A storage method and retrieval method of reinsurance contract text data file

Abstract

The present application relates to the field of reinsurance information retrieval, and provides a storage method and a retrieval method for reinsurance contract text data files, which comprises the following steps: obtaining input target EPI data and a plurality of target business categories; performing multi-level range retrieval in a preset fusion database of reinsurance contract text historical data based on the target EPI data and the target business categories to obtain a preliminary candidate contract list; performing retrieval in the preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list, and mixing the retrieval result and the preliminary candidate contract list to obtain a mixed candidate contract list; performing retrieval in a preset full-amount database according to the mixed candidate list to obtain a supplementary candidate contract list, and merging the mixed candidate contract list and the supplementary candidate contract list to obtain a contract list. The present application greatly improves the historical contract text retrieval efficiency, and can perform second-level retrieval on target text data according to EPI and business categories.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of reinsurance information retrieval, and in particular to a storage method and retrieval method of reinsurance contract text data files. BACKGROUND

[0002] In the insurance industry, especially in the field of reinsurance, undertaking non-proportional treaty reinsurance business is a complex and delicate task. Non-proportional treaty reinsurance, also known as non-proportional reinsurance, is a reinsurance method based on claim amount or loss degree to determine reinsurance liability and reinsurance premium. The formulation of its contract terms needs to fully consider various factors such as business risk, mutual rights and interests, and industry practices. In this process, it is particularly crucial to write a detailed and compliant contract text according to the specific business situation.

[0003] However, finding and retrieving these historical contract texts is not easy. Due to the particularity of the insurance industry, the number of historical contract texts is large and the types are diverse, involving a wide range of business fields, from property insurance to personal insurance, from domestic business to international business. Therefore, to find historical contract texts that are highly matched with the current business situation, a large amount of manual search and retrieval work is often required. This process not only consumes time and effort, but is also susceptible to human factors, such as the searchers' experience level, understanding of the business, etc., which can affect the accuracy and comprehensiveness of the search results.

[0004] Currently, existing technologies mainly rely on the experience of business personnel when dealing with this task. Business personnel, based on their professional knowledge and past experience, manually search for historical contract texts in an attempt to filter out content that has direct reference value for the current business. However, this method is not only inefficient, but also highly dependent on the personal experience and judgment ability of business personnel, and has certain subjectivity and uncertainty. Different business personnel have different levels of understanding and grasp of historical contracts, which may result in uneven reference value of the filtered historical contracts, and even may introduce content that does not conform to the actual business due to personal understanding bias, increasing the risk of contracts. SUMMARY

[0005] To address the above problems, the present application proposes a storage method and retrieval method of reinsurance contract text data files, comprising:

[0006] obtaining input target epi data and a plurality of target business categories;

[0007] performing multi-level range retrieval in a pre-set reinsurance contract text historical data fusion database based on the target EPI data and target business categories to obtain a preliminary candidate contract list;

[0008] retrieve in a preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list, and mix the retrieval result and the preliminary candidate contract list to obtain a mixed candidate contract list;

[0009] retrieve in a preset full-amount database according to the mixed candidate list to obtain a supplementary candidate contract list, and combine the mixed candidate contract list and the supplementary candidate contract list to obtain a contract list.

[0010] In some embodiments, the construction process of the preset fusion database of reinsurance contract text historical data is as follows:

[0011] dividing reinsurance contract text historical data into structured data and unstructured data;

[0012] training the structured data by XGBoost to obtain a structured feature vector;

[0013] generating a BERT semantic vector by BERT from the unstructured data;

[0014] splicing the structured feature vector and the BERT semantic vector to form a fusion feature vector;

[0015] storing the fusion feature vector and a mapping relationship in a database to obtain the fusion database of reinsurance contract text historical data.

[0016] In some embodiments, the step of retrieving in a preset fusion database of reinsurance contract text historical data based on the target EPI data and the target business type to obtain a preliminary candidate contract list includes:

[0017] defining a plurality of retrieval ranges according to the target EPI data;

[0018] extracting a structured feature vector corresponding to the target business type from the fusion database of reinsurance contract text historical data;

[0019] calculating the cosine similarity of the structured feature vector and the retrieval result of different retrieval ranges as a structured feature matching degree, and screening out candidate contracts meeting the conditions based on the same;

[0020] summarizing all the candidate contracts as a preliminary candidate contract list.

