Information reply method and device, computer equipment and storage medium

Abstract

The embodiment of the invention belongs to the field of financial science and technology, and relates to an information reply method and device, computer equipment and a storage medium, and the method comprises the following steps: obtaining insurance structure information and insurance non-structure information; constructing an insurance domain knowledge graph based on the insurance structure information and the insurance non-structure information; performing semantic analysis on the user dialogue information to obtain dialogue keywords and dialogue semantic features; reasoning in an insurance domain knowledge graph according to the dialogue keyword and the dialogue semantic feature to obtain a user intention category; calculating a confidence coefficient score of the user intention category, and judging whether the confidence coefficient score is smaller than a preset confidence coefficient threshold value or not; if yes, querying in the insurance field knowledge graph according to the user demand detailed information, and generating first reply content for replying; and if not, generating a second reply content for replying. According to the method, the dialogue content of the user can be effectively and accurately replied, so that the accuracy and effectiveness of reply are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, specifically to the field of financial technology, and in particular to an information reply method, device, computer equipment and storage medium. Background Art

[0002] In the traditional insurance business order recording process, cumbersome and complex processes and a large amount of data and rule verification have always been key factors restricting business efficiency and accuracy. Salesmen and underwriting personnel need to perform many operations manually, from data collection, information entry to rule verification, and each link consumes a lot of time and energy. In addition, in the face of complex and changeable order recording scenarios, manual operations are often difficult to deal with various emergencies, resulting in low order recording efficiency and difficulty in ensuring accuracy.

[0003] To solve this problem, the insurance industry has begun to explore intelligent order recording technology, aiming to simplify the order recording process and improve business efficiency through automated and intelligent means. However, in the insurance order recording scenario, the diversity and complexity of user intentions pose great challenges to the intelligent order recording system. Users may express multiple intentions at the same time in a conversation, such as consulting insurance terms, modifying policyholder information, and checking policy payment status, which requires the order recording system to have strong multi-intention recognition capabilities.

[0004] At the same time, multi-round dialogue management is also one of the key technical issues in intelligent insurance recording. During the multi-round dialogue process, the system needs to accurately understand the user's intentions and reasonably arrange the order and content of the dialogue according to the priority and relevance of the intentions. In addition, the transmission and association of contextual information is also the key to achieving dialogue coherence. For example, when the user modifies the policyholder information mentioned in the previous round of dialogue in the next round, the system should be able to automatically associate and fill in the relevant information to avoid repeated inquiries, improve the effectiveness and rationality of the dialogue response, and save time in the recording process. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide an information reply method, apparatus, computer device and storage medium to solve the problem of being unable to effectively and accurately reply to the content of user conversations.

[0006] In order to solve the above technical problems, the embodiment of the present application provides an information reply method, which adopts the following technical solution:

[0007] Obtain insurance structural information and insurance non-structural information;

[0008] Constructing an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information;

[0009] Acquire user conversation information, perform semantic analysis on the user conversation information, and obtain conversation keywords and conversation semantic features;

[0010] Inferring the insurance field knowledge graph based on the conversation keywords and the conversation semantic features, obtaining a user intent category;

[0011] Calculating a confidence score for the user intent category, and determining whether the confidence score is less than a preset confidence threshold;

[0012] If the confidence score is less than the preset confidence threshold, obtaining the user demand refinement information, and querying the insurance field knowledge graph according to the user demand refinement information to generate the first reply content for reply;

[0013] If the confidence score is greater than or equal to the preset confidence threshold, a search is performed in the insurance field knowledge graph according to the user intention category, the corresponding knowledge information is extracted, and the knowledge information is filled into the preset reply template to generate a second reply content for reply.

[0014] Furthermore, the step of obtaining insurance structural information and insurance non-structural information specifically includes:

[0015] Acquire a structural information extraction identifier, and extract insurance product information, underwriting rule information, and underwriting decision information from a database according to the knowledge information extraction identifier;

[0016] Acquire a non-structure information extraction identifier, and extract insurance clause information and claim case information from a database according to the non-structure information extraction identifier;

[0017] Performing a first preprocessing on the insurance product information, the underwriting rule information, and the underwriting decision information to obtain the insurance structure information;

[0018] The extracted insurance clause information and the claim case information are subjected to a second preprocessing to obtain the insurance non-structure information.

[0019] Furthermore, the step of constructing an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information specifically includes:

[0020] Define insurance business entities, insurance business attributes, and insurance business entity relationships, and construct insurance domain knowledge ontology;

[0021] Integrate the insurance structure information into the insurance domain knowledge ontology to obtain the insurance domain knowledge graph framework;

[0022] Key information is extracted from the insurance non-structured information to obtain non-structured key information, and the non-structured key information is integrated into the insurance field knowledge graph framework to obtain the insurance field knowledge graph.

[0023] Furthermore, the step of obtaining user conversation information, performing semantic analysis on the user conversation information, and obtaining conversation keywords and conversation semantic features specifically includes:

[0024] Accessing user conversation records, and extracting the user conversation information from the user conversation records;

[0025] Performing word segmentation and part-of-speech tagging on the user dialogue information to obtain dialogue representation information;

[0026] performing named entity recognition on the conversation representation information, and extracting the conversation keywords from the conversation representation information;

[0027] The conversation keywords are represented by word vectors to obtain keyword semantic information, and the keyword semantic information is used as the conversation semantic feature.

