Information generation method, device and equipment based on underwriting large model and storage medium

Abstract

The invention relates to the field of data analysis, is applied to the field of financial business science and technology, and discloses an information generation method, device and equipment based on an underwriting large model, and a storage medium, and the method comprises the steps: determining at least one user intention of a current user; distributing the user intention to a preset underwriting model; and performing parallel question answering processing on each user intention through the underwriting model to generate question answering sub-information, and integrating the question answering sub-information to generate target question answering information. Through the above mode, the user intention is allocated to each preset underwriting model matched with the type of each user intention, and different preset underwriting models concentrate on processing the specific type of user intention, so that each preset underwriting model processes the user intention in parallel, the target question answering information can be quickly generated, and the user experience is improved. In the underwriting scene in the financial insurance field, the consultation and question answering efficiency of the business personnel to the user is improved.

Description

Technical Field

[0001] The present application relates to the field of data analysis and is applied to the field of financial business technology, and in particular to an information generation method, device, equipment and storage medium based on an underwriting big model. Background Art

[0002] In the traditional financial and insurance business, underwriters need to frequently communicate with sales staff about product information during the stages of business development, quotation inquiry, policy transfer, and policy issuance. The types of questions and answers include but are not limited to product highlights, product consultation, quotation inquiry materials, product comparison, etc. The current status of business Q&A is that the average number of online consultations per underwriter is 21.3 per day, and the first response to a question takes 10 hours. The workload is extremely heavy and the efficiency is very low, resulting in insufficient front-line support and a lack of response from the back-line.

[0003] In order to solve this problem, the industry often adopts the RAG (Retrieval-augmented Generat ion) solution, which uses external knowledge to refer to the large model's replies and constrain the large model's reply content, so that the answers are based on something. However, due to the complexity of underwriting scenarios, it is impossible to complete all tasks with a single RAG retrieval tool. At the same time, calling multiple RAG tools is extremely time-consuming. The end-to-end process often exceeds 20 seconds, and the user experience is very poor. It is completely unable to replace manual consultation. Therefore, how to improve the efficiency of business personnel's consultation and answering questions for users in underwriting scenarios in the financial insurance field has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present application provides an information generation method, device, equipment and storage medium based on an underwriting big model, so as to improve the efficiency of business personnel in answering users' questions in underwriting scenarios in the financial insurance field.

[0005] In a first aspect, the present application provides a method for generating information based on an underwriting big model, the method comprising:

[0006] Analyze the user's initial query information to determine at least one user intention of the current user;

[0007] Allocating each of the user intentions to each preset underwriting model that matches the type of each of the user intentions;

[0008] The preset underwriting models are used to perform parallel question-answering processing on the user intentions to generate question-answering sub-information, and the question-answering sub-information is integrated to generate target question-answering information.

[0009] In a second aspect, the present application further provides an information generation device based on an underwriting big model, the device comprising:

[0010] A user intention analysis module is used to analyze the user's initial query information and determine at least one user intention of the current user;

[0011] A preset underwriting model determination module, used to assign each of the user intentions to each preset underwriting model that matches the type of each of the user intentions;

[0012] The target question-answering information generation module is used to perform parallel question-answering processing on each of the user intentions through each of the preset underwriting models, generate question-answering sub-information, and integrate each of the question-answering sub-information to generate the target question-answering information.

[0013] In a third aspect, the present application also provides a computer device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the information generation method based on the underwriting big model as described above when executing the computer program.

[0014] In a fourth aspect, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements the information generation method based on the underwriting big model as described above.

[0015] The present application discloses an information generation method, device, equipment and storage medium based on a large underwriting model, wherein the information generation method based on the large underwriting model includes parsing the user's initial inquiry information to determine at least one user intention of the current user; assigning each of the user intentions to each preset underwriting model that matches the type of each of the user intentions; performing parallel question-answering processing on each of the user intentions through each of the preset underwriting models to generate question-answering sub-information, and integrating each of the question-answering sub-information to generate target question-answering information. In the above manner, the present application assigns user intentions to each of the preset underwriting models that matches the type of each of the user intentions, and different preset underwriting models focus on processing specific types of user intentions. Therefore, each preset underwriting model processes user intentions in parallel, and can quickly generate target question-answering information, thereby improving the efficiency of business personnel in consulting and answering users in underwriting scenarios in the financial insurance field. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. 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.

