the insurance problem feature and the semantic features of each initial insurance sub-problem, compare the second semantic similarity value with a preset threshold, retain the initial insurance sub-problems corresponding to the second semantic similarity value greater than or equal to the preset threshold, delete the initial insurance sub-problems corresponding to the second semantic similarity value less than the preset threshold, and finally obtain a plurality of the insurance sub-problems.

[0116] In this embodiment, by splitting according to the insurance knowledge graph and the corresponding splitting strategy, it is possible to understand the various components of the problem more carefully, thereby achieving accurate positioning of the problem; by splitting complex insurance problems into multiple insurance sub-problems, the difficulty of handling insurance problems can be reduced, and multiple insurance sub-problems can be processed in parallel, thereby improving the efficiency of handling insurance problems.

[0117] In one embodiment, each of the insurance sub-questions is processed to generate an answer corresponding to each of the insurance sub-questions to obtain multiple initial answers, including:

[0118] Based on each of the insurance sub-problems, converting the insurance sub-problems into vector representation through the input layer of the feature extraction model to obtain an insurance sub-problem vector;

[0119] Extracting local features of the insurance sub-problem vector through the convolutional layer of the feature extraction model;

[0120] Inputting the local features of the insurance sub-problem vector into the pooling layer of the feature extraction model to perform a dimensionality reduction operation to obtain a key vector of the insurance sub-problem;

[0121] Performing similarity matching between the key vector and each entry in the insurance database to obtain a matching result;

[0122] A search path is constructed according to the matching results, and the search path is executed to obtain an initial answer to the insurance sub-question.

[0123] In this embodiment, each insurance sub-problem is processed and each insurance sub-problem is input into the model through the input layer of the feature extraction model. In the input layer, the insurance sub-problem is converted into a vector representation through a vector conversion technology (such as Word2Vec, etc.) to obtain an insurance sub-problem vector; feature extraction is performed on the insurance sub-problem vector through the convolution layer of the feature extraction model. Specifically, in the convolution layer, multiple convolution kernels are slid on the insurance sub-problem vector to extract local features of the insurance sub-problem vector. The local features are processed through the pooling layer of the feature extraction model to reduce the dimension of the local features and the amount of calculation, so as to obtain the key features of the insurance sub-problem.

[0124] Different from the conventional model, in this embodiment, after obtaining the key features, the key features are not directly output through the output layer of the feature extraction model, but the key features are matched with each entry in the insurance database connected to the feature extraction model. The entry refers to the basic component unit of the insurance database, and each entry contains information related to insurance, such as detailed information, terms and conditions of each insurance product, etc. Specifically, the entire insurance database is traversed, and each entry in the insurance database is converted into a vector representation to obtain multiple entry features. Specifically, the above-mentioned vector conversion technology (such as Word2Vec, etc.) can be used to calculate the similarity between the key features and each entry feature (such as pre-similarity, Euclidean distance, etc.). For each entry, its similarity with the relevant The similarity score of the key vector is calculated to obtain the matching result; after obtaining a series of matching results (each matching result contains the data of the entry and the similarity score of the corresponding entry feature and the key feature), all the matching results are sorted, and the entry with the highest score is selected as the starting point of the retrieval path, and other entries are retrieved based on the starting point. For example, whether the conditions for claim are met is the starting point, and the other entries are whether the conditions for claim are met - customer service consultation, whether the conditions for claim are met - self-inquiry, and finally the retrieval path is obtained, and the corresponding initial answer to the insurance sub-problem is executed. For example, the initial answer to the insurance sub-problem is, please call the customer service hotline or online APP to contact your customer service to confirm whether you have the conditions for claim; or please confirm whether you have the conditions for claim on the claims page of the APP.

[0125] In this embodiment, by matching the key features with each entry in the insurance database for similarity, the entry most relevant to the question can be accurately found, thereby improving the accuracy and pertinence of the initial answer; by constructing a reasonable search path and selecting the starting point based on the ranking of the matching results, the insurance database can be efficiently traversed to find the entry most relevant to the question, which is more accurate and efficient than traditional keyword-based retrieval.

[0126] In another embodiment, after constructing a search path according to the matching results and executing the search path to obtain an initial answer to the insurance sub-question, the method further includes:

[0127] identifying the answer number of the initial answer;

[0128] When the number of answers is greater than one, obtaining basic information and behavior data of the user;

[0129] Answers to questions that interest the user are selected from the initial answers based on the basic information and behavior data of the user.

