Intelligent recommendation method and device for insurance products, computing equipment and storage medium

Abstract

The embodiment of the invention belongs to the technical field of artificial intelligence, and particularly relates to an intelligent recommendation method and device for insurance products, computing equipment and a storage medium, and the method comprises the steps: extracting key information of the insurance products from an insurance product poster through an OCR technology and a large language model; enterprise information of the insurance applicant and personal information of the insured person are extracted, the enterprise information comprises enterprise names, unified credit codes, belonging industries, regions and operating ranges, and the personal information comprises employee names, position names, ages, genders and births; performing standardization processing on position names provided by enterprises and supplementing position description information; and carrying out region matching, occupational matching, age matching and number matching on the key information of the insurance product, the enterprise information of the insurance applicant and the personal information of the insured, and outputting an insurance product recommendation set for the insurance applicant by combining matching results.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of artificial intelligence, and in particular, to an intelligent recommendation method and device for insurance products, a computing device and a storage medium. BACKGROUND

[0002] Current insurance product recommendations mostly rely on recommendation algorithms similar to e-commerce platforms. E-commerce recommendation algorithms usually predict and recommend related products through user historical purchase records, browsing behavior, preferences, and other information. Such algorithms focus more on matching user historical behavior and product features. However, for the recommendation of accidental insurance, especially employer liability insurance and group accidental insurance, not only the personal characteristics of customers need to be considered, but also many additional complex limiting factors, such as industry, region, occupation, and other factors, which make insurance product recommendation more complex.

[0003] Specifically, the employees of a customer may involve different occupational categories, and the insurance products required by employees in different occupational categories often have great differences. Traditional customer feature vectors (such as age, gender, income, etc.) cannot effectively distinguish the needs of different occupational categories, resulting in that the recommendation system is difficult to provide accurate product recommendations when facing diversified customer needs. Accidental insurance, especially group accidental insurance, usually has many specific requirements for insurance. For example, different insurance products have restrictions on the region, industry, and occupation category of the insured, and if these restrictions are not effectively incorporated into the recommendation algorithm, the recommended insurance products may not meet the actual needs of the customer, affecting the efficiency and quality of insurance. Insurance products are usually presented in the form of posters, flyers, or product invitations, and these information is mostly manually input and extracted by humans. Different insurance agencies and single systems use different formats of product descriptions, which causes certain difficulties in processing and analysis. Since these information requires a large amount of manual operation and proofreading, manual methods are difficult to cope with the growing and diversified insurance product demand. SUMMARY

[0004] For insurance products, especially complex accidental insurance products, in order to provide more accurate recommendation schemes for the insured, the embodiments described herein provide an intelligent recommendation method and device for insurance products, a computing device, and a computer-readable storage medium storing a computer program.

[0005] According to a first aspect of the present disclosure, a method for intelligent recommendation of insurance products is provided, comprising: extracting key information of an insurance product from an insurance product poster through an OCR technique and a large language model; extracting enterprise information of a policyholder and personal information of a policyholder, the enterprise information including an enterprise name, a unified credit code, an industry to which the enterprise belongs, a region in which the enterprise is located, and a business scope, and the personal information including an employee name, a position name, an age, a gender, and a place of birth; performing standardized processing on the position name provided by the enterprise and supplementing position description information; and performing regional matching, occupation matching, age matching, and number matching on the key information of the insurance product and the enterprise information of the policyholder and the personal information of the policyholder, and outputting an insurance product recommendation set for the policyholder by merging the matching results.

[0006] In some embodiments of the present disclosure, the extracting of the key information of the insurance product from the insurance product poster through the OCR technique and the large language model comprises: performing preprocessing on the insurance product poster, the preprocessing including converting a PDF page into a picture format, performing denoising, enhancing contrast, and adjusting brightness on the picture; identifying the text in the picture using the OCR technique to obtain text information including a product introduction, a premium, a premium, and special terms; inputting the text information identified by the OCR into the large language model, and guiding the model to extract the key information of the product through a prompt word, the key information including a product name, a region of coverage, a region of refusal, a class of occupations of coverage, a class of occupations of refusal, a range of ages of coverage, and a range of numbers of coverage; and storing the key information returned by the large language model as structured data in a predetermined format.

[0007] In some embodiments of the present disclosure, the extracting of the enterprise information of the policyholder and the personal information of the policyholder comprises: using an API of an enterprise information service website or a crawler program to extract the unified credit code, the industry to which the enterprise belongs, the region in which the enterprise is located, and the business scope from the business registration information of the enterprise; extracting the employee name and the position name from an employee management system in the enterprise, and analyzing the age, the gender, and the place of birth from the ID number of the employee.