[0028] Furthermore, the step of performing reasoning in the insurance field knowledge graph based on the conversation keywords and the conversation semantic features to obtain the user intention category specifically includes:

[0029] Inputting the conversation keywords and the conversation semantic features into the insurance field knowledge graph, and performing reasoning on the insurance field knowledge graph based on a knowledge reasoning algorithm to obtain a knowledge reasoning result;

[0030] The knowledge reasoning result is input into a pre-built intention recognition model to obtain the user intention category.

[0031] Furthermore, the step of calculating the confidence score of the user intent category and determining whether the confidence score is less than a preset confidence threshold specifically includes:

[0032] Inputting the user conversation information and the user intention category into a pre-built confidence prediction model to obtain the confidence score;

[0033] Matching the confidence scores with the user intent categories to obtain an intent confidence score set;

[0034] Extracting a corresponding preset confidence threshold from a database according to the user intention category;

[0035] The intention confidence score set is sorted from high to low according to the confidence score, and the confidence score ranked first is extracted and compared with the preset confidence threshold.

[0036] Furthermore, the step of obtaining the detailed information of user needs, and querying the insurance field knowledge graph according to the detailed information of user needs, and obtaining the first reply content for replying specifically includes:

[0037] Determine a corresponding information database according to the user intention category, and extract corresponding initial reply information from the information database according to the user intention category;

[0038] Returning the initial reply information to the user interface for reply, and obtaining detailed information on user needs returned by the user;

[0039] Search the insurance field knowledge graph according to the user demand refinement information to obtain demand entity nodes and demand relationship edges;

[0040] Acquire node attribute information of related entity nodes according to the demand entity node and the demand relationship edge, and determine key attribute information in the node attribute information of the related entity nodes according to the user demand refinement information;

[0041] Matching a corresponding reply template in a preset reply template library according to the key attribute information and the user demand refinement information;

[0042] The key attribute information is filled into the reply template as a parameter to generate the first reply content, and the first reply content is sent to the user interface for reply.

[0043] In order to solve the above technical problems, the embodiment of the present application further provides an information reply device, which adopts the following technical solution:

[0044] An information acquisition module, used to acquire insurance structural information and insurance non-structural information;

[0045] A graph construction module, used to construct an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information;

[0046] An information analysis module is used to obtain user conversation information, perform semantic analysis on the user conversation information, and obtain conversation keywords and conversation semantic features;

[0047] An information reasoning module, used to perform reasoning in the insurance field knowledge graph according to the conversation keywords and the conversation semantic features to obtain a user intention category;

[0048] A score comparison module, used to calculate the confidence score of the user intention category and determine whether the confidence score is less than a preset confidence threshold;

[0049] A first processing module, configured to obtain user demand refinement information if the confidence score is less than the preset confidence threshold, and query the insurance field knowledge graph according to the user demand refinement information to generate a first reply content for reply;

[0050] The second processing module is used to search the insurance field knowledge graph according to the user intention category if the confidence score is greater than or equal to the preset confidence threshold, extract the corresponding knowledge information, fill the knowledge information into the preset reply template, and generate a second reply content for reply.

[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0052] A computer device comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the information reply method as described in any one of the above items when executing the computer-readable instructions.

[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0054] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of the information reply method as described in any one of the above items.

[0055] Compared with the prior art, the embodiment of the present application has the following beneficial effects: the embodiment obtains insurance structural information and insurance non-structural information; constructs an insurance field knowledge graph based on the insurance structural information and the insurance non-structural information; obtains user dialogue information, performs semantic analysis on the user dialogue information, obtains dialogue keywords and dialogue semantic features; performs reasoning in the insurance field knowledge graph according to the dialogue keywords and the dialogue semantic features, obtains user intent categories; calculates the confidence score of the user intent category, and determines whether the confidence score is less than a preset confidence threshold; if the confidence score is less than the preset confidence threshold, obtains user demand refinement information, and performs a query in the insurance field knowledge graph according to the user demand refinement information to generate a first reply content for reply; if the confidence score is greater than or equal to the preset confidence threshold, searches in the insurance field knowledge graph according to the user intent category, extracts the corresponding knowledge information, and fills the knowledge information into a preset reply template to generate a second reply content for reply. Thus, an accurate reply is effectively made to the user dialogue content to improve the accuracy and effectiveness of the reply. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] 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.

[0057] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0058] Figure 2 A flowchart of an embodiment of the information reply method according to the present application;

[0059] Figure 3 yes Figure 2 A flowchart of a specific implementation of step S20;

[0060] Figure 4 yes Figure 2 A flowchart of a specific implementation of step S30;

[0061] Figure 5 yes Figure 2 A flowchart of a specific implementation of step S40;

[0062] Figure 6 yes Figure 2 A flowchart of a specific implementation of step S50;

[0063] Figure 7 yes Figure 2 A flowchart of a specific implementation of step S60;

[0064] Figure 8 is a structural schematic diagram of an embodiment of an information reply device according to the present application;

[0065] Fig. 9 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0066] 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.

[0067] 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 does it refer to non-related or alternative embodiments that are 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.

[0068] 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.

[0069] like Figure 1 As 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.

[0070] 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.

[0071] 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.

[0072] 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 .

[0073] It should be noted that the information reply method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the information reply device is generally set in the server / terminal device.

[0074] 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.