[0130] In this embodiment, the application only requires an answer corresponding to one insurance sub-question, so after obtaining the initial answer to the insurance sub-question, it is also necessary to identify the number of answers to the initial answer. If the number of answers to the initial answer is one, the initial answer is integrated into the target answer through a subsequent integration step and fed back to the user. If the number of initial answers is greater than one, the user's basic information and behavior data are obtained, and the user's preference information is obtained from the customer's basic information and behavior data. For example, if the user's age is 60, it is considered that the user prefers to talk to manual customer service to directly confirm whether the conditions for claim are met; if the user's age is 25, it is considered that the user prefers to directly confirm whether the conditions for claim are met through APP page operations. The only initial answer corresponding to the sub-insurance question is selected according to the user's preference information.

[0131] Specifically, the above basic information includes but is not limited to age, gender, occupation, etc., and behavioral data includes but is not limited to the frequency and duration of using the APP; the collected basic information and behavioral data are preprocessed, and the basic information and behavioral data are processed through a preference model to output the predicted results of user preferences, such as which method is preferred to confirm claim eligibility, etc. For example, the user age is divided into multiple intervals, such as 18-24 years old, 25-34 years old, 35-44 years old, 45-54 years old, 55 years old and above, etc.; for each age range, based on historical data and user feedback, the claim confirmation method preferred by users in the range is inferred, and the preference results are: young users may prefer to confirm the claim conditions through APP page operations, while middle-aged and elderly users may prefer to confirm by calling manual customer service.

[0132] The preference model can use the same architecture as the feature extraction model, but the training data and other aspects are different.

[0133] In this embodiment, by collecting and analyzing the user's basic information and behavioral data, the user's preferences and needs can be identified more accurately; based on the identification results of the user's preferences, the answer that best meets the user's needs can be selected from multiple initial answers for feedback, thereby improving user satisfaction; when there are multiple initial answers, by identifying the user's preferences, it is possible to avoid feeding back all answers to the user, thereby reducing the interference of redundant information and improving the accuracy and pertinence of the answers.

[0134] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned insurance question and answer results, etc., the above-mentioned insurance question and answer results, etc. can also be stored in a node of a blockchain.

[0135] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0136] 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.

[0137] 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.

[0138] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0139] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0140] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of an insurance question-and-answer device. Figure 2 Corresponding to the method embodiment shown, the apparatus can be specifically applied to various computer devices.

[0141] like Figure 3 As shown, the insurance question-answering device 300 described in this embodiment includes: an extraction module 301, a splitting module 302, a generation module 303, and an integration module 304. Among them:

[0142] An extraction module 301 is used to receive an insurance question input by a user and extract insurance question features of the insurance question;

[0143] In one embodiment, the extraction module 301 includes:

[0144] A tagging submodule, used for performing part-of-speech tagging on the insurance question to obtain the insurance question after part-of-speech tagging;

[0145] An identification submodule, for performing topic identification on the insurance questions after part-of-speech tagging using a pre-trained topic model to obtain the topic of the insurance questions;

[0146] A first extraction submodule, for extracting insurance question keywords from the insurance question according to the subject of the insurance question;

[0147] The second extraction submodule is used to perform feature extraction based on the insurance question keywords to obtain insurance question features of the insurance question.

[0148] In one embodiment, the apparatus further comprises:

[0149] A type identification module, used for identifying the data type of the insurance question;

[0150] A form acquisition module, for acquiring a text conversion method of the non-text form if the data type is identified as a non-text form;

[0151] A form conversion module is used to convert the insurance question into a text form according to the text conversion method.

[0152] The splitting module 302 is used to select a splitting strategy corresponding to the characteristics of the insurance problem according to the characteristics of the insurance problem, and split the insurance problem according to the splitting strategy to obtain multiple insurance sub-problems.

[0153] In one embodiment, the splitting module 302 includes:

[0154] A classification submodule, used for processing the features of the insurance problem by using a pre-trained classifier to obtain the problem type of the insurance problem;

[0155] A selection submodule, used to select a splitting strategy corresponding to the insurance problem according to a pre-built strategy mapping table and the problem type;

[0156] The splitting submodule is used to split the insurance problem according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems.

[0157] In another embodiment, the splitting submodule comprises:

[0158] A matching subunit, used for semantically matching the insurance question feature with each node in the insurance knowledge graph to obtain a matching node;

[0159] A hierarchical relationship acquisition subunit, used to acquire the hierarchical relationship of the matching node in the insurance knowledge graph;

[0160] A first splitting subunit is used for splitting the insurance problem according to the horizontal hierarchical relationship of the hierarchical relationship to obtain a plurality of the insurance sub-problems if the splitting strategy is a process splitting strategy;

[0161] The second splitting subunit is used to split the insurance problem according to the vertical hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems if the splitting strategy is a conditional splitting strategy.