[0021] In some embodiments, the step of retrieving in a preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list includes:

[0022] obtaining a target semantic vector;

[0023] extracting corresponding semantic vectors from a preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list;

[0024] calculating semantic similarity between the target semantic vector and the semantic vectors of the preliminary candidate contract list based on cosine similarity.

[0025] In some embodiments, the step of mixing the retrieval result and the preliminary candidate contract list to obtain a mixed candidate contract list includes:

[0026] adjusting the contribution proportion of the structured feature vector and the semantic vector dynamically based on the input weight;

[0027] calculating the comprehensive similarity of each candidate contract according to the contribution proportion, the structured feature matching degree and the semantic similarity;

[0028] selecting candidate contracts with a comprehensive similarity higher than a preset range as the mixed candidate contract list.

[0029] In some embodiments, the method further includes dynamically adjusting the preset range according to the click rate of user feedback.

[0030] In some embodiments, the step of retrieving in the preset full-quantity library according to the mixed candidate list to obtain a supplementary candidate contract list includes:

[0031] checking whether the mixed candidate contract list covers all target business categories;

[0032] in response to all being covered, an empty list is taken as the supplementary candidate contract list;

[0033] in response to not all being covered, retrieving in the preset full-quantity library according to the missing business categories to obtain candidate contracts as the supplementary candidate contract list.

[0034] The present application provides a reinsurance contract text historical data processing system, which includes:

[0035] an acquisition unit configured to acquire input target EPI data and a plurality of target business categories;

[0036] a preliminary retrieval unit configured to perform multi-level range retrieval in a preset fusion database of reinsurance contract text historical data based on the target EPI data and the target business categories to obtain a preliminary candidate contract list;

[0037] a mixed retrieval unit configured to perform retrieval in the preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list, and mix the retrieval result and the preliminary candidate contract list to obtain a mixed candidate contract list;

[0038] The merging unit is configured to retrieve from a preset full database according to the mixed candidate list to obtain a supplementary candidate contract list, and merge the mixed candidate contract list and the supplementary candidate contract list to obtain a contract list.

[0039] The present application provides a computer device, comprising:

[0040] The present application provides a computer device, comprising:

[0041] The present application provides a computer device, comprising:

[0042] The present application has at least the following beneficial technical effects:

[0043] The present application provides a computer device, comprising:

[0044] The present application has at least the following beneficial technical effects: BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other embodiments can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0046] Figure 1 A storage method and retrieval method flow chart of a reinsurance contract text data file are provided for the present application;

[0047] Figure 2 A storage and retrieval system module diagram of a reinsurance contract text data file is provided for the present application;

[0048] Figure 3 A preliminary retrieval module diagram of a storage and retrieval system of a reinsurance contract text data file is provided for the present application;

[0049] Figure 4 A supplementary retrieval module of a storage and retrieval system of a reinsurance contract text data file is provided for the present application Figure 1 ;

[0050] Figure 5 A supplementary retrieval module of a storage and retrieval system of a reinsurance contract text data file is provided for the present application Figure 2 ;

[0051] Figure 6 A structure schematic diagram of an embodiment of a computer device provided for the present application is provided;

[0052] Figure 7 A structure schematic diagram of an embodiment of a computer readable storage medium provided for the present application is provided. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present application clearer and more apparent, the embodiments of the present application are further described in detail below, with reference to the accompanying drawings.

[0054] It should be noted that all the expressions of "first" and "second" in the embodiments of the present application are used to distinguish two same name non-same entities or non-same parameters, and it can be seen that "first" and "second" are only for the convenience of description, and should not be understood as a limitation of the embodiments of the present application, and the subsequent embodiments will not be described one by one.

