Historical insurance policy information structured filling method and device, medium and product

Abstract

The invention relates to the technical field of data processing, in particular to a historical insurance policy information structured filling method and device, a medium and a product, and the method comprises the steps: recognizing a business data identifier in a historical insurance policy image, and carrying out the image partitioning; performing multi-dimensional quality evaluation on each initial logic plate to obtain a corresponding plate quality score, and performing image preprocessing on each initial logic plate based on the plate quality score of each initial logic plate to obtain a respective corresponding target logic plate; based on the target business data identifier and the plate quality score of each target logic plate, determining a corresponding associated logic plate; performing field filling based on each target logic plate and the corresponding associated logic plate to obtain a corresponding structured information plate; and combining all the structured information plates to obtain target structured insurance policy information corresponding to the historical insurance policy image. According to the invention, the accuracy and speed of the historical insurance policy information in the structured filling process can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, device, medium, and product for structuring and filling historical insurance policy information. Background Technology

[0002] In the process of digital transformation in the insurance industry, "paperless process" has become one of the core trends in industry development. However, before the implementation of the "paperless process" policy, the insurance industry may have accumulated a large number of historical paper policies. These historical paper policies carry core data such as customers' insurance information, coverage responsibilities, and payment records. They are an important part of customers' assets and a key basis for insurance companies to carry out policy maintenance business such as renewal payment, information change, and claims.

[0003] With the continuous expansion of the insurance business and the upgrading of customer service needs, the demand for digital management of old insurance policies is becoming increasingly urgent. In existing technological solutions, the industry generally uses manual data entry for the input and digital processing of information on old policies. This method has significant efficiency and accuracy deficiencies: on the one hand, the input efficiency is extremely low, especially when dealing with a massive amount of old policies. Manual data entry typically requires a large amount of manpower and time, making it unsuitable for large-scale digitization. On the other hand, the accuracy of manual data entry may be difficult to guarantee. Prolonged repetitive data entry can easily lead to visual fatigue and operational errors among data entry personnel, resulting in problems such as incorrect policy numbers, discrepancies in insured amounts, and misspelled policyholder names. These errors directly affect subsequent policy maintenance business and may even lead to insurance contract disputes, harming the interests of both customers and insurance companies. Summary of the Invention

[0004] To improve the accuracy and speed of structured population of historical policy information, this application provides a method, device, medium, and product for structured population of historical policy information.

[0005] Firstly, this application provides a method for structured filling of historical insurance policy information, employing the following technical solution: A method for structured population of historical insurance policy information includes: Identify business data identifiers in historical policy images, and partition the historical policy images based on the business data identifiers to obtain multiple initial logical blocks, each with a different business data identifier. Based on preset evaluation dimensions, a multi-dimensional quality evaluation is performed on each initial logical block to obtain the block quality score corresponding to each initial logical block. Based on the block quality score of each initial logical block, image preprocessing is performed on each initial logical block to obtain the corresponding target logical block. The preset evaluation dimensions include character distinguishability, background interference degree, and field integrity. Based on the target business data identifier and segment quality score corresponding to each target logical segment, determine the associated logical segments corresponding to each target logical segment; Based on each target logical module and its corresponding associated logical module, the fields of the structured information template corresponding to each target logical module are filled to obtain the structured information module corresponding to each target logical module. By combining the structured information blocks corresponding to all target logical blocks, the target structured policy information corresponding to the historical policy image is obtained.

[0006] By adopting the above technical solution, business data identifiers in historical policy images are identified and divided into multiple initial logical blocks. This avoids confusion between different business fields and improves the accuracy of field positioning from the source. Each initial logical block is assessed and scored based on three dimensions: character distinguishability, background interference level, and field completeness. Targeted image preprocessing is then performed based on the scores, improving preprocessing accuracy and avoiding indiscriminate processing, thus optimizing preprocessing efficiency. By matching the target business data identifier and the block quality score of the target logical block with the corresponding associated logical blocks, and using the dual data from the target and associated logical blocks, the structured information template is filled with fields. This effectively avoids filling errors caused by quality defects in the target logical block such as blurred characters, background interference, and missing fields, thereby improving the filling speed. Finally, all structured information blocks are automatically combined into the final structured policy information, replacing manual integration and achieving end-to-end automated output from image to structured data, significantly improving processing speed.

[0007] In one possible implementation, based on the target business data identifier and the segment quality score corresponding to the target logical segment, the associated logical segments corresponding to the target logical segment are determined, including: Identify the multi-dimensional segment feature parameters corresponding to the target logical segment, and determine the associated filtering value corresponding to the target logical segment based on the multi-dimensional segment feature parameters and the segment quality score; Based on the target business data identifier corresponding to the target logic module, the logic module to be examined is determined from other logic modules. The logic module to be examined is other logic modules whose corresponding business data identifier has an identifier similarity to the target business data identifier. Based on the association filtering value and the identifier similarity corresponding to each examined logic block, the associated logic block corresponding to the target logic block is determined from the examined logic blocks.