[0008] In some embodiments of the present disclosure, the standardization and supplement of the name of the enterprise provided and the position description information comprises: based on a text similarity matching algorithm, matching the insurance company occupation classification table with the national standard occupation classification table to obtain the most matched occupation name, supplementing the occupation description information in the national standard occupation table to the occupation classification table of the insurance company product to form a standardized occupation classification table; constructing each occupation item in the standardized occupation classification table as a first industry text block and a first occupation text block; taking the intersection of the first occupation text block and the obtained underwriting occupation category in the insurance company product poster to filter out the insurable occupation category text block; retrieving and obtaining the supplementary description information of the enterprise position from the recruitment website according to the position name provided by the enterprise; integrating the industry to which the enterprise belongs, the position name and the position description information to generate a second industry text block and a second occupation text block.

[0009] In some embodiments of the present disclosure, constructing each occupation item in the standardized occupation classification table as a first industry text block and a first occupation text block comprises: extracting the industry information corresponding to each occupation from the standardized occupation classification table to construct a first industry text block, the first industry text block comprising an industry major category, an industry middle category, an industry minor category and an industry ID number; integrating each occupation item in the standardized occupation classification table into a first occupation text block, the first occupation text block comprising a standardized occupation name, occupation description information, an ID of the industry to which the occupation belongs and a risk level of the occupation.

[0010] In some embodiments of the present disclosure, the key information of the insurance product is regionally matched with the enterprise information of the applicant and the personal information of the insured, and the matched results are output by merging to obtain a recommended set of insurance products for the applicant, including: based on the first industry text block and the second industry text block, comparing in the order of industry category, industry sub-category, and industry sub-sub-category, and matching different categories based on vector similarity ranking, after matching to the industry, comparing the sentence vectors corresponding to the specific occupation description based on the insurable occupation category text block and the second occupation text block, and obtaining an occupation matching candidate set; standardizing the region description of the insurance product and the enterprise address information and the birthplace of the insured, comparing in the order of province, city, and district, and judging whether the enterprise address and the birthplace of the insured belong to the insurable region of the insurance product, screening the insurable insurance products to form a region matching candidate set; performing rule matching according to the age requirement of the insurance product and the age of the applicant, judging whether the employee age is greater than the minimum age and less than the maximum age of the product, if the age of the applicant fully meets the product requirement, adding to the age fully compliant candidate set, if the age of the applicant does not fully meet the product requirement, adding to the age partially compliant candidate set; judging whether the number of employees of the applicant meets the lower limit requirement of the number of applicants of the insurance product, if it meets, adding the insurance product to the number matching candidate set; based on the occupation matching candidate set, the region matching candidate set, the age fully compliant candidate set, the age partially compliant candidate set, and the number matching candidate set, obtaining a recommended set of insurance products for the applicant.

[0011] In some embodiments of the present disclosure, based on the occupation matching candidate set, the region matching candidate set, the age fully compliant candidate set, the age partially compliant candidate set, and the number matching candidate set, a recommended set of insurance products for the applicant is obtained, including: calculating the intersection of the occupation matching candidate set, the region matching candidate set, the age fully compliant candidate set, and the number matching candidate set, and obtaining the recommended set of insurance products according to the occupation similarity ranking; if the recommended set of insurance products is empty, calculating the intersection of the occupation matching candidate set, the region matching candidate set, the age partially compliant candidate set, and the number matching candidate set, and finally outputting the recommended set of insurance products for the applicant.

[0012] According to a second aspect of the present disclosure, an intelligent recommendation device for an insurance product is provided, including a product information extraction module, an applicant and insured information extraction module, an information processing module, and a matching recommendation module.

[0013] The product information extraction module is configured to extract key information of the insurance product from the insurance product poster by using an OCR technology and a large language model; the applicant and the insured person information extraction module is configured to extract enterprise information of the applicant and personal information of the insured person, the enterprise information including an enterprise name, a unified credit code, an industry to which the enterprise belongs, a region where the enterprise is located, and a business scope, and the personal information including an employee name, a position name, an age, a gender, and a birthplace; the information processing module is configured to standardize the position name provided by the enterprise and supplement position description information; and the matching and recommendation module is configured to perform region matching, occupation matching, age matching, and number matching on the key information of the insurance product, the enterprise information of the applicant, and the personal information of the insured person, and output an insurance product recommendation set for the applicant by merging the matching results.