[0075] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for information reply according to the present application. The information reply method comprises the following steps:

[0076] Step S10, obtaining insurance structural information and insurance non-structural information;

[0077] In this embodiment, insurance structural information refers to structured insurance knowledge information, which includes insurance product information, underwriting rule information, and underwriting decision information. Among them, insurance product information includes the name, type, coverage, insurance period, premium amount, etc. of the insurance product, underwriting rule information includes personal factors such as the age, health status, occupation, and living habits of the insured, as well as the nature, value, and risk level of the subject matter of insurance, and underwriting decision information includes underwriting conditions (such as premium adjustment, coverage limitation, etc.) and whether further investigation or physical examination is required. Insurance unstructured information refers to unstructured insurance-related texts, which includes insurance clause information and claim case information. Insurance clause information refers to the core content of the insurance contract, which stipulates the rights and obligations between the insurance company and the insured. Claim case information is the actual compensation case handled by the insurance company, which usually includes detailed information such as accident description, claim process, and compensation amount. Claim case information is usually recorded in text form and contains multimedia materials such as pictures and videos.

[0078] Step S20, constructing an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information;

[0079] In this embodiment, the insurance domain knowledge graph is a graph-based data structure that uses nodes (representing entities) and edges (representing relationships) to display various types of information in the insurance industry. These entities can include insurance products, insurance companies, customers, risk factors, etc., while relationships describe the interactions and connections between these entities. By constructing the insurance domain knowledge graph, it is convenient for users to quickly query relevant knowledge information and provide reliable and effective information responses.

[0080] Step S30, obtaining user conversation information, performing semantic analysis on the user conversation information, and obtaining conversation keywords and conversation semantic features;

[0081] In this embodiment, user dialogue information refers to the communication dialogue record of the user in the system question-answering robot, and the user dialogue information includes user input information (text input, voice input, etc.), system reply information (text reply information, voice reply information, multimedia reply information, etc.), and dialogue metadata (dialogue time, dialogue turn, session ID, etc.). By semantically parsing the user dialogue information, dialogue keywords and dialogue semantic features that reflect the information acquisition intention in the user dialogue can be accurately obtained.

[0082] Step S40, inferring in the insurance field knowledge graph according to the conversation keywords and the conversation semantic features to obtain the user intention category;

[0083] In this embodiment, the conversation keywords are insurance-related entities involved in the user conversation information, such as insurance products, customers, insurance companies, etc. The conversation semantic features are the keyword semantic information obtained after the conversation keywords are represented by word vectors. The word vector representation is a method of converting conversation keywords into word vectors to capture the semantic similarity and correlation between words. The user intent category refers to the category corresponding to the information acquisition intention revealed in the user conversation, which may include purchase consultation, claim application, product comparison or complaint, etc.

[0084] Step S50, calculating the confidence score of the user intention category, and determining whether the confidence score is less than a preset confidence threshold;

[0085] In this embodiment, the confidence score refers to the credibility corresponding to the judgment result of the user intention category. By calculating the confidence score of the user intention category, the reliability of the user intention category is judged to perform corresponding information generation reply processing.

[0086] Step S60: if the confidence score is less than the preset confidence threshold, obtaining user demand refinement information, and querying the insurance field knowledge graph according to the user demand refinement information to generate a first reply content for reply;

[0087] In this embodiment, the user demand refinement information is the reply information collected from the user after the initial reply information is generated. The initial reply information may include template information to guide the user to refine the demand information, such as "Please refine the current demand to obtain more accurate reply information". The first reply content is the system reply information generated by combining the key attribute information obtained by specific retrieval in the insurance field knowledge graph based on the user demand refinement information and the corresponding reply template. After the system generates the first reply content, it returns the first reply content to the user through the corresponding path of the user interface for display reply.

[0088] Step S70: If the confidence score is greater than or equal to the preset confidence threshold, a search is performed in the insurance field knowledge graph according to the user intention category, the corresponding knowledge information is extracted, and the knowledge information is filled into the preset reply template to generate a second reply content for reply.

[0089] In this embodiment, a search is performed in the insurance field knowledge graph based on the user intent category to match the corresponding entities and relationships, thereby obtaining the knowledge information corresponding to the entities and relationships, and filling this part of the knowledge information into the preset reply template to generate a reply text message that is returned to the user interface for reply.

[0090] This embodiment obtains insurance structural information and insurance non-structural information; constructs an insurance field knowledge graph based on the insurance structural information and the insurance non-structural information; obtains user dialogue information, performs semantic analysis on the user dialogue information, obtains dialogue keywords and dialogue semantic features; performs reasoning in the insurance field knowledge graph based on the dialogue keywords and the dialogue semantic features to obtain user intent categories; calculates the confidence score of the user intent category, and determines whether the confidence score is less than a preset confidence threshold; if the confidence score is less than the preset confidence threshold, obtains user demand refinement information, and performs a query in the insurance field knowledge graph based on the user demand refinement information to generate a first reply content for reply; if the confidence score is greater than or equal to the preset confidence threshold, searches the insurance field knowledge graph based on the user intent category, extracts the corresponding knowledge information, and fills the knowledge information into a preset reply template to generate a second reply content for reply. In this way, accurate replies are effectively made to the user dialogue content to improve the accuracy and effectiveness of the reply.

[0091] In some optional implementations of this embodiment, the obtaining of insurance structural information and insurance non-structural information includes the following steps:

[0092] Acquire a structural information extraction identifier, and extract insurance product information, underwriting rule information, and underwriting decision information from a database according to the knowledge information extraction identifier;

[0093] In this embodiment, the structural information extraction identifier is identification information corresponding to the insurance product information, underwriting rule information, and underwriting decision information. The structural information extraction identifier is used to perform a matching query in the database to effectively obtain the insurance product information, underwriting rule information, and underwriting decision information.

[0094] Acquire a non-structure information extraction identifier, and extract insurance clause information and claim case information from a database according to the non-structure information extraction identifier;

[0095] In this embodiment, the non-structure information extraction identifier is identification information corresponding to the insurance clause information and the claim case information. The non-structure information extraction identifier is used to perform a matching query in the database to effectively obtain the insurance clause information and the claim case information.