[0008] By adopting the above technical solution, the multi-dimensional feature parameters corresponding to the target logical module are identified and the correlation screening value is determined. Combined with the target business data identifier, the logical module to be examined is selected from other logical modules. Then, the related logical modules are determined based on the correlation screening value and the similarity of the identifier. This facilitates the accurate identification of the logical modules to be examined that are highly related to the target logical module, avoids the blindness of the related module matching, and improves the matching accuracy of the related logical modules from the source. Through the dual-dimensional screening of multi-dimensional feature parameters and identifier similarity, the interference of irrelevant or low-related logical modules is effectively eliminated, ensuring the business relevance and data compatibility between the related logical modules and the target logical module. This provides reliable dual data support for subsequent field filling, further avoids filling errors caused by improper matching of related modules, and eliminates the need to compare all other logical modules one by one, reducing invalid matching operations, optimizing the efficiency of determining related modules, and thus helping to improve the speed of overall field filling and the digital processing of historical insurance policies.

[0009] In one possible implementation, determining the association filtering value corresponding to the target logical segment based on the multi-segment feature parameters and the segment quality score includes: Based on the multi-segment feature parameters, the segment position of the target logical segment in the historical policy image is determined, and based on the preset segment position mapping relationship and the segment position, the segment weight corresponding to the target logical segment is determined. The preset segment position mapping relationship is the correspondence between segment position and segment weight. Based on the multi-segment feature parameters, the segment data volume and segment data type contained in the target logical segment are determined. Based on the preset data mapping relationship, the segment data volume, and the segment data type, the data weight corresponding to the target logical segment is determined. The preset data mapping relationship is the correspondence between the parameter combination of segment data volume and segment data type and the data weight. Based on the multi-segment feature parameters, the fuzzy positions and fuzzy regions contained in the target logic segment are determined. Based on the preset fuzzy mapping relationship, the fuzzy positions and the fuzzy regions, the fuzzy weights corresponding to the target logic segment are determined. The preset fuzzy mapping relationship is the correspondence between the parameter combination of fuzzy positions and fuzzy regions and the fuzzy weights. Based on the sector weight, the data weight, the fuzzy weight, and the sector quality score, the association filtering value is determined.

[0010] By adopting the above technical solution, and through the hierarchical weight design of segment position, data characteristics, and fuzzy state, it is easy to accurately characterize the core attributes of the target logical segment. This allows the association screening value to truly reflect the matching needs of the target logical segment for related segments, avoiding screening bias caused by a single parameter. Furthermore, it relies on a preset mapping relationship to achieve automated weight calculation without manual intervention, significantly improving the efficiency of determining the association screening value. At the same time, incorporating the segment quality score into the calculation makes it easier to strongly bind the association screening value with the quality status of the target logical segment. This ensures that the subsequent association logical segments matched based on the screening value can accurately adapt to the quality defects of the target logical segment, further improving the accuracy of historical policy structure filling.

[0011] In one possible implementation, based on the target logical module and its corresponding associated logical modules, the structured information template corresponding to the target logical module is populated with fields, including: Based on the fill identifier of the fill position in the structured information template, the corresponding original fill field and the original fuzzy parameter corresponding to the original fill field are identified from the target logical block. The original fuzzy parameter includes the character integrity of the original fill field and the degree of background interference of the original character. Based on the original fuzzy parameters corresponding to the original fill field, determine the original fuzzy score of the original fill field; Based on the original fuzzy score of the original fill field, the initial field to be filled corresponding to the position to be filled is determined. The initial field to be filled is then converted according to the format of the position to be filled to obtain the target field to be filled. Based on the target field to be filled, the field to be filled in the structured information template corresponding to the target logical block is filled.

[0012] By adopting the above technical solution, the original fields to be filled and the original fuzzy parameters containing character integrity and background interference are identified from the target logical block by using the fill-in identifier based on the structured information template. The original fuzzy score is then calculated, and the initial fields to be filled are determined accordingly, and the format conversion and filling are completed. This facilitates the accurate identification of quality defects in the original fields to be filled, avoids errors caused by directly using fuzzy fields for filling, and ensures the accuracy of the filled data. Furthermore, by relying on the targeted conversion of the format to be filled, the format requirements of the fields to be filled are fully matched with those of the structured template, eliminating obstacles to subsequent data application caused by inconsistent formats. At the same time, the entire process is automated, without the need for manual identification of fuzzy fields and adjustment of formats, greatly reducing manual intervention and improving the overall field filling speed, providing reliable support for the efficient digitization of historical insurance policies.

[0013] In one possible implementation, determining the initial field to be filled corresponding to the position to be filled based on the original fuzzy score of the original fill field includes: When the original fuzzy score of the original fill field is lower than the preset simulated score threshold, the original fill field is determined as the fill field to be examined, and based on the fill identifier corresponding to the fill field to be examined, the associated fill field corresponding to the fill field to be examined is identified from each associated logic block. Based on the associated section weight, associated data weight, and associated fuzzy weight corresponding to each associated logical section, the section priority of each associated logical section is determined; The similarity of all associated fields is fused to obtain the field similarity. When the field similarity is higher than the preset similarity threshold, the associated field corresponding to the associated logical section with the highest section priority is determined as the initial field to be filled.

[0014] By adopting the above technical solution, the low-quality fields that need to be optimized are accurately located by using fuzzy score thresholds, avoiding errors caused by directly filling fuzzy data. Furthermore, the priority of the fields is determined by multi-dimensional evaluation of related sector weights, data weights, and fuzzy weights, ensuring that the selected related fields have high reliability. At the same time, the matching degree between the related fields and the positions to be filled is further guaranteed by similarity fusion verification, which greatly improves the accuracy of field filling.