[0162] A generating module 303 is used to process each of the insurance sub-problems based on each of the insurance sub-problems, generate an answer corresponding to each of the insurance sub-problems, and obtain multiple initial answers;

[0163] In one embodiment, the generating module 303 includes:

[0164] A problem input submodule, used for converting each of the insurance sub-problems into a vector representation through the input layer of the feature extraction model to obtain an insurance sub-problem vector;

[0165] A problem local feature extraction submodule, used to extract local features of the insurance sub-problem vector through the convolution layer of the feature extraction model;

[0166] A problem key vector extraction submodule, used for inputting the local features of the insurance sub-problem vector into the pooling layer of the feature extraction model for dimensionality reduction operation to obtain the key vector of the insurance sub-problem;

[0167] A similarity calculation submodule, used to perform similarity matching between the key vector and each entry in the insurance database to obtain a matching result;

[0168] The path construction submodule is used to construct a search path according to the matching results, and execute the search path to obtain an initial answer to the insurance sub-question.

[0169] In another embodiment, the generating module 303 further includes:

[0170] A quantity identification submodule, used to identify the number of answers to the initial answer;

[0171] A user information acquisition submodule, used to acquire basic information and behavior data of the user when the number of answers is greater than one;

[0172] The interest answer selection submodule is used to select answers to questions that the user is interested in from the initial answers based on the user's basic information and behavior data.

[0173] The integration module 304 is used to integrate all the initial answers, generate a target answer, and feed back the target answer to the user.

[0174] In order to solve the above technical problems, the present application embodiment also provides a device (computer device). Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0175] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 4 with a memory 41, a processor 42, and a network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0176] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0177] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the insurance question-and-answer method, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0178] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the insurance question-and-answer method.

[0179] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0180] During the implementation process, the electronic device of the present application extracts the features of the question text and then splits and processes it based on the features, thereby avoiding the limitation of the traditional insurance question-and-answer method that can only answer based on the literal meaning, improving the flexibility of question-and-answer, and improving the processing ability of complex question texts, and improving the accuracy of the subsequent generation of target answers;

[0181] By selecting the splitting strategy corresponding to the characteristics of the insurance problem, the accuracy of the insurance sub-problem splitting is improved; by splitting the insurance problem into multiple insurance sub-problems, it is helpful to more accurately identify the comprehensiveness of the insurance problem. Based on the multiple insurance sub-problems after splitting, the comprehensiveness of the target answer generated subsequently is ensured, and the complex insurance problem is simplified, the difficulty of generating the target answer is reduced, and the efficiency of generating the subsequent target answer is improved;

[0182] By splitting insurance into multiple insurance sub-questions and generating initial answers to the insurance sub-questions based on the insurance sub-questions, the problem of being unable to generate in-depth answers in the existing technology can be avoided, thereby improving the accuracy of insurance question-answering.

[0183] By integrating all the initial answers and ultimately generating the target answers corresponding to the insurance questions, we can achieve in-depth answers to insurance questions, improve the accuracy and efficiency of answering insurance questions, and the integrated target answers are clear and concise, which further improves the user experience.

[0184] The present application also provides another implementation, namely, providing a storage medium (computer-readable storage medium), wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the insurance question-and-answer method as described above.

[0185] During the implementation process, the computer-readable storage medium of the present application extracts the features of the question text and then performs split processing based on the features, thereby avoiding the limitation of the traditional insurance question-and-answer method that can only answer based on the literal meaning, improving the flexibility of the question-and-answer method, and improving the processing capability of complex question texts, thereby improving the accuracy of the subsequent generation of target answers;

[0186] By selecting the splitting strategy corresponding to the characteristics of the insurance problem, the accuracy of the insurance sub-problem splitting is improved; by splitting the insurance problem into multiple insurance sub-problems, it is helpful to more accurately identify the comprehensiveness of the insurance problem. Based on the multiple insurance sub-problems after splitting, the comprehensiveness of the target answer generated subsequently is ensured, and the complex insurance problem is simplified, the difficulty of generating the target answer is reduced, and the efficiency of generating the subsequent target answer is improved;

[0187] By splitting insurance into multiple insurance sub-questions and generating initial answers to the insurance sub-questions based on the insurance sub-questions, the problem of being unable to generate in-depth answers in the existing technology can be avoided, thereby improving the accuracy of insurance question-answering.