[0014] According to a third aspect of the present disclosure, a computing device is provided, comprising at least one processor; and at least one memory having stored thereon a computer program. The processor executes the method for intelligent recommendation of insurance products according to the first aspect of the present disclosure.

[0015] According to a fourth aspect of the present disclosure, a computer-readable storage medium having stored thereon a computer program is provided, wherein the computer program, when executed by a processor, implements the steps of the method for intelligent recommendation of insurance products according to the first aspect of the present disclosure.

[0016] According to the method and device for intelligent recommendation of insurance products according to the embodiments of the present disclosure, key information is automatically extracted from an insurance product poster by using an OCR technology and a large language model, which can greatly reduce the time for manual input and data processing, and improve the efficiency and accuracy of information extraction; by standardizing the position name and supplementing the position description information, the problem of different enterprises using different position names or non-standard descriptions can be solved, so that the matching is more accurate; by combining the accurate screening of multi-dimensional matching and the fuzzy matching, the problem that the traditional recommendation method cannot adapt to the limitation clauses is avoided, and the appropriate insurance product can be more intelligently and accurately recommended according to the detailed information of the applicant. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments will be briefly described below. It should be noted that the drawings described below only relate to some embodiments of the present disclosure, rather than limiting the present disclosure, wherein: Figure 1 An exemplary flowchart of the method for intelligent recommendation of insurance products according to the embodiments of the present disclosure is shown; Figure 2 is a schematic block diagram of the device for intelligent recommendation of insurance products according to the embodiments of the present disclosure; Figure 3 is a schematic block diagram of the computing device according to the embodiments of the present disclosure.

[0018] It should be noted that the elements in the drawings are schematic and not drawn to scale. DETAILED DESCRIPTION

[0019] In order to make the purposes, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by a person of ordinary skill in the art without creative labor also belong to the scope of protection of the present disclosure.

[0020] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this present subject matter belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the specification and relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0021] The embodiments of the present disclosure provide more accurate recommendation solutions for complex accidental insurance products (such as employer liability insurance and group accidental insurance) by combining large language models, vector retrieval, and rule matching technologies, which not only improve the work efficiency of insurance agents / single companies, but also provide customers with insurance product selection that is more in line with personalized needs.

[0022] Figure 1 FIG. 1 is a flow diagram of an intelligent recommendation method 100 of an insurance product according to an embodiment of the present disclosure. Referring to FIG. 1, the method 100 includes the following steps. Figure 1 As shown in FIG. 1, at block S102, key information is extracted from an insurance product poster by an OCR technology and a large language model and is stored in a structured manner. Figure 1

[0023] ​The insurance product poster is usually saved in the form of a picture or a PDF file. The insurance product poster can be pre-processed first, including converting the PDF page to a picture format, denoising the picture, enhancing the contrast, adjusting the brightness, etc. Then, the OCR (Optical Character Recognition) technology is used to recognize the text in the picture to obtain text information including product introduction, coverage, premium, special provisions. After OCR recognition, the following main information can be extracted from the poster: product feature introduction: including basic description of the insurance product, guarantee content, coverage and premium information. Special provisions: including additional conditions, specific provisions, and exempted liabilities of the insurance. The text extracted by the OCR step can be lengthy and disordered, so a large language model is needed to deeply process these texts to extract structured insurance product information. The text information recognized by the OCR can be input into the large language model, and the model is guided by the prompt words to extract the key information of the product, including the product name, the insured area, the non-insured area, the insured occupation category, the non-insured occupation category, the insured age range, and the insured number range.