[0015] In one possible implementation, the method further includes: Target logical blocks with a quality score lower than a preset quality threshold are identified as blocks to be verified, and the target fields to be filled corresponding to the blocks to be verified are identified. The target verification dimension and the dimension weight of each target verification dimension are determined based on the quality score of the segment to be verified, and the verification credibility is calculated based on the verification information and dimension weight corresponding to each target verification dimension. The target field to be filled is compared with the corresponding original field to be filled to determine the field difference. Based on the verification credibility and the field difference, the audit type of the target field to be filled is determined and audited. The audit type includes direct entry into the database, random inspection, and manual review.

[0016] By adopting the above technical solution, target logical modules with quality scores below a preset threshold are designated as modules to be verified. This facilitates precise targeted verification of low-quality modules and avoids the waste of efficiency caused by indiscriminate review of all modules. By matching the verification dimensions and weights with the module quality scores, the verification content is made more relevant to the module's quality defects, thereby improving the accuracy of verification credibility. At the same time, based on the verification credibility and field difference, three types of review are divided: direct entry into the database, sampling inspection, and manual review. This ensures efficient entry of high-credibility and low-difference fields into the database, while controlling risky fields through sampling inspection and manual review, significantly reducing the probability of erroneous data flowing into the business system.

[0017] In one possible implementation, after obtaining the target structured policy information corresponding to the historical policy image, the method further includes: The policy status, validity period, and near-payment period of the structured policy information are detected. Based on the policy status validity period and the near-payment period, a renewal reminder corresponding to the target structured policy information is generated, and the renewal reminder is sent to the policyholder corresponding to the structured policy information.

[0018] By adopting the above technical solution, after obtaining the structured policy information corresponding to historical policies, the system detects the policy's validity period and the approaching payment period, generates corresponding renewal reminders, and sends them back to the policyholder. This achieves seamless integration between the digitization of historical policies and renewal services. Furthermore, by automating the detection of policy validity and payment periods, the system replaces the tedious process of manually checking and notifying each policy individually, significantly improving the timeliness and accuracy of renewal reminders and preventing policyholders from missing payment periods or policies from lapsed due to human oversight.

[0019] Secondly, this application provides an electronic device that adopts the following technical solution: An electronic device comprising: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, the at least one application being configured to: execute the above-described historical policy information structured population method.

[0020] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium includes: a computer program that can be loaded by a processor and execute the above-described method for structuring and filling in historical policy information.

[0021] Fourthly, this application provides a computer program product, which adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the above-mentioned method for structured filling of historical policy information.

[0022] In summary, this application includes at least one of the following beneficial technical effects: By identifying business data identifiers in historical policy images, they are divided into multiple initial logical modules to avoid confusion between different business fields and improve the accuracy of field positioning from the source. Each initial logical module is assessed and scored based on three dimensions: character distinguishability, background interference level, and field completeness. Targeted image preprocessing is then performed based on the scores, improving preprocessing accuracy and avoiding indiscriminate processing, thus optimizing preprocessing efficiency. By matching the target business data identifier and module quality score of the target logical module with corresponding associated logical modules, and using the dual data from the target and associated logical modules, the structured information template is filled with fields. This effectively avoids filling errors caused by quality defects in the target logical module such as blurred characters, background interference, and missing fields, thereby improving the filling speed. Finally, all structured information modules are automatically combined into the final structured policy information, replacing manual integration and achieving end-to-end automated output from images to structured data, significantly improving processing speed.

[0023] By defining target logical modules with quality scores below a preset threshold as modules to be verified, it is easier to achieve precise targeted verification of low-quality modules and avoid the inefficiency caused by indiscriminate review of all modules. By matching the verification dimensions and weights with the module quality scores, the verification content can be more closely aligned with the module's quality defects, thereby improving the accuracy of verification credibility. At the same time, based on the verification credibility and field difference, three types of review are divided: direct entry into the database, sampling inspection, and manual review. This ensures efficient entry of high-credibility and low-difference fields into the database, while controlling risky fields through sampling inspection and manual review, significantly reducing the probability of erroneous data flowing into the business system. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a method for structured filling of historical policy information in an embodiment of this application; Figure 2 This is a schematic diagram of an audit marking process in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0025] The following is in conjunction with the appendix Figures 1 to 3 This application will be described in further detail.

[0026] After reading this specification, those skilled in the art may make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] It should be noted that, in the optional embodiments of this application, the data related to object information, when applied to specific products or technologies, requires the permission or consent of the object. Furthermore, the collection, use, and processing of this data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of this application involve data related to an object, it must be obtained with the object's authorization and consent, the authorization and consent of relevant departments, and in accordance with the relevant laws, regulations, and standards of the country and region. If the embodiments involve personal information, the acquisition of all personal information requires the individual's consent. If sensitive information is involved, the separate consent of the information subject is required. The embodiments also need to be implemented with the object's authorization and consent.

[0029] Specifically, this application provides a method for structured population of historical insurance policy information, executed by an electronic device. This electronic device can be a server or a terminal device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this.

[0030] refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for structured filling of historical insurance policy information in an embodiment of this application. The method includes steps S110-S150, wherein: Step S110: Identify the business data identifier in the historical policy image, and partition the historical policy image based on the business data identifier to obtain multiple initial logical blocks, each with a different business data identifier.