[0096] Performing a first preprocessing on the insurance product information, the underwriting rule information, and the underwriting decision information to obtain the insurance structure information;

[0097] In this embodiment, the first preprocessing includes data cleaning (removing redundant information, correcting erroneous data), data standardization and structuring (defining data structure, data standardization), information extraction and induction (extracting key information, summarizing information), etc. The first preprocessing is performed on insurance product information, underwriting rule information, and underwriting decision information to obtain standard and effective insurance structure information.

[0098] The extracted insurance clause information and the claim case information are subjected to a second preprocessing to obtain the insurance non-structure information.

[0099] In this embodiment, the second preprocessing includes data cleaning (removing irrelevant information, standardizing formats, processing special characters), text segmentation and standards, information extraction (entity recognition, relationship extraction, event extraction), etc. The second preprocessing is performed on the insurance clause information and claim case information to obtain standard and effective insurance non-structured information.

[0100] This embodiment obtains a structural information extraction identifier, and extracts insurance product information, underwriting rule information, and underwriting decision information from the database according to the knowledge information extraction identifier; obtains a non-structural information extraction identifier, and extracts insurance clause information and claim case information from the database according to the non-structural information extraction identifier; performs a first preprocessing on the insurance product information, the underwriting rule information, and the underwriting decision information to obtain the insurance structural information; and performs a second preprocessing on the extracted insurance clause information and the claim case information to obtain the insurance non-structural information. In this way, structured insurance structural information and unstructured insurance non-structural information are effectively and accurately obtained, providing effective and reliable data support for the subsequent construction of the insurance field knowledge graph.

[0101] refer to Figure 3 In some optional implementations of this embodiment, step S20 includes the following steps:

[0102] Step S201, defining insurance business entities, insurance business attributes, and insurance business entity relationships, and constructing insurance domain knowledge ontology;

[0103] In this embodiment, an ontology modeling method is adopted to define the core business entities in the insurance field, wherein the core business entities include insurance policies, customers and claims. At the same time, the key attributes of the core business entities are defined. By analyzing the relationships between insurance business entities, an entity relationship model is constructed. The relationship includes the relationship between insurance policies and customers and the corresponding relationship between insurance policies and claims. The defined insurance business entities, key attributes and the relationships are integrated to form a knowledge ontology of the insurance field.

[0104] Step S202, integrating the insurance structure information into the insurance domain knowledge ontology to obtain an insurance domain knowledge graph framework;

[0105] In this embodiment, the key information elements of the insurance structure information are extracted by the knowledge extraction algorithm, and the extracted key information elements are extracted and mapped to the insurance field knowledge ontology for fusion to obtain the insurance field knowledge graph framework. Among them, the knowledge extraction algorithm is a technology that can automatically identify and extract key information from structured or semi-structured data. The insurance structure information is parsed and extracted by the knowledge extraction algorithm. This step includes identifying and extracting key information elements related to insurance business entities, attributes and relationships, such as the name of the insured, the identity of the insured, the type of insurance product, the insurance amount, the premium, the underwriting conditions, the claim results, etc. According to the structure of the insurance field knowledge ontology, the mapping rules between key information elements and entities, attributes and relationships in the knowledge ontology are defined. The extracted key information elements are mapped according to the defined mapping rules. This step includes matching the information elements with entities in the knowledge ontology, filling the attribute values ​​into the corresponding attribute fields, and establishing relationship links between entities. On the basis of the mapping, the extracted key information elements are integrated with the existing information in the knowledge ontology. This step includes dealing with issues such as information redundancy, conflict and missing information to ensure the accuracy and completeness of the knowledge graph.

[0106] Step S203: extract key information from the insurance non-structured information to obtain non-structured key information, and integrate the non-structured key information into the insurance field knowledge graph framework to obtain the insurance field knowledge graph.

[0107] In this embodiment, for the text data of insurance clauses, a rule-based and pattern matching method is used to extract the underwriting rules in the insurance clauses, wherein the underwriting rules include the insured object, insurance liability and exclusions; for the text data of claims cases, a knowledge extraction method based on deep learning is used to extract the claims decision information in the claims cases, wherein the claims decision information includes the amount of compensation and the conditions of compensation; the extracted underwriting rules and claims decision information are structured, and the structured underwriting rules and claims decision information are fused with the pre-built insurance field knowledge graph framework to obtain the fused insurance field knowledge graph, wherein the fusion includes entity alignment and relationship alignment. In the process of knowledge fusion, a knowledge conflict detection method can be used to check the consistency of the newly added knowledge with the existing knowledge. If a conflict is detected, it is reviewed and corrected to ensure that the final insurance field knowledge graph is valid and correct.

[0108] This embodiment constructs an insurance field knowledge ontology by defining insurance business entities, insurance business attributes, and insurance business entity relationships; integrates the insurance structural information into the insurance field knowledge ontology to obtain an insurance field knowledge graph framework; extracts key information from the insurance non-structural information to obtain non-structural key information, and integrates the non-structural key information into the insurance field knowledge graph framework, thereby obtaining an insurance field knowledge graph that effectively integrates insurance structural information and insurance non-structural information to facilitate subsequent knowledge reasoning processing.