[0055] The present application provides a storage method and retrieval method of a reinsurance contract text data file, please refer to Figure 1 , comprising:

[0056] S1: obtaining input target epi (expected premium income) data and a plurality of target business categories;

[0057] S2: based on the target EPI data and the target business category, performing multi-level range retrieval in a preset reinsurance contract text historical data fusion database to obtain a preliminary candidate contract list;

[0058] S3: searching in a preset fusion database of reinsurance contract text history data based on the preliminary candidate contract list, and mixing the searching result and the preliminary candidate contract list to obtain a mixed candidate contract list;

[0059] S4: searching in a preset full-amount database according to the mixed candidate list to obtain a supplementary candidate contract list, and merging the mixed candidate contract list and the supplementary candidate contract list to obtain a contract list.

[0060] The application narrows down the candidate range quickly through preliminary searching, reduces the data amount of subsequent searching, combines the preliminary candidate contract list and the further searching result, and ensures the comprehensiveness and accuracy of the candidate contract. The full-amount database is used to supplement the candidate contract, so that no possible contract text is missed. The combination of the target EPI data and the target business type makes the searching result closer to the user demand. The combination of the multi-level range searching and the mixed searching improves the matching degree of the contract text and the target business. The method of the application can be executed on the server in general, wherein the steps of obtaining the target EPI data and the target business type can be realized by obtaining the data transmitted by the client by the server.

[0061] Through the preset database and the searching rule, the time and the labor cost of manual searching and searching are reduced. The combination of the multi-level range searching and the mixed searching makes the target contract text be positioned more quickly. The supplement of the candidate contract list ensures the comprehensiveness and diversity of the contract text. The preset fusion database of reinsurance contract text history data provides rich historical contract texts as a reference.

[0062] In some embodiments, referring to Figure 1 , the construction process of the preset fusion database of reinsurance contract text history data is as follows:

[0063] The reinsurance contract text history data is divided into structured data and unstructured data;

[0064] The structured data is trained by XGBoost to obtain a structured feature vector; XGBoost (eXtreme Gradient Boosting) is a high-efficiency machine learning algorithm based on a gradient boosting framework, which is widely used in classification, regression and sorting tasks of structured data.

[0065] The unstructured data is generated into a BERT semantic vector by BERT (Bidirectional Encoder Representations from Transformers, a pre-training language model based on a Transformer architecture);

[0066] The structured feature vector is spliced with the BERT semantic vector to form a fusion feature vector;

[0067] The fusion feature vector and the mapping relationship are stored in a database to obtain a fusion database of reinsurance contract text historical data.

[0068] The historical proportional contract text is split and stored, the EPI data is extracted from the original text, the business type data is extracted using named entity recognition, the contract text is semantically segmented using BERT, and the abstract, self-retention table, exclusion clause, and long clause are identified.

[0069] The structured data includes epi data and business type data, and the unstructured data includes contract text abstract table information, contract text self-retention table information, contract text exclusion clause part, and contract text long clause title and specific content information.

[0070] The main folder is named proportional contract text name, and the part-0, part-1, part-2, part-3, and part-4 contents are placed in the sub-folders.

[0071] In the part-0 file, the first line stores the EPI (expected premium income) information of the proportional contract text, and the second line stores the business type of the proportional contract.

[0072] The part-1 file stores the contract text abstract table information.

[0073] The part-2 file stores the contract text self-retention table information.

[0074] The part-3 file stores the contract text exclusion clause part.

[0075] The part-4 file stores the contract text long clause title and specific content information.

[0076] The reinsurance contract data is divided into structured and unstructured parts, and efficient algorithms are used for processing: structured data is trained by XGBoost feature vector, and unstructured data is generated by BERT semantic vector. This hierarchical processing method can fully utilize the advantages of the two types of algorithms and avoid the waste of computing resources caused by unified processing of full data.

[0077] By splicing the structured feature vector and the BERT semantic vector, the numerical rules (such as premium, claim ratio) and semantic associations (such as clause description, risk definition) of the contract can be captured simultaneously. The fusion feature vector breaks through the limitations of single dimension, realizes the deep association of clause content and business logic, and improves the understanding ability of the model for complex text.