[0017] Figure 1is a schematic flow chart of an information generation method based on an underwriting big model provided in the first embodiment of the present application;

[0018] Figure 2 is a schematic flow chart of an information generation method based on an underwriting big model provided in the second embodiment of the present application;

[0019] Figure 3 A schematic block diagram of an information generation device based on an underwriting model provided in an embodiment of the present application;

[0020] Figure 4 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0023] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0024] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0025] The embodiments of the present application provide an information generation method, device, equipment and storage medium based on a large underwriting model. Among them, the information generation method based on the large underwriting model can be applied to a server, by assigning user intent to each preset underwriting model that matches the type of each user intent, and different preset underwriting models focus on processing specific types of user intent, so that each preset underwriting model processes the user intent in parallel, and can quickly generate target question-answering information, which improves the efficiency of business personnel in answering questions to users in underwriting scenarios in the financial insurance field. Among them, the server can be an independent server or a server cluster.

[0026] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0027] See also Figure 1 , Figure 1 This is a schematic flow chart of an information generation method based on an underwriting big model provided by the first embodiment of the present application. The information generation method based on the underwriting big model can be applied to a server, and is used to assign user intent to each preset underwriting model that matches the type of each user intent. Different preset underwriting models focus on processing specific types of user intents, so that each preset underwriting model processes user intents in parallel, and can quickly generate target question-answering information, which improves the efficiency of business personnel in answering questions to users in underwriting scenarios in the financial insurance field.

[0028] like Figure 1 As shown, the information generation method based on the underwriting big model specifically includes steps S10 to S30.

[0029] Step S10: parse the user's initial query information to determine at least one user intention of the current user;

[0030] Step S20: assigning each of the user intentions to each preset underwriting model that matches the type of each of the user intentions;

[0031] Step S30: Perform parallel question-answering processing on each of the user intentions through each of the preset underwriting models to generate question-answering sub-information, and integrate each of the question-answering sub-information to generate target question-answering information.

[0032] In one embodiment, this embodiment is applied to a multi-agent-based parallel LLM (Large Language Model 1) question answering model, by constructing multiple agents to participate in RAG and reasoning in parallel. Specifically, the agent architecture includes:

[0033] The document retrieval sub-model provides the system with the latest and relevant information by dynamically retrieving relevant documents in real time. The agent can retrieve information from a variety of data sources (such as internal company databases, public insurance information platforms, regulatory document libraries, etc.) to ensure the accuracy and timeliness of the answers;

[0034] FAQ (frequently asked questions) agent, i.e. question bank maintenance sub-model, can maintain and continuously update the FAQ bank and quickly respond to FAQs based on user input. Specifically, through machine learning and data mining technology, the question bank maintenance sub-model can automatically identify and collect new FAQs and optimize the coverage and accuracy of answers;

[0035] The product recommendation sub-model is used to recommend the most suitable insurance products based on the user's specific needs and intentions. The agent combines information such as user profile, historical purchase records, and current market trends to provide personalized recommendations, thereby improving user satisfaction and the commercial value of the system.

[0036] The industry prediction sub-model can output the risk type of the industry or job mentioned by the user based on the user's specific input and different insurance types, assisting the intelligent agent in reasoning and decision-making, such as high-risk industries and 5-6 types of occupations that require caution when insuring, thereby improving the system's underwriting decision-making capabilities.

[0037] The present embodiment discloses an information generation method, apparatus, device and storage medium based on a large underwriting model, wherein the information generation method based on the large underwriting model includes parsing the user's initial inquiry information to determine at least one user intention of the current user; assigning each of the user intentions to each preset underwriting model that matches the type of each of the user intentions; performing parallel question-answering processing on each of the user intentions through each of the preset underwriting models to generate question-answering sub-information, and integrating each of the question-answering sub-information to generate target question-answering information. In the above manner, the present application assigns user intentions to each of the preset underwriting models that matches the type of each of the user intentions, and different preset underwriting models focus on processing specific types of user intentions. Therefore, each preset underwriting model processes user intentions in parallel, and can quickly generate target question-answering information, thereby improving the efficiency of business personnel in answering questions to users in underwriting scenarios in the financial insurance field.

[0038] See also Figure 2 , Figure 2This is a schematic flow chart of an information generation method based on an underwriting big model provided by the second embodiment of the present application. The information generation method based on the underwriting big model can be applied to a server, and is used to assign user intent to each preset underwriting model that matches the type of each user intent. Different preset underwriting models focus on processing specific types of user intents, so that each preset underwriting model processes user intents in parallel, and can quickly generate target question-answering information, which improves the efficiency of business personnel in answering questions to users in underwriting scenarios in the financial insurance field.