[0109] refer to Figure 4 In some optional implementations of this embodiment, step S30 includes the following steps:

[0110] Step S301, accessing the user session record, and extracting the user conversation information from the user session record;

[0111] In this embodiment, the user conversation record is a historical conversation record between the user and the system question-answering robot in the system. The user conversation information required for the current question-answering is extracted from the user conversation record through the corresponding timestamp to ensure the validity of the user conversation information. For example, if the user starts to ask questions on the system at 3:10, the historical conversation records after 3:10 are extracted according to the timestamp to obtain the user conversation information.

[0112] Step S302, performing word segmentation and part-of-speech tagging on the user dialogue information to obtain dialogue representation information;

[0113] In this embodiment, word segmentation refers to the process of cutting a continuous text string into independent words or phrases, and the word segmentation operation of the user dialogue information can be performed through the corresponding word segmentation tool. Part-of-speech tagging refers to assigning a part-of-speech tag to each word on the basis of word segmentation, such as noun, verb, adjective, etc., and the part-of-speech standard can also be performed through the word segmentation tool.

[0114] Step S303, performing named entity recognition on the conversation representation information, and extracting the conversation keywords from the conversation representation information;

[0115] In this embodiment, named entity recognition of dialogue representation information can be implemented through a pre-trained NER model (such as BERT), by inputting the dialogue representation information into the pre-trained NER model to output the corresponding NER model results, and then filtering out dialogue keywords in the dialogue representation information based on the NER model results.

[0116] Step S304, performing word vector representation on the conversation keywords to obtain keyword semantic information, and using the keyword semantic information as the conversation semantic feature.

[0117] In this embodiment, a pre-trained word vector model (a BERT model may be used) is used to represent the conversation keywords as word vectors. The conversation keywords are input into the pre-trained word vector model to obtain the corresponding word vectors. The cosine similarity of the word vectors is calculated, cluster analysis, and dimensionality reduction projection analysis are performed to obtain the semantic relationship and potential semantic categories between the keywords, thereby obtaining the keyword semantic information.

[0118] This embodiment accesses the user conversation record, extracts the user conversation information from the user conversation record; performs word segmentation and part-of-speech tagging on the user conversation information to obtain conversation representation information; performs named entity recognition on the conversation representation information to extract the conversation keywords from the conversation representation information; performs word vector representation on the conversation keywords to obtain keyword semantic information, and uses the keyword semantic information as the conversation semantic feature. This enables accurate acquisition of conversation keywords and conversation semantic features in the user communication process, so as to facilitate subsequent reasoning operations based on user intent categories.

[0119] Continue to refer Figure 5 In some optional implementations of this embodiment, step S40 includes the following steps:

[0120] Step S401, inputting the conversation keywords and the conversation semantic features into the insurance field knowledge graph, and performing reasoning on the insurance field knowledge graph based on a knowledge reasoning algorithm to obtain a knowledge reasoning result;

[0121] In this embodiment, by taking conversation keywords and conversation semantic features as input, the input is mapped to the corresponding nodes and relationships of the insurance field knowledge graph, where nodes represent entities in the insurance field, such as insurance products, customers, risks, etc.; relationships describe the associations between these entities, such as "belongs to", "contains", "related to", etc., and a rule-based reasoning algorithm is applied to reason on nodes and relationships to obtain knowledge reasoning results.

[0122] Step S402: input the knowledge reasoning result into a pre-built intention recognition model to obtain the user intention category.

[0123] In this embodiment, the user intention category refers to the user's intention to obtain information during the conversation, and the user intention category may include purchase consultation, claim application, product comparison or complaint, etc. The pre-built intention recognition model can adopt a decision tree model, and the decision tree model is trained and optimized by obtaining labeled sample data, so that the decision tree model can accurately identify the corresponding user intention category according to the input data.

[0124] In this embodiment, the conversation keywords and the conversation semantic features are input into the insurance field knowledge graph, and the knowledge reasoning algorithm is used to perform reasoning on the insurance field knowledge graph to obtain the knowledge reasoning result; the knowledge reasoning result is input into the pre-built intention recognition model to obtain the user intention category. This effectively realizes accurate user conversation intention recognition, so as to facilitate the subsequent confidence judgment operation based on the recognized user intention category.

[0125] Continue to refer Figure 6 In some optional implementations of this embodiment, step S50 includes the following steps:

[0126] Step S501, inputting the user conversation information and the user intention category into a pre-built confidence prediction model to obtain the confidence score;

[0127] In this embodiment, the pre-built confidence prediction model can adopt a support vector machine model, and the support vector machine model is trained by using a labeled data set (including user dialogue information and corresponding intent categories and confidence labels) to obtain a confidence prediction model that can accurately output confidence score results. Then, by extracting corresponding features from the user dialogue information and user intent categories (using word embedding, such as Word2Vec, BERT, etc.), the features are input into the confidence prediction model to obtain the corresponding confidence score. In this embodiment, the confidence score can be an integer or a percentage value in a percentage system.

[0128] Step S502, matching the confidence scores with the user intent categories to obtain an intent confidence score set;

[0129] In this embodiment, each user intent category corresponds to a confidence score. For user dialogue information, multiple user intent categories may be included, so that the output reply content is more in line with the user's personality and needs. By correspondingly associating the user intent categories and confidence scores, an intent confidence score set including the categories and their corresponding confidence scores is obtained.

[0130] Step S503, extracting a corresponding preset confidence threshold from a database according to the user intention category;

[0131] In this embodiment, each user intent category corresponds to a different preset confidence threshold. For example, if the user intent category is product comparison, the preset confidence threshold is 70; if the user intent category is claim application, the preset confidence threshold is 80. This is because different user intent categories correspond to different conversation words and emotional values. In the specific implementation process, the preset confidence threshold can be set and adjusted according to actual conditions.