[0078] The fusion feature vector and its mapping relationship are stored in the database, and the index association of structured data and unstructured data can be established. During subsequent retrieval, similarity can be directly calculated based on the feature vector, without the need to repeatedly analyze the original text, thereby significantly improving the retrieval efficiency and response speed.

[0079] The structured data extracts numerical features through XGBoost, and the unstructured data generates semantic vectors through BERT, thereby breaking through the limitation of a single data type and improving the understanding ability of the model for complex contract risks. When a new clause type is added, only the BERT model needs to be fine-tuned, without the need to reconstruct the entire feature engineering process. The traditional method relies on artificial rules or a single model, and it is difficult to handle ambiguous expressions in clauses. The fusion features capture the semantic differences of the clauses through BERT, and combine the numerical rules of XGBoost, thereby more accurately identifying high-risk contracts.

[0080] In some embodiments, referring to Figure 1 and Figure 3 , the step of performing multi-level range retrieval in the fusion database of preset reinsurance contract text historical data based on the target EPI data and the target business category to obtain a preliminary candidate contract list includes:

[0081] defining a plurality of retrieval ranges according to the target EPI data;

[0082] extracting a structured feature vector corresponding to the target business category from the fusion database of reinsurance contract text historical data;

[0083] calculating the cosine similarity of the structured feature vector and the retrieval results of different retrieval ranges as the structured feature matching degree, and screening out candidate contracts that meet the conditions based on the same;

[0084] collecting all candidate contracts as a preliminary candidate contract list.

[0085] As shown in Figure 3 , the contract folder is traversed to find the part-0 file, and the EPI and business category data of all historical texts are read. According to the target EPI data, 5-level EPI search ranges are dynamically defined, for example

(0.8, 1.2), (0.7, 1.3), (0.5, 1.5), (0.5, 1.8), (0.3, 2.0)

[0086] The preliminary candidate contract list includes contract ID, structured feature matching degree, key features, etc., and the user can intuitively understand the screening basis.

[0087] The search range is dynamically adjusted to avoid wasting time in irrelevant data. The cosine similarity can compare multiple features such as premium, claim conditions, and term, thereby avoiding the limitation of single-dimensional screening. By setting a similarity threshold of 0.8, highly matched candidate contracts are automatically screened out. The search range is dynamically adjusted based on the target EPI data, so that the data subset highly related to the current business demand can be accurately located, the efficiency loss caused by full library scanning is avoided, and the screening results are focused on the core target field. Through the dual filtering of multi-range search and feature matching, the preliminary candidate contract list has high relevance, providing a high-quality basis for subsequent in-depth analysis or manual review, and shortening the decision-making link.

[0088] In some embodiments, referring to Figure 1 the step of searching in the preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list comprises:

[0089] obtaining a target semantic vector;

[0090] extracting a corresponding semantic vector from the preset fusion database of reinsurance contract text historical data based on the preliminary candidate contract list;

[0091] calculating the semantic similarity between the target semantic vector and the semantic vector of the preliminary candidate contract list based on the cosine similarity.

[0092] Encode the user query (such as keywords, description) into a semantic vector as a target semantic vector, and calculate the cosine similarity between each candidate contract and the target semantic vector.

[0093] Through the cosine similarity calculation between the target semantic vector and the semantic vector of the candidate contract, the deep semantic association of the text can be captured, which surpasses the superficiality of keyword matching and accurately identifies implicit information such as clause wording and risk description, thereby improving the matching accuracy. In view of the characteristics of various clauses and flexible wording in reinsurance contracts, the semantic similarity calculation can adapt to different text styles, avoid missing matches due to wording differences, and is especially suitable for contract retrieval scenarios with updated clauses or cross-domain contracts.

[0094] The cosine similarity calculation converts the text matching into vector space distance measurement, has high calculation efficiency, can quickly sort the similarity of the candidate contracts, preferentially outputs high matching degree results, and reduces the cost of manual screening.