[0024] The specific prompt words are designed as follows: 1. Extract the product name: Example prompt word: "This is an insurance product description: {OCR recognized text content}, please give the name of the product." 2. Extract the insured area: Example prompt word: "Please refer to the format of "Jiangsu Province Suzhou Xiangcheng District, Hebei Cangzhou Dongguang County" to list the insured areas of this insurance product. If there is none, please answer no insured area list." 3. Extract the non-insured area: Example prompt word: "Please refer to the format of "Jiangsu Province Suzhou Xiangcheng District, Hebei Cangzhou Dongguang County" to list the non-insured areas of this insurance product. If there is none, please answer no non-insured area list." 4. Extract the insured occupation category: Example prompt word: "Please refer to the format of "cleaning workers, assembly workers" to list the insured occupation categories of this insurance product. If there is none, please answer no insured occupation category list." 5. Extract the non-insured occupation category: Example prompt word: "Please refer to the format of "cleaning workers, assembly workers" to list the non-insured occupation categories of this insurance product. If there is none, please answer no non-insured occupation category list." 6. Extract the insured age range: Example prompt word: "Please refer to the format of "insured age [18:65]" to explain the insured age range. If there is none, please answer no insured age limit." 7. Extract the insured number range: Example prompt: "Please refer to the 'number of insured persons [2:15]' format to indicate the range of the number of insured persons. If not, please answer that there is no limit on the number of insured persons." In some cases, the poster may not contain certain information or the information may be unclear. Large language models can return "no relevant information" or "no limit" according to pre-set rules. For example, if the insured area is not listed, return "no insured area list".

[0025] It should be noted that the embodiments of the present disclosure do not limit the specific technical methods of OCR, such as PaddleOCR, EasyOCR or qwenOCR, and do not limit the specific large language models, such as Wenxin Yiyang, Tongyi Qianwen, DeepSeek, ChatGPT or Claude.

[0026] Finally, the key information returned by the large language model is stored as structured data according to the predetermined format. Structured storage can make data more easily queried, analyzed and subsequently processed. Product information is stored in table format, and the structured format is as follows:

[0027] These structured data can be stored in various databases, such as relational databases (such as MySQL, PostgreSQL), NoSQL databases (such as MongoDB). Through this structured storage method, the extracted product key information can be quickly used for subsequent recommendation systems, data analysis and other tasks.

[0028] Subsequently, in block S104, the enterprise information of the insured person and the personal information of the insured person are extracted.

[0029] The enterprise information of the policyholder (the employing unit) usually includes the enterprise name, the unified social credit code, the industry to which the enterprise belongs, the region in which the enterprise is located, and the scope of business, etc. The policyholder provides the enterprise name in the application form, but other information may not be completely provided. First, the enterprise name of the policyholder is extracted from the form or the application file, and then the missing enterprise information is automatically supplemented through external data sources. For example, the enterprise information of the policyholder is captured from the enterprise business registration information through the API or the crawler technology of the enterprise information service website, which specifically includes the unified social credit code, the industry to which the enterprise belongs, the region in which the enterprise is located, the scope of business, etc. Among them, the unified social credit code of the enterprise is the unique identifier of each enterprise, which can be obtained by calling the API of the enterprise information service website. Through the scope of business or related business data of the enterprise, the industry category to which the enterprise belongs can be inferred or directly obtained. The registered address or the region in which the business place of the enterprise is located can be obtained through the business registration information of the enterprise. The scope of business is the business scope that the enterprise is allowed to engage in, which usually includes production, sales, service, etc., and can also be obtained by querying the business registration information or the API interface of the enterprise.

[0030] The extraction of the personal information of the insured person involves the basic information of the employee, including the employee name, the position name, the ID number, etc. The employee name and the position name are usually explicitly listed in the insurance product application form. The ID number of the employee contains rich personal information, from which the age, the gender, and the place of birth can be parsed. For example, the age of the employee can be calculated from the birth date part in the ID number. The first six digits of the Chinese ID number are the place of birth, and the seventh to fourteenth digits are the birth date (year, month, and day), according to which the age of the employee can be calculated. The 17th digit of the ID number is the gender identifier, with odd numbers representing male and even numbers representing female. The gender of the employee can be obtained by parsing this digit. The first six digits of the ID number are the administrative division code, representing the place of birth of the employee, which can be used to find the corresponding region.

[0031] Finally, the complete information of the policyholder and the insured person includes the following contents:

[0032] Then in block S106, the position name provided by the enterprise is standardized and the position description information is supplemented.

[0033] Insurance companies usually design the occupational classification table according to their risk assessment model, but the occupational names contained in the table are often self-defined, non-standardized, or even colloquial. In order to unify these occupational names into a standardized system, the insurance company occupational classification table can be compared with the national standard occupational classification table based on text similarity matching algorithms such as cosine similarity, Jaccard similarity, etc. to obtain the most matched occupational name. The occupational description information (such as occupational responsibilities, working environment, etc.) in the national standard occupational table is supplemented to the occupational classification table of the insurance company product to form a standardized occupational classification table A.