[0031] Specifically, images corresponding to old paper-based insurance policies signed before the implementation of the paperless process policy are considered historical policy images. These images can be used to identify business data identifiers based on preset identifier granularity and preset feature recognition algorithms. Business data identifiers can include policy validity period, policyholder information, insured information, premium amount, and payment information. The preset identifier granularity is used to limit the level of business data identifiers and is related to the initial logical module granularity. This granularity can be pre-uploaded to electronic devices by relevant staff according to actual needs. For example, if it is necessary to directly generate large-granularity logical modules such as basic information, policyholder information, and payment information after image partitioning, the preset identifier granularity is generally larger (avoiding overly fine field-level identifiers, such as policy number or premium amount). Instead, focus should be placed on module-level feature identifiers. These identifiers can directly anchor the boundaries and attributes of the entire business module, avoiding the appearance of scattered small field modules after partitioning.

[0032] After partitioning the historical policy image based on the identified business data identifiers, multiple initial logical blocks can be obtained, each with a different business data identifier. For single-identifier block partitioning (no adjacent identifiers): if a certain business data identifier has no adjacent identifiers in the historical policy image (such as an isolated "premium amount" identifier), the blocks are partitioned according to a preset catch-all boundary rule. The preset catch-all boundary rule is: taking the pixel coordinate range of the business data identifier as the core, expanding outward by 10%-20% pixels to form a rectangular area, which is the initial logical block corresponding to the business data identifier, such as the "premium amount initial logical block". For multi-identifier segmentation (with adjacent identifiers): If multiple business data identifiers exist and are adjacent in position, the segments are divided according to the principle of prioritizing explicit boundaries and using relative positional boundaries as a fallback. That is, first check whether there are explicit boundaries (such as separator lines) between adjacent business data identifiers. If they exist, the image area corresponding to each business data identifier is divided according to the explicit boundary. If there are no explicit boundaries, the distance between the geometric centers of two adjacent business data identifiers is calculated, and two non-overlapping rectangular areas are divided according to the perpendicular bisector of the distance, which serve as the initial logical segments corresponding to the two business data identifiers. Regarding special area handling (handwritten annotations / additional clauses): If the area corresponding to a business data identifier contains handwritten annotations or additional clauses, and the annotation content belongs to the business scope of that business data identifier, the annotation area is included in the corresponding initial logical segment. If the annotation belongs to independent business content and has a dedicated identifier, it is divided into a new initial logical segment separately. The method of image partitioning of historical policy images based on business data identifiers is not specifically limited in this application embodiment, as long as it can ensure that the business data identifiers corresponding to each initial logical segment are different.

[0033] Step S120: Perform multi-dimensional quality assessment on each initial logical block based on preset evaluation dimensions to obtain the block quality score corresponding to each initial logical block, and perform image preprocessing on each initial logical block based on the block quality score of each initial logical block to obtain the corresponding target logical block. The preset evaluation dimensions include character distinguishability, background interference degree and field integrity.

[0034] Specifically, the preset evaluation dimensions can be uploaded to the electronic device in advance by relevant staff according to the actual evaluation needs. Character distinguishability is used to evaluate the clarity of character edges, stroke completeness, and character-background contrast of the fields recorded in the initial logic block. Background interference level is used to evaluate the background noise density, stain coverage area ratio, and texture interference intensity in the initial logic block. Field integrity is used to evaluate the regional integrity and key information missing ratio of the fields in the initial logic block.

[0035] For any initial logic block, a multi-dimensional quality assessment is performed based on preset evaluation dimensions to obtain the dimension score for each preset evaluation dimension. Finally, the quality score of the initial logic block is obtained by summing all the dimension scores. Different preset evaluation dimensions correspond to different scoring rules. For example, the scoring rule for character discernibility is as follows: Gradient operators (such as the Sobel operator) are used to calculate the character edge gradient value. A gradient value ≥ 80 indicates clear character (90-100 points); a gradient value of 50-79 indicates blurred character (50-89 points); and a gradient value < 50 indicates severely blurred character (0-49 points). The scoring rule for background interference is as follows: Noise density is calculated using a median filter and a difference method. Noise density < 10% indicates low interference (90-100 points); 10%-30% indicates medium interference (50-89 points); and > 30% indicates high interference (0-49 points). The scoring rules for field completeness are as follows: A field with no truncation and a complete field area receives full marks (100 points); 20 points are deducted for every 10% increase in the truncation rate; and 15 points are deducted for every 5% increase in the missing rate of key information (such as policy number or amount). The scoring rules for each preset evaluation dimension can also be uploaded to an electronic device in advance.

[0036] After determining the initial quality score of the logical blocks, the corresponding image preprocessing strategy can be determined based on the preset processing strategy mapping relationship. Different quality scores correspond to different image preprocessing strategies. For example, when the quality score is 80-100 (high-quality block), it indicates that the characters in the logical block are clear, the background is clean, and the fields are complete, requiring no complex processing. The corresponding image preprocessing strategy can be to perform only mild grayscale conversion and normalization. When the quality score is 50-79 (medium-quality block), it indicates that the logical block is slightly blurred and has a small amount of background noise. Targeted optimization of character contrast is needed, and the corresponding image preprocessing strategy can be to perform adaptive binarization and mild noise reduction. When the quality score is 0-49 (low-quality block), it indicates that the logical block is severely blurred, has high background interference, and incomplete fields. A combination of strategies is used to improve quality, and the corresponding image preprocessing strategy can be to perform multi-scale sharpening, morphological noise reduction, and field completion and repair. The preset processing strategy mapping relationship is the correspondence between the block quality score and the image preprocessing strategy. It can be determined by relevant personnel based on historical experimental data and then uploaded to the electronic device.