[0095] In some embodiments, referring to Figure 1 the step of mixing the search results and the preliminary candidate contract list to obtain a mixed candidate contract list comprises:

[0096] dynamically adjusting the contribution proportion of the structured feature vector and the semantic vector based on the input weight;

[0097] According to the contribution proportion, the structured feature matching degree and the semantic similarity, a comprehensive similarity of each candidate contract is calculated.

[0098] A candidate contract with a comprehensive similarity higher than a preset range is selected as a hybrid candidate contract list.

[0099] The comprehensive similarity = the contribution proportion of the structured feature vector x the structured matching degree + the contribution proportion of the semantic vector x the semantic similarity.

[0100] By dynamically adjusting the contribution proportion of the structured feature vector and the semantic vector, different business scenarios can be flexibly adapted to. In a scenario emphasizing numerical matching of clauses, the weight of the structured feature is increased; in a scenario focusing on semantic understanding of clauses, the weight of the semantic vector is increased, achieving precise adaptation.

[0101] The comprehensive similarity calculation combines the structured feature matching degree and the semantic similarity, avoiding the limitations of single-dimensional screening. The structured feature captures explicit rules, and the semantic vector mines implicit associations. The combination of the two can more comprehensively evaluate the similarity of the candidate contract and the target contract, and improve the matching accuracy.

[0102] The dynamic weight adjustment mechanism can cope with data changes or abnormal situations. When the structured feature data is missing or has a large amount of noise, the weight of the semantic vector can be automatically increased to ensure the stability and reliability of the matching result.

[0103] The candidate contract with a comprehensive similarity higher than a preset range is preferentially screened to form a hybrid candidate contract list, providing high-quality decision-making basis for users, reducing manual screening costs, and improving decision-making efficiency.

[0104] In some embodiments, referring to Figure 1 , the method further comprises dynamically adjusting the preset range according to the click rate of user feedback.

[0105] By dynamically adjusting the preset range through the user click rate, changes in user preferences can be captured in real time. If the user's click rate on a certain type of candidate contract increases, the screening range of this type of contract will be automatically expanded to ensure that the recommended results better meet the user's actual needs. The click rate feedback mechanism has self-optimization capability. When user needs change over time or market environment, the screening threshold can be automatically adjusted to ensure the timeliness and accuracy of the recommended results, avoiding matching deviations caused by fixed thresholds. The dynamic adjustment mechanism makes the screening results more targeted, reducing the user's manual screening costs. Users do not need to repeatedly adjust the screening conditions to provide candidate contracts that meet their preferences, significantly improving operational efficiency and satisfaction.

[0106] In some embodiments, referring to Figure 1 , Figure 4 and Figure 5The step of retrieving from a preset full database according to the mixed candidate list to obtain a supplementary candidate contract list comprises:

[0107] checking whether the mixed candidate contract list covers all target business categories;

[0108] in response to all being covered, an empty list is taken as the supplementary candidate contract list;

[0109] in response to not all being covered, retrieval is performed in the preset full database according to missing business categories, and candidate contracts obtained are taken as the supplementary candidate contract list.

[0110] As shown in Figure 4 and Figure 5 , it is judged whether the target business categories have been all covered in the mixed contract text list, if all are covered, the obtained contract text list is directly returned, which is the target file expected to be obtained.

[0111] If all cannot be covered, retrieval is performed in the full historical contract text database according to missing business categories, all contract texts containing the missing categories are found, and the two new contracts with the smallest difference from the target EPI value are selected as the supplementary candidate contract list and added to the target file, to obtain the historical contract text covering the target EPI and the target business categories.

[0112] The full database is a full historical contract database, through checking the coverage of the mixed candidate contract list on the target business categories, missing categories can be actively identified, missing due to limited screening range can be avoided, and it is ensured that the final candidate set covers all business requirements, and the comprehensiveness of the matching result is improved. The dynamic coverage checking and supplementing mechanism can flexibly cope with changes in business requirements. When new business categories are added or the target range is adjusted, quick adaptation is achieved without the need to reconstruct the underlying logic, and the expansion capability is enhanced. Through the perfection of the supplementary candidate contract list, users can make decisions based on a more comprehensive candidate set, reduce risks caused by missing information, and improve the credibility and effectiveness of the final selection.