[0034] Each occupational entry in the standardized occupational classification table is constructed as a first industry text block and a first occupational text block. The industry information corresponding to each occupation can be extracted from the standardized occupational classification table A to construct the first industry text block, which includes industry categories, industry sub-categories, industry small categories, and industry ID numbers. Each occupational entry in the standardized occupational classification table A is integrated into the first occupational text block, which includes the standardized occupational name, occupational description information, the ID of the industry to which the occupation belongs, and the risk level of the occupation (usually assessed by the insurance company according to the occupational risk assessment model).

[0035] In the insurance product poster or underwriting platform, the covered occupational categories of the product or the refusal to insure certain high-risk occupations will usually be listed, so all refused occupations need to be removed from the intersection. The intersection of the first occupational text block and the covered occupational categories obtained from the insurance company product poster can be matched to filter out the insurable occupational category text block. That is, these covered occupational categories are compared with the first occupational text block A2 to obtain their intersection, generating an insurable occupational classification table C, which corresponds to the insurable occupational category text block C2 and contains the occupations and detailed information that meet the insurance company's underwriting conditions.

[0036] To further supplement the job information provided by the enterprise, the enterprise's job description information can be automatically retrieved from the recruitment website based on the job title provided by the enterprise. For example, the job title provided by the enterprise is used as a keyword to automatically search the recruitment website (such as Joblink, Zhaopin.com, etc.) to obtain more detailed occupational description information, especially for those colloquial and ambiguous job titles. These obtained standardized job description information will be supplemented to the enterprise's job description to ensure that the enterprise's job information is closer to the standardized occupational definition. For example, the job title provided by the enterprise is "Sales Manager", and "Sales Manager" is used as a keyword to search the job description in the recruitment website, such as "responsible for formulating and implementing sales targets, managing sales teams, and improving product sales", which is supplemented to the job information as the enterprise's job description.

[0037] Finally, the industry, position name and position description information of the enterprise are integrated to generate a second industry text block and a second occupation text block. For example, the second industry text block B1 contains the industry information of the enterprise, and the format is the same as the industry text block A1. Example: manufacturing industry - mechanical manufacturing - precision machinery - 001. The second occupation text block B2 includes the position name, position description and industry ID information, and the format is {position name}: {position description}, industry: {industry ID}. Example: software development engineer: responsible for developing and maintaining company software systems, industry: 001. These two text blocks can help map the specific position information of the enterprise to the standardized industry and occupation description.

[0038] Finally, in step S108, the key information of the insurance product is matched with the enterprise information of the applicant and the personal information of the insured person in terms of region, occupation, age and number of people, and the matched results are combined to output a set of insurance product recommendations for the applicant.

[0039] According to one embodiment of the present disclosure, the insurance product and the regional description of the enterprise address information and the birthplace of the insured person are standardized, and are compared level by level according to the province, city and district to determine whether the address of the enterprise and the birthplace of the insured person belong to the insurance product's underwriting region at the same time, and to filter the insurance products that can be insured to form a region matching candidate set. Among them, the region matching is to ensure that the address of the applicant and the underwriting or rejection region of the insurance product are consistent. Since the address description is omitted (such as province, city, county, etc.), the address description of the enterprise and the insurance product is first standardized. A standard province-city-district / county three-level address library can be created. Call the map service API, such as Gaode Map or Baidu Map, to perform geocoding and convert the address to a standard three-level address format: {province / municipality, city / region, county / district}. Compare according to the address level (from large to small), and allow small addresses to be matched as subsets of large addresses. For example, the enterprise address is "Hubei Huanggang City Macheng City", and after standardization, it is {Hubei Province, Huanggang City, Macheng City}, and the rejection region of the insurance product is "Hubei Province Huanggang City", and after standardization, it is {Hubei Province, Huanggang City}. At this time, the level-by-level matching is performed, and it is found that the region of the enterprise meets the rejection condition. For all insurance products, the products that can be underwritten are selected according to the above standardized address information. The birthplace of the insured person is matched in the same way. All insurance products that meet the region conditions of the enterprise and the insured person will enter the candidate set {D}, i.e. the region matching candidate set.