[0132] Step S504: sort the intention confidence score set from high to low according to the confidence score, and extract the confidence score ranked first for comparison with the preset confidence threshold.

[0133] In this embodiment, by sorting the intent confidence score set from high to low, the confidence score ranked first can be directly extracted, and the user intent category corresponding to the confidence score is used as the main user intent for reply judgment. In the specific implementation process, multiple confidence scores can also be taken in the order of the sorted intent confidence score set for comparison, and different reply contents are generated according to the comparison results, and the different reply contents are merged according to certain rules to obtain diversified reply information for user reply.

[0134] This embodiment obtains the confidence score by inputting the user conversation information and the user intention category into a pre-built confidence prediction model; matches the confidence score with the user intention category to obtain an intention confidence score set; extracts the corresponding preset confidence threshold from the database according to the user intention category; sorts the intention confidence score set from high to low according to the confidence score, and extracts the first ranked confidence score and compares it with the preset confidence threshold. This effectively realizes the accuracy judgment of the user intention category, so as to facilitate the subsequent corresponding processing operation according to the judgment result.

[0135] Continue to refer Figure 7 In some optional implementations of this embodiment, step S60 includes the following steps:

[0136] Step S601, determining a corresponding information database according to the user intention category, and extracting corresponding initial reply information from the information database according to the user intention category;

[0137] In this embodiment, the user intent category is used as a query condition to match the corresponding information database, and then the category keyword corresponding to the user intent category is used to traverse the query in the information database to extract the corresponding initial reply information. The initial reply information is a standard template reply information pre-set according to the user intent category, and the content can be modified according to the actual situation.

[0138] Step S602, returning the initial reply information to the user interface for reply, and obtaining the user demand refinement information returned by the user;

[0139] In this embodiment, the initial reply information may include content to guide the user to refine the demand, such as "Please refine the current demand to obtain more accurate reply information". By returning the initial reply information to the user interface for reply, the user is effectively guided to propose more detailed user demand refinement information to facilitate subsequent processing.

[0140] Step S603, searching the insurance field knowledge graph according to the user demand refinement information to obtain demand entity nodes and demand relationship edges;

[0141] In this embodiment, a corresponding search query is constructed based on user demand refinement information. The search query can be a combination of one or more keywords or a more complex query statement, which is used to locate relevant entity nodes and relationship edges in the knowledge graph. The constructed search query is then input into the knowledge graph system in the insurance field to retrieve and return relevant entity nodes and relationship edges, thereby effectively obtaining entity nodes and demand relationship edges.

[0142] Step S604, acquiring node attribute information of related entity nodes according to the demand entity node and the demand relationship edge, and determining key attribute information in the node attribute information of the related entity nodes according to the user demand refinement information;

[0143] In this embodiment, starting from the demand entity node, traversing along the relationship edge in the knowledge graph to other related entity nodes, and then extracting the node attribute information for each traversed related entity node. For example, the demand entity node represents insurance products, customers, risk events, etc., the demand relationship edge represents the purchase relationship between the product and the customer, the association between the risk event and the insurance product, etc., and the node attribute information includes the premium, coverage, claim conditions, etc. of the insurance product, the age, occupation, health status, etc. of the customer, and the type and probability of occurrence of the risk event. By analyzing the description of the user demand refinement information, including the detailed information that has an important impact on the decision, the detailed information involves the user's specific needs for insurance products (such as coverage, premium budget, etc.), the degree of concern for risk events (such as risk events with high probability of occurrence), and the description of the customer's own situation (such as age, health status, etc.). According to the user demand refinement information, the key attribute information directly related to the user's needs is screened out from the extracted node attribute information. In this embodiment, the screened key attribute information can be sorted so that the most important information can be found more easily.

[0144] Step S605, matching a corresponding reply template in a preset reply template library according to the key attribute information and the user demand refinement information;

[0145] In this embodiment, similarity matching is performed in the preset response template library through the information content of key attribute information and user demand refinement information, and a text similarity algorithm (such as cosine similarity, Jaccard similarity, edit distance, etc.) can be used to calculate the similarity between the extracted information and the preset response template, so as to find the corresponding response template.

[0146] Step S606: fill the key attribute information into the reply template as a parameter, generate the first reply content, and send the first reply content to the user interface for reply.

[0147] In this embodiment, after the key attribute information is filled into the reply template, the generated reply information can be checked. The checking steps include data format check, internal logic check, external logic check, language quality check, etc. When the check is completed, the generated reply information is sent as the first reply content to the user interface for reply.

[0148] This embodiment determines the corresponding information database according to the user intention category, and extracts the corresponding initial reply information from the information database according to the user intention category; returns the initial reply information to the user interface for reply, and obtains the user demand refinement information returned by the user; searches the insurance field knowledge graph according to the user demand refinement information to obtain the demand entity node and demand relationship edge; obtains the node attribute information of the relevant entity node according to the demand entity node and the demand relationship edge, and determines the key attribute information in the node attribute information of the relevant entity node according to the user demand refinement information; matches the corresponding reply template in the preset reply template library according to the key attribute information and the user demand refinement information; fills the key attribute information into the reply template as a parameter, generates the first reply content, and sends the first reply content to the user interface for reply. This effectively realizes the regeneration of the first reply content that better meets the user's intention and needs for user reply, so as to improve the accuracy of replying user information.

[0149] 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).

[0150] 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.