[0039] based on Figure 1 The embodiment shown, this embodiment Figure 2 As shown, step S10 includes steps S101 to S103.

[0040] Step S101: calling a sensitive word detection model, and detecting whether there are query sensitive words in the user's initial query information through the sensitive word detection model;

[0041] Step S102: if the sensitive words exist in the initial query information of the user, filter the sensitive words to generate query information to be identified;

[0042] Step S103: extracting key field values ​​from the query information to be identified, and determining the intentions of each user according to the key field values ​​and a preset task database.

[0043] In one embodiment, the multi-agent-based parallel LLM question answering agent also includes a sensitive word interception agent, which is used to detect and intercept sensitive words in user input to ensure the compliance and security of the system. Through natural language processing technology, the sensitive word interception agent can identify sensitive content in complex contexts and improve the compliance detection capability of the system.

[0044] The present embodiment discloses an information generation method, apparatus, device and storage medium based on an underwriting big model. The information generation method based on the underwriting big model includes calling a sensitive word detection model, and detecting whether there are inquiry sensitive words in the user's initial inquiry information through the sensitive word detection model; when the sensitive words exist in the user's initial inquiry information, filtering the sensitive words to generate inquiry information to be identified; extracting key field values ​​in the inquiry information to be identified, and determining each user intention based on the key field values ​​and a preset task database; assigning each user intention to each preset underwriting model that matches the type of each user intention; performing parallel question-answering processing on each user intention through each preset underwriting model to generate question-answering sub-information, and integrating each question-answering sub-information to generate target question-answering information. Through the above method, the present application assigns user intentions to preset underwriting models that match the types of each user intention. Different preset underwriting models focus on processing specific types of user intentions. Therefore, each preset underwriting model processes user intentions in parallel, and can quickly generate target question and answer information. In the underwriting scenario in the financial insurance field, the efficiency of business personnel in answering users' questions is improved.

[0045] In a more preferred embodiment, step S101 includes:

[0046] Performing word segmentation processing on the user's initial query information to generate at least one query word segmentation, and determining irrelevant characters in each query word segmentation;

[0047] Matching each query segmentation word with each preset sensitive word in the preset sensitive word list one by one;

[0048] The query segmentation word that matches the preset sensitive word is determined as the query sensitive word.

[0049] In one embodiment, the query information is input through the interface, the user's original query information is received and recorded, and the user's query information is preprocessed, including conversion to a unified character encoding, conversion to lowercase (if necessary), removal of redundant spaces, etc. The user's query information is segmented using a word segmentation algorithm to generate at least one query segmentation word. Commonly used word segmentation algorithms include forward maximum matching method, reverse maximum matching method, conditional random field, etc.

[0050] Analyze the characters in each query segmentation to determine irrelevant characters, such as punctuation marks, special symbols, etc. You can use a predefined list of irrelevant characters for matching, or use a stop word list in a natural language processing library and remove the determined irrelevant characters from the query segmentation to retain meaningful words.

[0051] The query segmented words after removing irrelevant characters are matched one by one with the sensitive words in the preset sensitive word list. Specifically, a simple string matching algorithm or a more complex algorithm such as fuzzy matching, regular expression matching, etc. can be used, and the query segmented words that match the preset sensitive words are determined as query sensitive words.

[0052] The present embodiment discloses an information generation method, device, equipment and storage medium based on an underwriting big model. The information generation method based on the underwriting big model includes performing word segmentation processing on the user's initial inquiry information, generating at least one inquiry word segmentation, and determining irrelevant characters in each of the inquiry word segmentations; matching each of the inquiry word segmentations with each preset sensitive word in a preset sensitive word list one by one; determining the inquiry word segmentation that matches the preset sensitive word as the inquiry sensitive word; if the sensitive word exists in the user's initial inquiry information, filtering the sensitive word to generate inquiry information to be identified; extracting key field values ​​in the inquiry information to be identified, and determining each of the user intentions based on the key field values ​​and a preset task database; assigning each of the user intentions to each preset underwriting model that matches the type of each of the user intentions; performing parallel question-answering processing on each of the user intentions through each of the preset underwriting models to generate question-answering sub-information, and integrating each of the question-answering sub-information to generate target question-answering information. Through the above method, the present application assigns user intentions to preset underwriting models that match the types of each user intention. Different preset underwriting models focus on processing specific types of user intentions. Therefore, each preset underwriting model processes user intentions in parallel, and can quickly generate target question and answer information. In the underwriting scenario in the financial insurance field, the efficiency of business personnel in answering users' questions is improved.