[0040] Based on the first industry text block and the second industry text block, the order of the industry category, the industry middle category and the industry small category is compared level by level, the different categories are matched based on the vector similarity ranking, after matching to the industry, the specific occupation description corresponding sentence vector is compared based on the insurable occupation category text block and the second occupation text block, and the occupation matching candidate set is obtained. The purpose of occupation matching is to select the appropriate occupation category for the insured person, and to consider the risk level difference of different occupations in different industries. Occupation matching is divided into two parts of industry matching and occupation name matching. Industry matching is mainly to match according to the hierarchy of industry (category, middle category, small category), and vector similarity calculation is used. First, the first industry text block A1 (containing industry name and description) constructed in step S106 is embedded respectively to generate industry classification vector {V1}. Similarly, the second industry text block B1 generates another set of vectors {V2}. Using vector similarity calculation methods such as cosine similarity or Jaccard similarity, the vector similarity of industry category, middle category and small category is compared level by level. According to the similarity from large to small, the top N most matched industries are selected, and their industry IDs are recorded.

[0041] Secondly, the occupation name is matched according to the occupation description. Occupation matching combines industry matching and occupation description matching. First, from the N industry categories, all occupation names and descriptions under the corresponding industry are obtained, and are read from the insurable occupation category text block C2. Each occupation description is converted into a sentence vector {V3} by a sentence vector generation model such as Sentence-BERT or TextCNN. The job title and description to be matched (second occupation text block B2) are converted into a sentence vector {V4}. The sentence vector {V4} is compared with the occupation descriptions in {V3} for similarity, and the K most similar occupation descriptions are obtained. According to the similarity ranking, the occupation name with the highest similarity is selected to construct the occupation matching candidate set {J}.

[0042] Age matching is mainly to judge whether the employees of the applicant meet the age requirements of the insurance product. According to the age requirements of the insurance product and the age of the applicant, the rule matching is performed to judge whether the employee age is greater than the minimum age of the product and less than the maximum age. If the age of the applicant completely meets the product requirements, the age completely meets the candidate set {L1} is added. If the age of the applicant does not completely meet the product requirements, i.e., part of the employees do not meet the requirements, but the number of overage personnel is less than a certain threshold, the product is added to the age partially meets the candidate set {L2}.

[0043] It is judged whether the number of employees of the applicant meets the lower limit requirement of the number of applicants of the insurance product. If it meets, the insurance product is added to the number matching candidate set {P}.

[0044] After the above matching is completed, based on the occupation matching candidate set, the region matching candidate set, the age full match candidate set, the age partial match candidate set and the number matching candidate set, an insurance product recommendation set for the applicant is obtained. First, the intersection of the region matching candidate set {D}, the occupation matching candidate set {J}, the age candidate set {L1} and the number candidate set {P} is calculated, sorted according to the occupation matching similarity, and an insurance product recommendation set {R} is obtained. If the insurance product recommendation set is empty, the intersection of the occupation matching candidate set {J}, the region matching candidate set {D}, the age partial match candidate set {L2} and the number matching candidate set {P} is calculated, and finally the insurance product recommendation set for the applicant is output. That is, new products that meet the conditions can be recalculated and selected by relaxing the age constraint and the like. Through the above matching process, accurate and required insurance product recommendations can be provided for each applicant, improving the accuracy of insurance products and customer satisfaction.

[0045] The embodiments of the present disclosure can realize accurate screening and fuzzy matching by combining vector retrieval and rule matching. Vector retrieval can calculate the similarity between different products and customer requirements, thereby accurately screening insurance products that meet customer requirements. At the same time, rule matching can ensure that fuzzy matching can be performed even in the case of incomplete matching, thereby expanding the scope of application.

[0046] Figure 2 is a schematic block diagram of an intelligent recommendation device for insurance products according to an embodiment of the present disclosure. As shown in Figure 2 the device 200 can be implemented through the following modules: The product information extraction module 210 can extract the key information of the insurance product from the insurance product poster through OCR technology and a large language model.

[0047] The applicant and insured person information extraction module 220 can extract the enterprise information of the applicant and the personal information of the insured person. The enterprise information includes the enterprise name, the unified credit code, the industry to which the enterprise belongs, the region where the enterprise is located and the scope of business, and the personal information includes the employee name, the position name, the age, the gender and the place of birth.

[0048] The information processing module 230 can standardize the position name provided by the enterprise and supplement the position description information.

[0049] The matching recommendation module 240 can perform region matching, occupation matching, age matching and number matching on the key information of the insurance product and the enterprise information of the applicant and the personal information of the insured person, and output an insurance product recommendation set for the applicant by merging the matching results.

[0050] The specific functions of the above modules refer to the description of the intelligent recommendation method for insurance products, and this solution will not be repeated here.

[0051] Figure 3 is a schematic block diagram of a computing device according to an embodiment of the present disclosure. Figure 3 As shown, the computing device 300 may include a processor 310 and a memory 320 storing a computer program. When the computer program is executed by the processor 310, the computing device 300 may perform the following operations: Figure 1 The steps of method 100 are shown.