[0037] The above method can determine the image preprocessing strategy for each initial logical block. After image preprocessing using the corresponding strategy, the target logical block for each initial logical block can be obtained. Quality assessment and score calculation are performed on the initial logical blocks, and then targeted image preprocessing is implemented based on the scores. This approach not only improves preprocessing accuracy but also avoids indiscriminate processing, optimizing preprocessing efficiency.

[0038] Step S130: Based on the target business data identifier and segment quality score corresponding to each target logical segment, determine the associated logical segment corresponding to each target logical segment.

[0039] Specifically, the associated logical modules are other logical modules that have similar or related target business data representations to the target logical module. Due to common quality defects in older insurance policies, such as yellowing, stains, non-standard formatting, and blurred characters, relying solely on the target logical module's own identification data for filling in errors or omissions is highly likely. The core value of the associated logical modules is to provide dual protection for the target logical module through data verification, supplementation, and correction. Furthermore, to improve the accuracy of determining associated logical modules, the method provided in this application, when determining the associated logical modules corresponding to the target logical module based on the target business data identifier and module quality score, may specifically include: Identify the multi-dimensional segment feature parameters corresponding to the target logical segment, and determine the association screening value corresponding to the target logical segment based on the multi-dimensional segment feature parameters and segment quality score; based on the target business data identifier corresponding to the target logical segment, determine the logical segment to be examined from other logical segments, and the logical segment to be examined is other logical segments whose corresponding business data identifier has an identifier similarity to the target business data identifier; based on the association screening value and the identifier similarity corresponding to each logical segment to be examined, determine the associated logical segment corresponding to the target logical segment from the logical segments to be examined.

[0040] Specifically, for any target logical block, a preset feature recognition algorithm can be used to identify corresponding multi-dimensional block feature parameters from the target logical block. These multi-dimensional block feature parameters include, but are not limited to, block position feature parameters, used to describe the geometric position attributes of the target logical block in historical policy images; block data feature parameters, used to characterize the data attributes carried within the target logical block; and block fuzzy state feature parameters, used to quantify the image quality defect status of the target logical block. Determining the association screening value corresponding to the target logical block based on the multi-dimensional block feature parameters and block quality scores essentially involves constructing a multi-dimensional weighted quantitative screening criterion. This criterion accurately reflects the matching needs and priority requirements of the target logical block for associated logical blocks, thereby achieving accurate screening and efficient matching of associated logical blocks. Specifically, determining the association screening value corresponding to the target logical block based on the multi-dimensional block feature parameters and block quality scores can include: The system determines the location of the target logical segment in the historical policy image based on multi-segment feature parameters. Based on a preset segment location mapping relationship and the segment location, it determines the corresponding segment weight. The preset segment location mapping relationship is the correspondence between segment location and segment weight. The system also determines the segment data volume and data type contained in the target logical segment based on multi-segment feature parameters. Based on a preset data mapping relationship, the segment data volume, and the segment data type, it determines the corresponding data weight. The preset data mapping relationship is the correspondence between the parameter combination of segment data volume and segment data type and the data weight. Furthermore, the system determines the proportion of fuzzy locations and fuzzy regions contained in the target logical segment based on multi-segment feature parameters. Based on a preset fuzzy mapping relationship, the fuzzy locations, and the fuzzy region proportion, it determines the corresponding fuzzy weight. The preset fuzzy mapping relationship is the correspondence between the parameter combination of fuzzy locations and the fuzzy region proportion and the fuzzy weight. Finally, based on the segment weight, data weight, fuzzy weight, and segment quality score, it determines the association filtering value.

[0041] Specifically, the position of target logic modules in historical policy images may follow industry-standard layout patterns. The positions of key business modules are fixed, and their positional attributes directly reflect their core importance. Therefore, it is necessary to quantify the position of target logic modules in historical policy images based on a pre-defined module position mapping relationship to obtain the corresponding module weight. For example, the top / upper part of the policy typically contains core identifiers such as the policy number, application date, and insurance type. These modules are crucial for the policy's uniqueness and business attributes, and their data accuracy directly determines the effectiveness of subsequent filling and verification. The bottom / edge areas of the policy, on the other hand, often contain auxiliary modules such as remarks and agent information, which have a lower impact on the core business logic. Without distinguishing module position weights, invalid matching of "auxiliary modules with core related modules" may occur (e.g., matching "remarks module" with "premium amount module"). The pre-defined module position mapping relationship is the correspondence between module position and module weight, which can be determined by relevant personnel based on historical experimental data and uploaded to electronic devices in advance.

[0042] The amount and type of data contained in the target logic module are directly related to business value. The richness of the data and the coreness of the data types determine the supporting role of the target logic module in filling the policy structure. Therefore, it is necessary to quantify the amount and type of data in the target logic module based on a preset data mapping relationship to obtain the corresponding data weight. For example, the policyholder information module of historical policies usually contains multiple core data such as policyholder name, ID number, contact information, and address, with a data volume of 5-8 items, and the data type is core business data for identity authentication. The data in this type of module is directly related to the effectiveness of subsequent key businesses such as policyholder identity verification and renewal reminder delivery, and is a core component of the policy structure data. On the other hand, the remarks and explanations module may only contain 1-2 items of auxiliary explanatory data, with a data type of unstructured text, and its supporting role in the core business logic of the policy is weak. If the amount and type of data in the target logic module are not distinguished, there may be invalid situations where "auxiliary text modules match core amount modules", which will lead to a significant decrease in the reliability of matching related logic modules. The preset data mapping relationship is the correspondence between the parameter combination of the data volume and data type of the sector and the data weight. The specific relationship can be determined by relevant staff based on historical experimental data and the priority rules of business data in the insurance industry, and then uploaded to electronic devices in advance.