[0053] Furthermore, in a more preferred embodiment, after matching each query segmentation word with each preset sensitive word in the preset sensitive word list one by one, the following steps are included:

[0054] If there is no query participle matching the preset sensitive word, determining the contextual semantics of each query participle in the user's initial query information;

[0055] Detecting whether the context semantics includes the preset sensitive word or whether there is a sensitive associated word whose similarity with the preset sensitive word is higher than a preset threshold;

[0056] If the context semantics includes at least one of the preset sensitive words and / or the sensitive associated words, the query participle corresponding to the context semantics is determined as the query sensitive word.

[0057] In one embodiment, after completing the word segmentation and removing irrelevant characters, context semantic analysis is performed on each query word. Natural language processing technology, such as dependency syntactic analysis or semantic role labeling, is used to understand the semantic role and meaning of the query word in the original query information, so as to construct a context window for each query word, which usually includes a certain number of words before and after the word, to capture the surrounding context of the query word so as to more accurately analyze its semantics.

[0058] The contextual semantics of each query segment is matched with the entries in the preset sensitive word library, and the semantic similarity is evaluated using a semantic matching algorithm, such as a pre-trained language model such as a vector space model. For the contextual semantics of each query segment, its similarity with the preset sensitive words is calculated.

[0059] A preset threshold is set to determine when the similarity is considered high enough to be considered as a sensitive associated word, and whether the context semantics contains any preset sensitive words is detected. At the same time, whether there are sensitive associated words in the context semantics whose similarity with the preset sensitive words is higher than the preset threshold is detected.

[0060] The present embodiment discloses an information generation method, device, equipment and storage medium based on an underwriting big model. The information generation method based on the underwriting big model includes segmenting the user's initial query information, generating at least one query segmentation, and determining irrelevant characters in each query segmentation; matching each query segmentation with each preset sensitive word in a preset sensitive word list one by one; if there is no query segmentation matching the preset sensitive word, determining the context semantics of each query segmentation in the user's initial query information; detecting whether the context semantics contains the preset sensitive word or there is a sensitive associated word whose similarity with the preset sensitive word is higher than a preset threshold; if the context segmentation matches the preset sensitive word, determining the context semantics of each query segmentation in the user's initial query information; detecting whether the context semantics contains the preset sensitive word or there is a sensitive associated word whose similarity with the preset sensitive word is higher than a preset threshold; If the semantics contains at least one of the preset sensitive words and / or the sensitive associated words, the query segmentation corresponding to the context semantics is determined as the query sensitive word; if the sensitive words exist in the user's initial query information, the sensitive words are filtered to generate the query information to be identified; the key field values ​​in the query information to be identified are extracted, and the user intentions are determined according to the key field values ​​and the preset task database; each user intention is assigned to each preset underwriting model that matches the type of each user intention; each user intention is processed in parallel through each preset underwriting model to generate question-answering sub-information, and each question-answering sub-information is integrated to generate target question-answering information. In the above manner, the present application assigns user intentions to each preset underwriting model that matches the type of each user intention, and different preset underwriting models focus on processing specific types of user intentions. Therefore, each preset underwriting model processes user intentions in parallel, and can quickly generate target question-answering information, which improves the efficiency of business personnel in consulting and answering users in the underwriting scenario in the financial insurance field.

[0061] based on Figure 1 In the embodiment shown, in this embodiment, step S20 includes:

[0062] Classifying each of the user intentions through a semantic analysis model to determine the type of each of the user intentions, wherein the type of the user intentions includes claim information, policy query information and / or business consulting information;

[0063] The preset underwriting model is called according to the type of the user's intention.

[0064] In one embodiment, the multi-agent based parallel LLM question answering agent further includes:

[0065] The user intention distribution agent is used to parse the query content entered by the user, identify its intention, and distribute different types of requests to the corresponding processing agents. Unlike traditional systems, the intention distribution agent can understand complex intentions through multi-level analysis, such as users' specific needs in the insurance field (such as claims, policy inquiries, new product consultations, etc.), thereby improving the accuracy of task allocation.