[0052] In an embodiment of the present disclosure, the processor 310 may be, for example, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a processor based on a multi-core processor architecture, etc. The memory 320 may be any type of memory implemented using data storage technology, including but not limited to random access memory, read-only memory, semiconductor-based memory, flash memory, disk storage, etc.

[0053] Furthermore, in an embodiment of the present disclosure, the computing device 300 may also include an input device 330, such as a keyboard, a mouse, etc. Furthermore, the computing device 300 may also include an output device 340, such as a display, etc., for outputting a set of insurance product recommendations for the policyholder.

[0054] In other embodiments of the present disclosure, a computer-readable storage medium storing a computer program is further provided, wherein the computer program can achieve the following when executed by a processor: Figure 1 The steps of the intelligent recommendation method 100 for insurance products are shown.

[0055] In summary, according to the intelligent recommendation method and device for insurance products of the embodiments of the present disclosure, key information is automatically extracted from insurance product posters through OCR technology and large language models, which can greatly reduce the time of manual input and data processing, and improve the efficiency and accuracy of information extraction; by standardizing job titles and supplementing job description information, the problem of different companies using different job titles or non-standard descriptions can be solved, making the matching more accurate; by combining precise screening and fuzzy matching of multi-dimensional matching, the problem of traditional recommendation methods being unable to adapt to restrictive clauses is avoided, and suitable insurance products can be recommended more intelligently and accurately based on the insured's detailed information.

[0056] The diagrams of the flowcharts and block diagrams in the drawings show the architecture, functionality, and operation of possible implementations of apparatuses and methods according to various embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams can represent a module, a segment, or a portion of code which comprises one or more executable instructions for implementing the specified logical functions. In some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession can in fact be executed substantially concurrently or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by dedicated hardware-based systems which perform the specified functions or acts or combinations thereof, or can be implemented by a combination of dedicated hardware and computer instructions.

[0057] The singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a" or "the" element is

[0058] Further aspects and scope of adaptations will become apparent from the description provided herein. It should be understood that the various aspects of the present application can be practiced alone or in combination with one or more other aspects. It should also be understood that the description and specific examples herein are intended to be illustrative only and are not intended to limit the scope of the present application.

[0059] The above detailed description of several embodiments of the present disclosure has been presented for the purposes of illustration and description. It is apparent to those skilled in the art that various modifications and variations can be made to the embodiments of the present disclosure without deviating from the spirit and scope of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. An intelligent recommendation method for insurance products, characterized in that: include: Extract key information about insurance products from insurance product posters using OCR technology and large language models; Extract the policyholder's business information and the insured's personal information. The business information includes the company name, unified credit code, industry, location, and business scope. The personal information includes the employee's name, position title, age, gender, and place of birth. Standardize the job titles provided by companies and supplement job description information; as well as The key information of the insurance product is matched with the enterprise information of the policyholder and the personal information of the insured in terms of region, occupation, age and number of people, and a set of insurance product recommendations for the policyholder is output by merging the matching results.

2. The intelligent recommendation method for insurance products according to claim 1, characterized in that: The key information of insurance products extracted from insurance product posters using OCR technology and large language models includes: Preprocessing the insurance product poster, including converting the PDF page into an image format, removing noise from the image, enhancing contrast, and adjusting brightness; Use OCR technology to recognize text in images and obtain text information including product introduction, insurance amount, premium, and special terms; Input the text information recognized by OCR into the large language model, and use prompt words to guide the model to extract key information of the product, such as product name, coverage area, rejection area, coverage occupation category, rejection occupation category, coverage age range, and coverage number range; The key information returned by the large language model is stored as structured data in a predetermined format.

3. The intelligent recommendation method for insurance products according to claim 1, characterized in that: The extraction of the policyholder's business information and the insured's personal information includes: Use the API of the enterprise information service website or a crawler program to crawl the insured's unified credit code, industry, region, and business scope from the enterprise's industrial and commercial registration information; Extract employee names and job titles from the company's internal employee management system, and parse age, gender, and place of birth from employee ID numbers.