[0043] The proportion of fuzzy locations and fuzzy regions within a target logical block directly reflects the degree of image quality defects within that block. The core nature of the fuzzy location and the coverage of the fuzzy region determine the reliability of the target logical block's own data and its dependence on related logical blocks. Therefore, it is necessary to quantify the proportion of fuzzy locations and fuzzy regions within the target logical block based on a pre-defined fuzzy mapping relationship to obtain the corresponding fuzzy weights. For example, if the premium amount block of historical insurance policies (the core data area) exhibits localized fuzziness, and the fuzzy location is concentrated in the amount number area (the core location), with the fuzzy region accounting for more than 30%, the core values ​​cannot be accurately identified. In this case, the reliability of the data within that block is extremely low, requiring a high degree of reliance on data from related logical blocks (such as the insurance type block and the payment period block) for supplementation and correction. Conversely, if the fuzzy location is merely a blank area at the edge of the block (a non-core location), with the fuzzy region accounting for less than 10%, the impact on the identification of the core data is minimal, the reliability of the data within the block is high, and its dependence on related logical blocks is low. Without distinguishing the weights of fuzzy location and fuzzy region proportions, it's possible to have the same weight for "slightly fuzzy core locations and heavily fuzzy edge locations," leading to inaccurate matching of the associated screening values ​​to the defect repair needs of the relevant sectors, thus affecting the targeted matching of the associated logic sectors. The preset fuzzy mapping relationship is the correspondence between the parameter combination of fuzzy location and fuzzy region proportion and the fuzzy weight. This can be determined by relevant staff based on historical policy damage cases, core data location distribution patterns, and image recognition accuracy requirements, and then uploaded to electronic devices in advance. The fuzzy weight ranges from 0 to 1; a higher value indicates a more severe fuzzy defect in the sector, lower data reliability, and a higher dependence on the associated logic sectors, requiring priority matching of high-quality associated sectors.

[0044] Finally, the weighted average of the target logic segment's segment weight, data weight, and fuzzy weight, along with the segment quality score, is summed to obtain the association screening value for the target logic segment. Based on the above method, the association screening value for each target logic segment can be obtained.

[0045] The target business data identifier of the target logical module is compared with the business data identifiers of other logical modules to obtain the identifier similarity between the target logical module and any other logical module. Other logical modules with identifier similarity higher than a preset similarity limit are identified as the logical modules under consideration. Further filtering of the logical modules under consideration is then performed based on the determined association screening value to obtain the associated logical modules of the target logical module. The associated logical modules are the logical modules under consideration whose identifier similarity is higher than the association screening value, which is higher than the preset similarity limit. The preset similarity limit is used for preliminary filtering of other logical modules; the specific value is not specifically limited in this embodiment and can be determined by relevant personnel based on historical experimental data and then uploaded to an electronic device.

[0046] Step S140: Based on each target logical module and its corresponding associated logical module, fill in the fields of the structured information template corresponding to each target logical module to obtain the structured information module corresponding to each target logical module.

[0047] Specifically, based on the dual data support of the target logical module and its corresponding related logical modules, precise field filling is performed on the structured information template corresponding to each target logical module. The core approach is to address the field recognition errors and missing information caused by image quality defects in older insurance policies through a method of "prioritizing the extraction of target module data + supplementing and verifying data from related modules." This ultimately generates accurate and formatted structured information modules that meet the needs of subsequent digital management and business applications of insurance policies. The structured information templates are standardized templates pre-built based on insurance industry business specifications. Each target logical module corresponds to a unique structured information template, which contains fields to be filled that match the business attributes of the corresponding target logical module, as well as unified format requirements, such as date field format and decimal places for amount fields. These templates can be uploaded to electronic devices in advance. Furthermore, to improve the overall field filling speed, when filling fields on the structured information template corresponding to the target logical module based on the target logical module and its corresponding related logical modules, the specific steps may include: Based on the fill identifier of the position to be filled in the structured information template, the corresponding original fill field and the original fuzzy parameter corresponding to the original fill field are identified from the target logical block. The original fuzzy parameter includes the character integrity of the original fill field and the degree of background interference of the original character. Based on the original fuzzy parameter corresponding to the original fill field, the original fuzzy score of the original fill field is determined. Based on the original fuzzy score of the original fill field, the initial fill field corresponding to the position to be filled is determined. The format of the initial fill field is converted according to the fill format corresponding to the position to be filled to obtain the target fill field. Based on the target fill field, the field is filled in the position to be filled in the structured information template corresponding to the target logical block.

[0048] Specifically, for any target logical module, the structured information template corresponding to the target logical module may contain multiple positions to be filled. First, based on a preset feature recognition algorithm, the identifier for each position to be filled can be identified from the structured information template. This identifier can be considered a unique business attribute label and data matching anchor for the position to be filled. In the field filling process, by identifying the identifier for the position to be filled, fields consistent with the business attributes to be filled can be extracted from the target logical module and related logical modules for filling. For any position to be filled, the original filling field corresponding to the identifier can be identified from the target logical module, and the character integrity and background interference level of the original filling field can be identified. The original filling field is the field in the target logical module corresponding to the identifier to be filled. The character integrity of the original filling field is used to characterize the stroke integrity of the characters within the original filling field and whether there are missing / blurred / erroneous characters. The core of the identification is standard character template comparison + stroke feature extraction. The background interference level of the original characters is used to characterize the interference intensity of background noise such as stains, specks, and yellowing paper within the original filling field area. The core of the identification is background noise extraction + interference ratio quantification. The core is to extract and compare the pixel features and character shapes of the original fill field area. The specific method is not limited in this application embodiment.