[0066] This embodiment discloses an information generation method, device, equipment and storage medium based on an underwriting big model, wherein the information generation method based on the underwriting big model includes parsing the user's initial inquiry information to determine at least one user intention of the current user; classifying each of the user intentions through a semantic analysis model to determine the type of each of the user intentions, wherein the type of the user intention includes claim information, policy query information and / or business consulting information; calling the preset underwriting model according to the type of the user intention; performing parallel Q&A processing on each of the user intentions through each of the preset underwriting models to generate Q&A sub-information, and integrating each of the Q&A sub-information to generate target Q&A information. In the above manner, the present application assigns user intentions to each of the preset underwriting models that match the type of each of the user intentions, and different preset underwriting models focus on processing specific types of user intentions, so that each preset underwriting model processes the user intention in parallel, and can quickly generate target Q&A information, thereby improving the efficiency of business personnel in answering questions to users in the underwriting scenario in the financial insurance field.

[0067] In a more preferred embodiment, step S30 includes:

[0068] Maintaining the sub-model through the question library, determining historical questions and / or historical question-answering information matching the user's intention from a preset question library, and generating the question-answering sub-information matching the user's intention based on the historical questions and / or the historical question-answering information;

[0069] If the historical questions and / or historical question-answering information matching the user's intention do not exist in the preset question library, then the document retrieval sub-model is used to dynamically search according to the user's intention, determine the relevant information related to the user's intention, generate the question-answering sub-information, and update the user's intention and the question-answering sub-information to the preset question library;

[0070] Analyze the user intention through the product recommendation sub-model, determine the target product matching the user intention, and use the target product as the question-answering sub-information;

[0071] The industry prediction sub-model is used to identify industry keywords in the user's intention, and based on each of the industry keywords, the industry information of the current user is predicted, and the industry information is used as the question-answering sub-information.

[0072] In a specific embodiment, at the first moment of user input, the system simultaneously calls the user intention distribution agent and the sensitive word interception agent to ensure rapid analysis of user intention and filtering of non-compliant content, while accurately matching the user input with the underlying knowledge base question. If the similarity is 100%, the corresponding answer is directly output. This parallel processing significantly shortens the processing time of the front-end link. Compared with traditional serial processing, the user experience is greatly improved and the response time is significantly shortened.

[0073] After the user intent is distributed, the document retrieval submodel, question base maintenance submodel, industry prediction submodel, and product recommendation submodel can work in parallel based on the results of the intent distribution. For example, when processing a complex query, the document retrieval submodel can retrieve relevant documents, the question base maintenance submodel can provide direct answers to common questions, and the product recommendation submodel can recommend suitable insurance products. The application of multi-threaded parallel processing enables these agents to work simultaneously, significantly improving the system's response speed and overall processing efficiency.

[0074] Through multi-threading technology, agents can collaborate efficiently. For example, when the document retrieval sub-model finds relevant documents, it can promptly notify the question library maintenance sub-model and product recommendation sub-model to provide more comprehensive answers. The collaborative work between agents not only improves the efficiency of individual agents, but also improves the overall coordination and user experience of the system.

[0075] Furthermore, step S30 also includes:

[0076] Pre-acquire a question-answering information template, wherein the question-answering information template includes various information dimensions of the question-answering information;

[0077] According to the type of each question-answering sub-information, each question-answering sub-information is encoded to generate a question-answering encoding field value;

[0078] Fill each of the question-answering coded field values ​​into each of the information dimensions matching the type of the question-answering sub-information in the question-answering information template to generate question-answering information to be decoded;

[0079] The to-be-decoded question-answering information is decoded to generate the target question-answering information.

[0080] In a specific embodiment, a Q&A information template is pre-designed and obtained. The Q&A information template defines various information dimensions of the Q&A information, such as problem description, answer, relevant regulations, reference document link, etc. Each Q&A sub-information is encoded according to the type of each Q&A sub-information (such as product information, claims process, FAQ, etc.). The code can be a number, letter or a string in a specific format, which is used to uniquely identify each type of Q&A sub-information.

[0081] Generate a Q&A coding field value for each Q&A sub-information. This value will be used to identify and fill in the corresponding information dimension in the template. Fill the generated Q&A coding field value into the information dimension that matches the Q&A sub-information type in the Q&A information template. Integrate the filled Q&A information template to form the Q&A information to be decoded. This information to be decoded may contain coding field values ​​and some predefined formats, which need to be further decoded to generate the final readable Q&A information.

[0082] The question-and-answer information to be decoded is decoded, and the encoded field value is converted back to the original question-and-answer sub-information. The decoding process may involve looking up the encoding mapping table, performing reverse operations, or using a decoding algorithm to integrate the decoded question-and-answer sub-information into the corresponding position of the question-and-answer information template to generate the final target question-and-answer information.