[0188] By integrating all the initial answers, the target answers corresponding to the insurance questions are finally generated, which enables in-depth answers to insurance questions, improves the accuracy and efficiency of answering insurance questions, and the integrated target answers are clear and concise, which improves the user experience.

[0189] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

[0190] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0191] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An insurance question-and-answer method, characterized in that: The steps include: Receiving an insurance question input by a user, and extracting insurance question features of the insurance question; According to the characteristics of the insurance problem, a splitting strategy corresponding to the characteristics of the insurance problem is selected, and the insurance problem is split according to the splitting strategy to obtain multiple insurance sub-problems; Based on each of the insurance sub-problems, each of the insurance sub-problems is processed to generate an answer corresponding to each of the insurance sub-problems to obtain multiple initial answers; All the initial answers are integrated to generate a target answer, and the target answer is fed back to the user.

2. The insurance question-and-answer method according to claim 1, characterized in that: The extracting the insurance problem features of the insurance problem comprises: Performing part-of-speech tagging on the insurance question to obtain the insurance question after part-of-speech tagging; Using a pre-trained topic model to perform topic identification on the insurance questions after part-of-speech tagging, to obtain the topic of the insurance questions; extracting insurance problem keywords from the insurance problem according to the subject of the insurance problem; Feature extraction is performed based on the insurance issue keywords to obtain insurance issue features of the insurance issue.

3. The insurance question-and-answer method according to claim 1, characterized in that: After receiving the insurance question input by the user, the method further includes: Identify the type of data of the insurance issue; If the data type is identified as a non-text form, obtaining a text conversion method for the non-text form; According to the text conversion method, the insurance question is converted into text form.

4. The insurance question-and-answer method according to claim 1, wherein: According to the characteristics of the insurance problem, a splitting strategy corresponding to the characteristics of the insurance problem is selected, and the insurance problem is split according to the splitting strategy to obtain multiple insurance sub-problems, including: Processing the features of the insurance problem by a pre-trained classifier to obtain the problem type of the insurance problem; According to the pre-built strategy mapping table and the problem type, select the splitting strategy corresponding to the insurance problem; The insurance problem is split according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems.

5. The insurance question-and-answer method according to claim 4, characterized in that: The insurance problem is split according to the splitting strategy corresponding to the insurance problem to obtain multiple insurance sub-problems, including: Semantically matching the insurance problem feature with each node in the insurance knowledge graph to obtain matching nodes; Obtaining the hierarchical relationship of the matching node in the insurance knowledge graph; If the splitting strategy is a process splitting strategy, the insurance problem is split according to the horizontal hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems; If the splitting strategy is a conditional splitting strategy, the insurance problem is split according to the vertical hierarchical relationship of the hierarchical relationship to obtain a plurality of insurance sub-problems.

6. The insurance question-and-answer method according to claim 1, wherein: The step of processing each of the insurance sub-problems and generating an answer corresponding to each of the insurance sub-problems to obtain multiple initial answers includes: Based on each of the insurance sub-problems, converting them into vector representations through the input layer of the feature extraction model to obtain insurance sub-problem vectors; Extracting local features of the insurance sub-problem vector through the convolutional layer of the feature extraction model; Inputting the local features of the insurance sub-problem vector into the pooling layer of the feature extraction model to perform a dimensionality reduction operation to obtain a key vector of the insurance sub-problem; Performing similarity matching between the key vector and each entry in the insurance database to obtain a matching result; A search path is constructed according to the matching results, and the search path is executed to obtain an initial answer to the insurance sub-question.

7. The insurance question-and-answer method according to claim 6, characterized in that: After constructing a search path according to the matching results and executing the search path to obtain an initial answer to the insurance sub-question, the method further includes: identifying the answer number of the initial answer; When the number of answers is greater than one, obtaining basic information and behavior data of the user; Answers to questions that interest the user are selected from the initial answers based on the basic information and behavior data of the user.

8. An insurance question-and-answer device, characterized in that: The device comprises: An extraction module, used for receiving an insurance question input by a user and extracting insurance question features of the insurance question; A splitting module, configured to select a splitting strategy corresponding to the characteristics of the insurance problem according to the characteristics of the insurance problem, and split the insurance problem according to the splitting strategy to obtain a plurality of insurance sub-problems; A generating module, used for processing each of the insurance sub-problems based on each of the insurance sub-problems, generating an answer corresponding to each of the insurance sub-problems, and obtaining a plurality of initial answers; The integration module is used to integrate all the initial answers, generate a target answer, and feed back the target answer to the user.

9. A computer device, characterized in that: The computer device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the insurance question and answer method as described in any one of claims 1 to 7.

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

Images