[0151] Further references Figure 8 , as a response to the above Figure 1 The present application provides an embodiment of an information reply device, and the device embodiment is similar to Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0152] like Figure 8 As shown, the information reply device 800 described in this embodiment includes: an information acquisition module 801, a graph construction module 802, an information analysis module 803, an information reasoning module 804, a score comparison module 805, a first processing module 806, and a second processing module 807. Among them:

[0153] The information acquisition module 801 is used to acquire insurance structural information and insurance non-structural information;

[0154] A graph construction module 802, used to construct an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information;

[0155] The information analysis module 803 is used to obtain user conversation information, perform semantic analysis on the user conversation information, and obtain conversation keywords and conversation semantic features;

[0156] An information reasoning module 804 is used to perform reasoning in the insurance field knowledge graph according to the conversation keywords and the conversation semantic features to obtain a user intention category;

[0157] The score comparison module 805 is used to calculate the confidence score of the user intention category and determine whether the confidence score is less than a preset confidence threshold;

[0158] The first processing module 806 is used to obtain the user demand refinement information if the confidence score is less than the preset confidence threshold, and query the insurance field knowledge graph according to the user demand refinement information to generate a first reply content for reply;

[0159] The second processing module 807 is used to search the insurance field knowledge graph according to the user intention category if the confidence score is greater than or equal to the preset confidence threshold, extract the corresponding knowledge information, fill the knowledge information into the preset reply template, and generate a second reply content for reply.

[0160] By adopting the above-mentioned information reply device, this embodiment can obtain insurance structural information and insurance non-structural information; construct an insurance field knowledge graph based on the insurance structural information and the insurance non-structural information; obtain user dialogue information, perform semantic analysis on the user dialogue information, and obtain dialogue keywords and dialogue semantic features; perform reasoning in the insurance field knowledge graph based on the dialogue keywords and the dialogue semantic features to obtain user intention categories; calculate the confidence score of the user intention category, and determine whether the confidence score is less than a preset confidence threshold; if the confidence score is less than the preset confidence threshold, obtain user demand refinement information, and query the insurance field knowledge graph based on the user demand refinement information to generate a first reply content for reply; if the confidence score is greater than or equal to the preset confidence threshold, search the insurance field knowledge graph based on the user intention category, extract the corresponding knowledge information, and fill the knowledge information into the preset reply template to generate a second reply content for reply. In this way, accurate replies are effectively made to the user dialogue content to improve the accuracy and effectiveness of the reply.

[0161] To solve the above technical problems, the present application also provides a computer device. Fig. 9 , Fig. 9 This is a basic structural block diagram of the computer device in this embodiment.

[0162] The computer device 9 includes a memory 91, a processor 92, and a network interface 93 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 9 with components 91-93, 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.

[0163] 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.

[0164] The memory 91 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 91 can be an internal storage unit of the computer device 9, such as a hard disk or memory of the computer device 9. In other embodiments, the memory 91 can also be an external storage device of the computer device 9, such as a plug-in hard disk equipped on the computer device 9, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 91 can also include both the internal storage unit of the computer device 9 and its external storage device. In this embodiment, the memory 91 is generally used to store the operating system and various application software installed on the computer device 9, such as computer-readable instructions of the information reply method, etc. In addition, the memory 91 can also be used to temporarily store various types of data that have been output or are to be output.

[0165] The processor 92 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 92 is generally used to control the overall operation of the computer device 9. In this embodiment, the processor 92 is used to run computer-readable instructions stored in the memory 91 or process data, such as computer-readable instructions for running the information reply method.

[0166] The network interface 93 may include a wireless network interface or a wired network interface. The network interface 93 is generally used to establish a communication connection between the computer device 9 and other electronic devices.

[0167] By adopting the above-mentioned computer device, this embodiment can obtain insurance structural information and insurance non-structural information; construct an insurance field knowledge graph based on the insurance structural information and the insurance non-structural information; obtain user dialogue information, perform semantic analysis on the user dialogue information, and obtain dialogue keywords and dialogue semantic features; perform reasoning in the insurance field knowledge graph based on the dialogue keywords and the dialogue semantic features to obtain user intention categories; calculate the confidence score of the user intention category, and determine whether the confidence score is less than a preset confidence threshold; if the confidence score is less than the preset confidence threshold, obtain user demand refinement information, and query the insurance field knowledge graph based on the user demand refinement information to generate a first reply content for reply; if the confidence score is greater than or equal to the preset confidence threshold, search the insurance field knowledge graph based on the user intention category, extract the corresponding knowledge information, and fill the knowledge information into the preset reply template to generate a second reply content for reply. In this way, accurate replies are effectively made to the user dialogue content to improve the accuracy and effectiveness of the reply.

[0168] The present application also provides another embodiment, namely, providing a 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 information reply method as described above.

[0169] By adopting the above-mentioned computer-readable storage medium, this embodiment can obtain insurance structural information and insurance non-structural information; construct an insurance field knowledge graph based on the insurance structural information and the insurance non-structural information; obtain user dialogue information, perform semantic analysis on the user dialogue information, and obtain dialogue keywords and dialogue semantic features; perform reasoning in the insurance field knowledge graph according to the dialogue keywords and the dialogue semantic features to obtain user intention categories; calculate the confidence score of the user intention category, and determine whether the confidence score is less than a preset confidence threshold; if the confidence score is less than the preset confidence threshold, obtain user demand refinement information, and query the insurance field knowledge graph according to the user demand refinement information to generate a first reply content for reply; if the confidence score is greater than or equal to the preset confidence threshold, search the insurance field knowledge graph according to the user intention category, extract the corresponding knowledge information, and fill the knowledge information into the preset reply template to generate a second reply content for reply. In this way, accurate replies are effectively made to the user dialogue content to improve the accuracy and effectiveness of the reply.