[0113] The present application provides a storage and retrieval system for reinsurance contract text data files, please refer to Figure 2 , comprising:

[0114] The acquisition unit 100 is configured to acquire input target EPI data and target business categories;

[0115] The preliminary retrieval unit 200 is configured to perform multi-level range retrieval in a preset fusion database of historical reinsurance contract text data based on the target EPI data and target business categories, to obtain a preliminary candidate contract list;

[0116] The mixing searching unit 300 is configured to search in a preset fusion database of reinsurance contract text history data based on the preliminary candidate contract list, and mix the searching result and the preliminary candidate contract list to obtain a mixed candidate contract list.

[0117] The merging unit 400 is configured to search in a preset full-amount database according to the mixed candidate list to obtain a supplementary candidate contract list, and merge the mixed candidate contract list and the supplementary candidate contract list to obtain the contract list.

[0118] Each step in the present application is based on a preset database and searching rule, and has high flexibility and scalability. With the update of business data and historical contract text, the method can be continuously optimized to improve the accuracy and efficiency of searching.

[0119] Based on the same inventive concept, according to another aspect of the present application, as shown in Figure 6 The embodiment of the present application also provides a computer device 30, which comprises a processor 310 and a memory 320, and the memory 320 stores a computer program 321 which can run on the processor, and the processor executes the steps of the method as above when executing the program.

[0120] Based on the same inventive concept, according to another aspect of the present application, as shown in Figure 7 The embodiment of the present application also provides a computer readable storage medium 40, which stores a computer program 410 which is executed by a processor to execute the method as above.

[0121] The embodiment of the present application can also include a corresponding computer device. The computer device comprises a memory, at least one processor, and a computer program stored in the memory and executable on the processor, and the processor executes the program to execute any method as above.

[0122] The memory is a non-volatile computer readable storage medium, which can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules in the embodiment of the present application. The processor executes various function applications and data processing of the device by running the non-volatile software programs, instructions and modules stored in the memory, that is, implements the method as above.

[0123] The memory can include a program storage area and a data storage area. The program storage area can store an operating system, application programs required for at least one function, etc. The data storage area can store data created according to use of the device, etc. In addition, the memory can include a high-speed random access memory, and can further include a nonvolatile memory such as at least one of a magnetic disk storage device, a flash memory device, or other nonvolatile solid state memory device. In an embodiment, the memory can optionally include a memory remotely located with respect to the processor, which can be connected to the local module through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0124] Finally, it should be noted that a person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by a computer program instructing relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. The storage medium of the program can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), etc. The above-mentioned computer program embodiments can achieve the same or similar effects as the corresponding method embodiments.

[0125] Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans can implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present embodiments.

[0126] The above are exemplary embodiments disclosed in the present disclosure, but it should be noted that various changes and modifications can be made without departing from the scope of the embodiments disclosed in the present disclosure defined by the claims. The functions, steps and / or actions of the method claims in accordance with the embodiments disclosed herein need not be performed in any particular order. The above-mentioned embodiment numbers of the embodiments disclosed in the present disclosure are merely for description and do not represent the advantages and disadvantages of the embodiments. In addition, although the elements of the embodiments disclosed in the present disclosure can be described or claimed in singular form, they can also be understood as plural unless explicitly limited to singular.

[0127] It should be understood that, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising", or "includes" and / or "including" when used herein, specify the presence of stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0128] Those skilled in the art will understand that the above description is intended to be illustrative, and not restrictive, and that many other implementations of the present embodiments are possible within the spirit and scope of the present embodiments, as defined by the appended claims.

Claims

1. A method for storing and retrieving reinsurance contract text data files, characterized in that, include: Obtain the target EPI data and several target service types from the input; Based on the target EPI data and target business type, a multi-level range search is performed in the preset fusion database of reinsurance contract text historical data to obtain a preliminary candidate contract list; Based on the preliminary candidate contract list, a search is performed in the preset fusion database of reinsurance contract text historical data, and the search results are combined with the preliminary candidate contract list to obtain a hybrid candidate contract list; Based on the mixed candidate list, a supplementary candidate contract list is obtained by searching the preset full database. The mixed candidate contract list and the supplementary candidate contract list are then merged to obtain the contract list.