[0049] By normalizing and quantizing the original fuzzy parameters of the original fill field, and summing the results, the original fuzzy score of the original fill field can be obtained. The original fuzzy score is used to evaluate the reliability of the original fill field. Based on the original fuzzy score of the original fill field, the initial field to be filled corresponding to the position to be filled is determined. When the original fuzzy score of the original fill field is not lower than a preset simulated score threshold, it indicates that the original fill field has high reliability. In this case, the original fill field can be directly determined as the initial field to be filled at the position to be filled. The specific preset simulated score threshold is not specifically limited in this embodiment.

[0050] When the original fuzzy score of the original fill field is lower than the preset simulated score threshold, the original fill field is determined as the fill field to be examined. Based on the fill identifier corresponding to the fill field to be examined, the associated fill field corresponding to the fill field to be examined is identified from each associated logical block. According to the associated block weight, associated data weight and associated fuzzy weight corresponding to each associated logical block, the block priority of each associated logical block is determined. The similarity of all associated fill fields is fused to obtain the field similarity. When the field similarity is higher than the preset similarity threshold, the associated fill field corresponding to the associated logical block with the highest block priority is determined as the initial fill field to be examined.

[0051] Specifically, when the original fuzzy score of the original fill field is lower than the preset simulated score threshold, it indicates that the reliability of the original fill field may be low. If the original fill field is directly determined as the initial fill field, it may lead to errors, incompleteness, or logical contradictions in the fill data of the structured information template. In this case, the original fill field can be first determined as the fill field under examination. Based on the fill identifier corresponding to the examination fill field, all related logical modules are traversed, and related fill fields with the same business attributes as the examination fill field are identified. For example, if the fill identifier is the premium amount field, the target logical module is the premium information module, and the original fill field is "￥5**0.00", with the middle two digits blurred due to yellowing paper, the original fuzzy score is 32 points (lower than the preset threshold of 60 points). This is determined as the fill field under examination. Based on the business attributes of the fill identifier (the total premium amount agreed upon in the policy, equal to the sum of the basic premium and the supplementary premium), its related logical modules are traversed to obtain: Related Logic Block 1 (Payment Information Block): Contains the fields "Annual premium of 5800 yuan" and "Payment period of 10 years". After calculation, 5800 yuan / year × 10 years = 5800 yuan. The calculation result is extracted as the related fill field. Related Logic Section 2 (Insurance Information Section): Contains fields such as "Basic Sum Insured 500,000 Yuan" and "Premium Rate 0.116%". After calculation, 500,000 Yuan × 0.116% × 10 Years = 58,000 Yuan. This calculation result is extracted as the related fill field.

[0052] After determining the associated fill fields corresponding to each associated logical module, it is not necessary to directly select any associated fill field as the initial field to be filled. Instead, it is necessary to first perform similarity fusion on all associated fill fields to obtain field similarity, which is used to verify the consistency of multiple associated fill fields and avoid possible deviations in the data of a single associated module. When the field similarity is higher than the preset similarity threshold, it indicates that the associated fill fields extracted from multiple associated logical modules have a high degree of consistency in business attributes and data content. It can be determined that the data of these associated fill fields is highly reliable and there is no contradiction between them. At this time, the total weight of each associated logical module can be obtained by summing the associated module weight, associated data weight, and associated fuzzy weight. The higher the total weight, the higher the priority of the corresponding module. The priority of each associated logical module can be determined based on the preset priority mapping relationship, and the associated fill field corresponding to the associated logical module with the highest priority is determined as the initial field to be filled. The preset priority mapping relationship is the correspondence between the total weight and the module priority. The specific content is not specifically limited in this embodiment of the application.

[0053] By accurately locating low-quality fields that need optimization through fuzzy score thresholds, errors caused by directly filling in fuzzy data are avoided. Furthermore, by using multi-dimensional evaluation of related sector weights, data weights, and fuzzy weights to determine sector priorities, the selected related fields to be filled are ensured to have high reliability. At the same time, through similarity fusion verification, the matching degree between related fields to be filled and the positions to be filled is further guaranteed, which greatly improves the accuracy of field filling.

[0054] Based on the above method, the initial field to be filled for each position in the structured information template can be determined. Then, the initial field to be filled is formatted according to the format corresponding to each position to be filled, so as to obtain the target field to be filled for each position. Finally, the target field to be filled is filled into the corresponding position in the structured information template to obtain a complete structured information block. This process is fully automated and does not require manual identification of ambiguous fields or adjustment of formats, which greatly reduces the manual intervention and improves the overall field filling speed, providing reliable support for the efficient digitization of historical insurance policies.

[0055] Step S150: Combine the structured information blocks corresponding to all target logical blocks to obtain the target structured policy information corresponding to the historical policy image.