[0170] 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.

[0171] 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. A method for replying information, characterized in that: The steps include: Obtain insurance structural information and insurance non-structural information; Constructing an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information; Acquire user conversation information, perform semantic analysis on the user conversation information, and obtain conversation keywords and conversation semantic features; Inferring the insurance field knowledge graph based on the conversation keywords and the conversation semantic features, obtaining a user intent category; Calculating a confidence score for the user intent category, and determining whether the confidence score is less than a preset confidence threshold; If the confidence score is less than the preset confidence threshold, obtaining the user demand refinement information, and querying the insurance field knowledge graph according to the user demand refinement information to generate the first reply content for reply; If the confidence score is greater than or equal to the preset confidence threshold, a search is performed in the insurance field knowledge graph according to the user intention category, the corresponding knowledge information is extracted, and the knowledge information is filled into the preset reply template to generate a second reply content for reply.

2. The information reply method according to claim 1, characterized in that: The step of obtaining insurance structural information and insurance non-structural information specifically includes: Acquire a structural information extraction identifier, and extract insurance product information, underwriting rule information, and underwriting decision information from a database according to the knowledge information extraction identifier; Acquire a non-structure information extraction identifier, and extract insurance clause information and claim case information from a database according to the non-structure information extraction identifier; Performing a first preprocessing on the insurance product information, the underwriting rule information, and the underwriting decision information to obtain the insurance structure information; The extracted insurance clause information and the claim case information are subjected to a second preprocessing to obtain the insurance non-structure information.

3. The information reply method according to claim 1, characterized in that: The step of constructing an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information specifically includes: Define insurance business entities, insurance business attributes, and insurance business entity relationships, and construct insurance domain knowledge ontology; Integrate the insurance structure information into the insurance domain knowledge ontology to obtain the insurance domain knowledge graph framework; Key information is extracted from the insurance non-structured information to obtain non-structured key information, and the non-structured key information is integrated into the insurance field knowledge graph framework to obtain the insurance field knowledge graph.

4. The information reply method according to claim 1, characterized in that: The step of acquiring user conversation information, performing semantic analysis on the user conversation information, and obtaining conversation keywords and conversation semantic features specifically includes: Accessing user conversation records, and extracting the user conversation information from the user conversation records; Performing word segmentation and part-of-speech tagging on the user dialogue information to obtain dialogue representation information; performing named entity recognition on the conversation representation information, and extracting the conversation keywords from the conversation representation information; The conversation keywords are represented by word vectors to obtain keyword semantic information, and the keyword semantic information is used as the conversation semantic feature.

5. The information reply method according to claim 1, characterized in that: The step of performing reasoning in the insurance field knowledge graph based on the conversation keywords and the conversation semantic features to obtain the user intention category specifically includes: Inputting the conversation keywords and the conversation semantic features into the insurance field knowledge graph, and performing reasoning on the insurance field knowledge graph based on a knowledge reasoning algorithm to obtain a knowledge reasoning result; The knowledge reasoning result is input into a pre-built intention recognition model to obtain the user intention category.

6. The information reply method according to claim 1, characterized in that: The step of calculating the confidence score of the user intention category and determining whether the confidence score is less than a preset confidence threshold specifically includes: Inputting the user conversation information and the user intention category into a pre-built confidence prediction model to obtain the confidence score; Matching the confidence scores with the user intent categories to obtain an intent confidence score set; Extracting a corresponding preset confidence threshold from a database according to the user intention category; The intention confidence score set is sorted from high to low according to the confidence score, and the confidence score ranked first is extracted and compared with the preset confidence threshold.

7. The information reply method according to claim 1, characterized in that: The step of obtaining the detailed information of user needs, and querying the insurance field knowledge graph according to the detailed information of user needs, and obtaining the first reply content for replying specifically includes: Determine a corresponding information database according to the user intention category, and extract corresponding initial reply information from the information database according to the user intention category; Returning the initial reply information to the user interface for reply, and obtaining detailed information on user needs returned by the user; Search the insurance field knowledge graph according to the user demand refinement information to obtain demand entity nodes and demand relationship edges; Acquire node attribute information of related entity nodes according to the demand entity node and the demand relationship edge, and determine key attribute information in the node attribute information of the related entity nodes according to the user demand refinement information; Matching a corresponding reply template in a preset reply template library according to the key attribute information and the user demand refinement information; The key attribute information is filled into the reply template as a parameter to generate the first reply content, and the first reply content is sent to the user interface for reply.

8. An information reply device, characterized in that: include: An information acquisition module, used to acquire insurance structural information and insurance non-structural information; A graph construction module, used to construct an insurance field knowledge graph based on the insurance structure information and the insurance non-structure information; An information analysis module is used to obtain user conversation information, perform semantic analysis on the user conversation information, and obtain conversation keywords and conversation semantic features; An information reasoning module, used to perform reasoning in the insurance field knowledge graph according to the conversation keywords and the conversation semantic features to obtain a user intention category; A score comparison module, used to calculate the confidence score of the user intention category and determine whether the confidence score is less than a preset confidence threshold; A first processing module, configured to obtain user demand refinement information if the confidence score is less than the preset confidence threshold, and query the insurance field knowledge graph according to the user demand refinement information to generate a first reply content for reply; The second processing module is used to search the insurance field knowledge graph according to the user intention category if the confidence score is greater than or equal to the preset confidence threshold, extract the corresponding knowledge information, fill the knowledge information into the preset reply template, and generate a second reply content for reply.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the information reply method as claimed in any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the information reply method according to any one of claims 1 to 7.

Images