4. The intelligent recommendation method for insurance products according to claim 1, characterized in that: The standardization of job titles provided by enterprises and supplementation of job description information include: Based on a text similarity matching algorithm, the insurance company's occupational classification table is compared with the national standard occupational classification table to obtain the most matching occupational name. The occupational description information in the national standard occupational table is supplemented into the occupational classification table of the insurance company's products to form a standardized occupational classification table. Constructing each occupation entry in the standardized occupation classification table into a first industry text block and a first occupation text block; Intersection of the first occupation text block and the insured occupation categories obtained from the insurance company's product poster is taken to filter out insurable occupation category text blocks; Based on the job title provided by the enterprise, retrieve additional description information of the enterprise position from the recruitment website; and Integrate the industry to which the enterprise belongs, the job title and the job description information to generate the second industry text block and the second occupation text block.

5. The intelligent recommendation method for insurance products according to claim 4, characterized in that: The step of constructing each occupation entry in the standardized occupation classification table into a first industry text block and a first occupation text block comprises: Extracting industry information corresponding to each occupation from the standardized occupation classification table to construct a first industry text block, wherein the first industry text block includes a major industry category, a medium industry category, a minor industry category, and an industry ID number; Each occupation entry in the standardized occupation classification table is integrated into a first occupation text block, which includes a standardized occupation name, occupation description information, the ID of the industry to which the occupation belongs, and the risk level of the occupation.

6. The intelligent recommendation method for insurance products according to claim 5, characterized in that: The matching of the key information of the insurance product with the enterprise information of the policyholder and the personal information of the insured by region, occupation, age and number of people, and outputting a set of insurance product recommendations for the policyholder by combining the matching results includes: Standardize the regional descriptions of insurance products and enterprise address information, as well as the insured's birthplace, and compare them step by step at the province, city, and district levels to determine whether the enterprise's address and the insured's birthplace are both within the insurance product's coverage area. This helps screen eligible insurance products and form a regional matching candidate set. Based on the first industry text block and the second industry text block, a step-by-step comparison is performed in the order of industry major category, industry medium category, and industry minor category. Different categories are matched based on vector similarity sorting. After matching an industry, a similarity comparison is performed on the sentence vectors corresponding to the specific occupation description based on the insurable occupation category text block and the second occupation text block to obtain an occupation matching candidate set; Based on the age requirements of the insurance product and the age of the policyholder, a rule matching is performed to determine whether the employee's age is greater than the minimum age of the product and less than the maximum age. If the policyholder's age fully meets the product requirements, the policyholder is added to the fully-matched age alternative set. If the policyholder's age does not fully meet the product requirements, the policyholder is added to the partially-matched age alternative set. Determine whether the number of insured employees of the policyholder meets the minimum number of insured employees of the insurance product. If so, add the insurance product to the number of employees matching candidate set; and Based on the occupation matching alternative set, region matching alternative set, age fully matching alternative set, age partially matching alternative set and number matching alternative set, a set of insurance product recommendations for the policyholder is obtained.

7. The intelligent recommendation method for insurance products according to claim 6, characterized in that: The insurance product recommendation set for the policyholder obtained based on the occupation matching candidate set, the region matching candidate set, the age fully matching candidate set, the age partially matching candidate set, and the number of people matching candidate set includes: Calculate the intersection of the occupation matching candidate set, region matching candidate set, age matching candidate set, and number matching candidate set, and sort them by occupation similarity to obtain a set of insurance product recommendations; If the insurance product recommendation set is an empty set, the intersection of the occupation matching alternative set, the region matching alternative set, the age partial matching alternative set and the number of people matching alternative set is calculated, and finally the insurance product recommendation set for the insured is output.

8. An intelligent recommendation device for insurance products, characterized in that: include: The product information extraction module is used to extract key information about insurance products from insurance product posters using OCR technology and a large language model; The insured and policyholder information extraction module is used to extract the policyholder's corporate information and the insured's personal information. The corporate information includes the company name, unified credit code, industry, location, and business scope; the personal information includes the employee's name, position title, age, gender, and place of birth; Information processing module, used to standardize job titles provided by enterprises and supplement job description information; as well as The matching recommendation module is used to match the key information of the insurance product with the enterprise information of the policyholder and the personal information of the insured person by region, occupation, age and number of people, and output a set of insurance product recommendations for the policyholder by merging the matching results.

9. A computing device, characterized in that include: at least one processor; as well as at least one memory storing a computer program; Wherein, when the computer program is executed by the at least one processor, the computing device is caused to perform the steps of the intelligent recommendation method for insurance products according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When executed by a processor, the computer program implements the steps of the intelligent recommendation method for insurance products according to any one of claims 1 to 7.

Images