[0083] See also Figure 3 , Figure 3 The embodiment of the present application provides a schematic block diagram of an information generation device based on an underwriting big model, and the information generation device based on the underwriting big model is used to execute the aforementioned information generation method based on the underwriting big model. The information generation device based on the underwriting big model can be configured on a server.

[0084] like Figure 3 As shown, the information generation device 400 based on the underwriting big model includes:

[0085] The user intention analysis module 410 is used to analyze the user's initial query information and determine at least one user intention of the current user;

[0086] A preset underwriting model determination module 420, configured to assign each of the user intentions to each preset underwriting model that matches the type of each of the user intentions;

[0087] The target question-answering information generating module 430 is used to perform parallel question-answering processing on each of the user intentions through each of the preset underwriting models, generate question-answering sub-information, and integrate each of the question-answering sub-information to generate the target question-answering information.

[0088] Furthermore, the user intention analysis module 410 includes:

[0089] A query sensitive word detection unit, used to call a sensitive word detection model and detect whether there is a query sensitive word in the user's initial query information through the sensitive word detection model;

[0090] a query information generation unit to be identified, configured to filter the sensitive words and generate the query information to be identified when the sensitive words exist in the initial query information of the user;

[0091] The user intention determination unit is used to extract the key field value in the query information to be identified, and determine each user intention according to the key field value and a preset task database.

[0092] Furthermore, the query sensitive word detection unit includes:

[0093] An irrelevant character determination subunit, configured to perform word segmentation processing on the user's initial query information, generate at least one query word segmentation, and determine irrelevant characters in each of the query word segmentations;

[0094] A preset sensitive word matching subunit, used to match each of the query segment words with each preset sensitive word in the preset sensitive word list one by one;

[0095] The query sensitive word determination subunit is used to determine the query segmentation word that matches the preset sensitive word as the query sensitive word.

[0096] Furthermore, the query sensitive word detection unit includes:

[0097] A context semantics determination subunit, configured to determine the context semantics of each query participle in the user's initial query information if there is no query participle matching the preset sensitive word;

[0098] A sensitive associated word detection subunit, used to detect whether the context semantics contains the preset sensitive word or whether there is a sensitive associated word whose similarity with the preset sensitive word is higher than a preset threshold;

[0099] The query sensitive word determination subunit is used to determine the query participle corresponding to the context semantics as the query sensitive word if the context semantics includes at least one of the preset sensitive words and / or the sensitive associated words.

[0100] Furthermore, the preset underwriting model determination module 420 includes:

[0101] A user intention type determination unit, configured to classify each of the user intentions through a semantic analysis model to determine the type of each of the user intentions, wherein the type of the user intention includes claim information, policy query information and / or business consulting information;

[0102] The preset underwriting model calling unit is used to call the preset underwriting model according to the type of the user intention.

[0103] Furthermore, the target question answering information generating module 430 includes:

[0104] A question library maintenance unit, configured to determine historical questions and / or historical question-answering information matching the user's intention from a preset question library through the question library maintenance sub-model, and generate the question-answering sub-information matching the user's intention based on the historical questions and / or the historical question-answering information;

[0105] A question-answering sub-information updating unit is used for, if the historical question and / or the historical question-answering information matching the user intention does not exist in the preset question library, to dynamically search according to the user intention through the document retrieval sub-model, determine the relevant information related to the user intention, generate the question-answering sub-information, and update the user intention and the question-answering sub-information to the preset question library;

[0106] a target product determination unit, configured to analyze the user intention through the product recommendation sub-model, determine a target product matching the user intention, and use the target product as the question-answering sub-information;

[0107] The industry information prediction unit is used to identify the industry keywords in the user's intention through the industry prediction sub-model, and predict the industry information of the current user based on each of the industry keywords, and use the industry information as the question-answering sub-information.

[0108] Furthermore, the target question answering information generating module 430 includes:

[0109] A question-answering information template acquiring unit, used to pre-acquire a question-answering information template, wherein the question-answering information template includes various information dimensions of the question-answering information;

[0110] A question-answering code field value generating unit, configured to encode each question-answering sub-information according to the type of each question-answering sub-information, and generate a question-answering code field value;

[0111] A unit for generating question-answering information to be decoded, configured to fill each of the question-answering coding field values ​​into each of the information dimensions in the question-answering information template that matches the type of the question-answering sub-information, to generate question-answering information to be decoded;

[0112] The target question-answering information generating unit is used to decode the question-answering information to be decoded to generate the target question-answering information.