2. The method for storing and retrieving reinsurance contract text data files according to claim 1, characterized in that, The process of constructing the fusion database of the preset reinsurance contract text historical data is as follows: The historical data of reinsurance contract texts are divided into structured data and unstructured data; The structured data is trained using XGBoost to obtain structured feature vectors; The unstructured data is used to generate BERT semantic vectors; The structured feature vector is concatenated with the BERT semantic vector to form a fused feature vector; The fused feature vectors and mapping relationships are stored in the database to obtain a fused database of reinsurance contract text historical data.

3. The method for storing and retrieving reinsurance contract text data files according to claim 1, characterized in that, The steps for obtaining a preliminary candidate contract list by performing multi-level range searches in a pre-defined fusion database of historical reinsurance contract texts based on the target EPI data and target business types include: Several search ranges are dynamically defined based on the target EPI data; Extract structured feature vectors corresponding to the target business type from the fusion database of the historical data of the reinsurance contract text; Calculate the cosine similarity between the structured feature vector and the retrieval results of different retrieval ranges as the structured feature matching degree, and select candidate contracts that meet the conditions based on it. All candidate contracts are compiled into a preliminary candidate contract list.

4. The method for storing and retrieving reinsurance contract text data files according to claim 3, characterized in that, The steps of retrieving data from the pre-defined fusion database of historical reinsurance contract texts based on the preliminary candidate contract list include: Obtain the target semantic vector; Based on the preliminary candidate contract list, corresponding semantic vectors are extracted from a fusion database of pre-set reinsurance contract text historical data; The semantic similarity between the target semantic vector and the semantic vector of the preliminary candidate contract list is calculated based on cosine similarity.

5. The method for storing and retrieving reinsurance contract text data files according to claim 4, characterized in that, The steps for combining the search results and the preliminary candidate contract list to obtain a mixed candidate contract list include: Based on the input weights, the contribution ratio of structured feature vectors and semantic vectors is dynamically adjusted. The overall similarity of each candidate contract is calculated based on the contribution ratio, structured feature matching degree, and semantic similarity. Candidate contracts with a comprehensive similarity higher than a preset range are selected as a mixed candidate contract list.

6. The method for storing and retrieving reinsurance contract text data files according to claim 5, characterized in that, Also includes: The preset range is dynamically adjusted based on the click-through rate reported by users.

7. The method for storing and retrieving reinsurance contract text data files according to claim 1, characterized in that, The steps for obtaining a supplementary candidate contract list by searching the pre-defined full database based on the mixed candidate list include: Check whether the mixed candidate contract list covers all target business types; If all are covered, an empty list will be used as a supplementary candidate contract list. If not all services are covered, a search is performed in the pre-defined full database based on the missing service type to obtain candidate contracts as a supplementary candidate contract list.

8. A system for storing and retrieving reinsurance contract text data files, characterized in that, include: The acquisition unit is configured to acquire the input target EPI data and several target service types; The preliminary retrieval unit is configured to perform multi-level range retrieval in a preset fusion database of historical reinsurance contract texts based on the target EPI data and the target business type to obtain a preliminary candidate contract list. The hybrid retrieval unit is configured to perform retrieval based on the preliminary candidate contract list in the preset fusion database of reinsurance contract text historical data, and to mix the retrieval results with the preliminary candidate contract list to obtain a hybrid candidate contract list; The merging unit is configured to retrieve a supplementary candidate contract list from a preset full database based on the mixed candidate list, and then merge the mixed candidate contract list and the supplementary candidate contract list to obtain a contract list.

9. Computer equipment, including: At least one processor; The processor also includes a memory storing a computer program that can run on the processor, wherein the processor executes the program by performing the steps of a method for storing and retrieving a reinsurance contract text data file as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it performs the steps of the method for storing and retrieving reinsurance contract text data files as described in any one of claims 1 to 7.

Images