[0113] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0114] The above-mentioned device can be implemented in the form of a computer program. Figure 4 Runs on the computer device shown.

[0115] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0116] See also Figure 4 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0117] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any information generation method based on the underwriting big model.

[0118] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0119] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any information generation method based on the underwriting model.

[0120] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will appreciate that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0121] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0122] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0123] Analyze the user's initial query information to determine at least one user intention of the current user;

[0124] Allocating each of the user intentions to each preset underwriting model that matches the type of each of the user intentions;

[0125] The preset underwriting models are used to perform parallel question-answering processing on the user intentions to generate question-answering sub-information, and the question-answering sub-information is integrated to generate target question-answering information.

[0126] In one embodiment, the user's initial query information is parsed to determine at least one user intention of the current user, so as to achieve:

[0127] Calling a sensitive word detection model, and detecting whether there are query sensitive words in the user's initial query information through the sensitive word detection model;

[0128] When the sensitive words exist in the initial query information of the user, filtering the sensitive words to generate query information to be identified;

[0129] Extract key field values ​​from the query information to be identified, and determine each user intention based on the key field values ​​and a preset task database.

[0130] In one embodiment, a sensitive word detection model is called, and the sensitive word detection model is used to detect whether there are query sensitive words in the user's initial query information, so as to achieve:

[0131] Performing word segmentation processing on the user's initial query information to generate at least one query word segmentation, and determining irrelevant characters in each query word segmentation;

[0132] Matching each query segmentation word with each preset sensitive word in the preset sensitive word list one by one;

[0133] The query segmentation word that matches the preset sensitive word is determined as the query sensitive word.

[0134] In one embodiment, after matching each query segment with each preset sensitive word in the preset sensitive word list one by one, it is used to achieve:

[0135] If there is no query participle matching the preset sensitive word, determining the contextual semantics of each query participle in the user's initial query information;

[0136] Detecting whether the context semantics includes the preset sensitive word or whether there is a sensitive associated word whose similarity with the preset sensitive word is higher than a preset threshold;

[0137] If the context semantics includes at least one of the preset sensitive words and / or the sensitive associated words, the query participle corresponding to the context semantics is determined as the query sensitive word.

[0138] In one embodiment, each of the user intentions is assigned to each preset underwriting model that matches the type of each of the user intentions, so as to achieve:

[0139] Classifying each of the user intentions through a semantic analysis model to determine the type of each of the user intentions, wherein the type of the user intentions includes claim information, policy query information and / or business consulting information;

[0140] The preset underwriting model is called according to the type of the user's intention.

[0141] In one embodiment, each of the preset underwriting models is used to perform parallel question-answering processing on each of the user intentions to generate question-answering sub-information for achieving:

[0142] Maintaining the sub-model through the question library, determining historical questions and / or historical question-answering information matching the user's intention from a preset question library, and generating the question-answering sub-information matching the user's intention based on the historical questions and / or the historical question-answering information;

[0143] If the historical questions and / or historical question-answering information matching the user's intention do not exist in the preset question library, then the document retrieval sub-model is used to dynamically search according to the user's intention, determine the relevant information related to the user's intention, generate the question-answering sub-information, and update the user's intention and the question-answering sub-information to the preset question library;

[0144] Analyze the user intention through the product recommendation sub-model, determine the target product matching the user intention, and use the target product as the question-answering sub-information;

[0145] The industry prediction sub-model is used to identify industry keywords in the user's intention, and based on each of the industry keywords, the industry information of the current user is predicted, and the industry information is used as the question-answering sub-information.

[0146] In one embodiment, each of the question-answering sub-information is integrated to generate target question-answering information, which is used to achieve:

[0147] Pre-acquire a question-answering information template, wherein the question-answering information template includes various information dimensions of the question-answering information;

[0148] According to the type of each question-answering sub-information, each question-answering sub-information is encoded to generate a question-answering encoding field value;

[0149] Fill each of the question-answering coded field values ​​into each of the information dimensions matching the type of the question-answering sub-information in the question-answering information template to generate question-answering information to be decoded;

[0150] The to-be-decoded question-answering information is decoded to generate the target question-answering information.

[0151] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any one of the information generation methods based on the underwriting big model provided in the embodiments of the present application.

[0152] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Medi a Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card (Flash Card), etc., equipped on the computer device.

[0153] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for generating information based on an underwriting big model, characterized in that: include: Analyze the user's initial query information to determine at least one user intention of the current user; Allocating each of the user intentions to each preset underwriting model that matches the type of each of the user intentions; The preset underwriting models are used to perform parallel question-answering processing on the user intentions to generate question-answering sub-information, and the question-answering sub-information is integrated to generate target question-answering information.

2. The information generation method based on the underwriting big model according to claim 1 is characterized in that: The parsing of the initial user query information to determine at least one user intention of the current user includes: Calling a sensitive word detection model, and detecting whether there are query sensitive words in the user's initial query information through the sensitive word detection model; When the sensitive words exist in the initial query information of the user, filtering the sensitive words to generate query information to be identified; Extract key field values ​​from the query information to be identified, and determine each user intention based on the key field values ​​and a preset task database.

3. The information generation method based on the underwriting big model according to claim 2 is characterized in that: The calling of the sensitive word detection model and detecting whether there is a query sensitive word in the user's initial query information by the sensitive word detection model includes: Performing word segmentation processing on the user's initial query information to generate at least one query word segmentation, and determining irrelevant characters in each query word segmentation; Matching each query segmentation word with each preset sensitive word in the preset sensitive word list one by one; The query segmentation word that matches the preset sensitive word is determined as the query sensitive word.

4. The information generation method based on the underwriting big model according to claim 3 is characterized in that: After matching each query segment word with each preset sensitive word in the preset sensitive word list one by one, the method further includes: If there is no query participle matching the preset sensitive word, determining the contextual semantics of each query participle in the user's initial query information; Detecting whether the context semantics includes the preset sensitive word or whether there is a sensitive associated word whose similarity with the preset sensitive word is higher than a preset threshold; If the context semantics includes at least one of the preset sensitive words and / or the sensitive associated words, the query participle corresponding to the context semantics is determined as the query sensitive word.

5. The information generation method based on the underwriting big model according to claim 1 is characterized in that: The allocating each of the user intentions to each of the preset underwriting models matching the type of each of the user intentions includes: Classifying each of the user intentions through a semantic analysis model to determine the type of each of the user intentions, wherein the type of the user intentions includes claim information, policy query information and / or business consulting information; The preset underwriting model is called according to the type of the user's intention.

6. The information generation method based on the underwriting big model according to claim 1 is characterized in that: The preset underwriting model includes a document retrieval sub-model, a question library maintenance sub-model, a product recommendation sub-model and / or an industry prediction sub-model. The preset underwriting model is used to perform parallel question-answering processing on each user intention to generate question-answering sub-information, including: Maintaining the sub-model through the question library, determining historical questions and / or historical question-answering information matching the user's intention from a preset question library, and generating the question-answering sub-information matching the user's intention based on the historical questions and / or the historical question-answering information; If the historical questions and / or historical question-answering information matching the user's intention do not exist in the preset question library, then the document retrieval sub-model is used to dynamically search according to the user's intention, determine the relevant information related to the user's intention, generate the question-answering sub-information, and update the user's intention and the question-answering sub-information to the preset question library; Analyze the user intention through the product recommendation sub-model, determine the target product matching the user intention, and use the target product as the question-answering sub-information; The industry prediction sub-model is used to identify industry keywords in the user's intention, and based on each of the industry keywords, the industry information of the current user is predicted, and the industry information is used as the question-answering sub-information.

7. The information generation method based on the underwriting big model according to claim 6 is characterized in that: The step of integrating the question-answering sub-information to generate target question-answering information includes: Pre-acquire a question-answering information template, wherein the question-answering information template includes various information dimensions of the question-answering information; According to the type of each question-answering sub-information, each question-answering sub-information is encoded to generate a question-answering encoding field value; Fill each of the question-answering coded field values ​​into each of the information dimensions matching the type of the question-answering sub-information in the question-answering information template to generate question-answering information to be decoded; The to-be-decoded question-answering information is decoded to generate the target question-answering information.

8. An information generation device based on an underwriting big model, characterized in that: include: A user intention analysis module is used to analyze the user's initial query information and determine at least one user intention of the current user; A preset underwriting model determination module, used to assign each of the user intentions to each preset underwriting model that matches the type of each of the user intentions; The target question-answering information generation module is used to perform parallel question-answering processing on each of the user intentions through each of the preset underwriting models, generate question-answering sub-information, and integrate each of the question-answering sub-information to generate the target question-answering information.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the information generation method based on the underwriting big model as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the information generation method based on the underwriting model as described in any one of claims 1 to 7.

Images