[0056] Specifically, the structured information modules corresponding to each target logic module are combined. The core is the standardized integration of modules based on insurance business logic, linking the scattered structured data of sub-modules into a complete, standardized, and directly usable target structured policy information for business systems. A hierarchical combination framework for structured policy information can be pre-established based on insurance industry policy business standards, clarifying the hierarchy and combination order of each structured information module to ensure that the combined data conforms to business application logic.

[0057] The default framework hierarchy is as follows: First level: The core policy information layer, which includes the structured modules of basic policy information and policy expiration time structure. It is the unique identifier of the policy and serves as an index for the entire structured policy information. Second-level layer: Main information layer, which includes the structured information of the policyholder and the structured information of the insured, and contains the main data related to the rights and obligations of the policyholder; The third level is the protection and payment information layer, which includes structured modules for insurance product information, premium amount, and payment information. It contains the core business terms data of the insurance policy. Level 4: Auxiliary information layer, which includes a structured section of agent information and a structured section of remarks, and is supplementary information data for the policy.

[0058] According to the preset framework hierarchy, the structured information blocks corresponding to each target logical block are embedded and integrated from top to bottom to obtain the target structured policy information corresponding to the historical policy image.

[0059] In this embodiment, by identifying business data identifiers in historical policy images, they are divided into multiple initial logical blocks to avoid confusion between different business fields and improve the accuracy of field positioning from the source. Each initial logical block is assessed and scored based on three dimensions: character distinguishability, background interference level, and field completeness. Targeted image preprocessing is then performed based on the scores, improving preprocessing accuracy and avoiding indiscriminate processing, thus optimizing preprocessing efficiency. By matching the target business data identifier and block quality score of the target logical block with the corresponding associated logical blocks, and using the dual data from the target and associated logical blocks, the structured information template is filled with fields. This effectively avoids filling errors caused by quality defects in the target logical block such as blurred characters, background interference, and missing fields, thereby improving the filling speed. Finally, all structured information blocks are automatically combined into the final structured policy information, replacing manual integration and achieving end-to-end automated output from image to structured data, significantly improving processing speed.

[0060] Furthermore, in order to significantly reduce the probability of erroneous data flowing into the business system, the method provided in this application embodiment may further include steps S210-S240, such as... Figure 2 As shown, where: Step S210: Identify target logical blocks whose block quality scores are lower than the preset quality threshold as blocks to be verified, and identify the target fields to be filled corresponding to the blocks to be verified.

[0061] Specifically, the section to be verified is the target logical section whose section quality score is lower than the preset quality threshold. Because the section to be verified has a low section quality score, it may have image quality defects such as blurry characters, severe background interference, and poor field integrity. Therefore, when determining the corresponding target fields to be filled based on the section to be verified, there may be problems such as insufficient reliability of the original data, complex traceability of field sources, and easy omission of core field determination, which directly affects the accuracy of the target fields to be filled. Therefore, it is necessary to further verify the target fields to be filled in the section to be verified.

[0062] Step S220: Determine the target verification dimension and the dimension weight of each target verification dimension based on the quality score of the segment to be verified, and calculate the verification credibility based on the verification information and dimension weight corresponding to each target verification dimension.

[0063] Specifically, different sections have different target verification dimensions corresponding to their quality scores. The target verification dimensions for the sections to be verified can be determined based on the preset verification dimension mapping relationship. The preset verification dimension mapping relationship is the correspondence between the section quality score and the target verification dimension. The target verification dimensions include, but are not limited to, format dimension, business logic dimension, cross-section consistency dimension, and cross-policy knowledge verification dimension.

[0064] Different target verification dimensions correspond to different methods for extracting verification information. Extraction can be based on preset extraction rules, which include: Format dimension: mainly extracts "format standard benchmark information" to verify whether the format of the target field to be filled conforms to the preset specifications of the policy template. The extraction rules can be: extract the format constraint rules corresponding to the target field to be filled from the verification section; Business logic dimension: mainly extracts "business rule constraint information" to verify whether the value of the target field to be filled conforms to the internal logic of insurance business. The extraction rule can be: extract business constraint conditions that are strongly associated with the target field from the product business rule library; Cross-segment consistency dimension: mainly extracts "matching information of related segments within the same policy", used to verify whether the value of the target field to be filled is consistent with the corresponding field of other segments in the same policy. The extraction rule can be: based on the unique identifier of the policy, locate all related segments under the same policy and extract the field values ​​that have the same semantics as the target field; Cross-policy knowledge verification dimension: mainly extracts "reference information from historical policies / industry knowledge base" to verify whether the value of the target field to be filled conforms to the policyholder's historical policy records or industry common knowledge. The extraction rules can be: based on the policyholder's unique identifier (such as ID number), extract the historical values ​​of the same field from the historical policy database.

[0065] Compare the target field to be populated with the validation information for each dimension to determine the score for each dimension, for example: Format dimension: Matching scores 100, non-matching scores 0; Business logic dimension: Field values ​​within the constraints score 100, exceeding the range deducts points based on deviation; Cross-sector consistency dimension: Consistent with values ​​from most sectors scores 100, contradictions deduct points based on the number of contradictions; Cross-policy knowledge verification dimension: Consistent with historical records / industry knowledge scores 100, inconsistent scores 0.

[0066] Finally, based on the dimensional weight of each target verification dimension, the scores of each dimension of the verification information corresponding to each target verification dimension are weighted and summed to obtain the verification credibility.

[0067] Step S230: Compare the target field to be filled with the corresponding original field to determine the field difference.

[0068] Specifically, the target field to be filled can be compared with the corresponding original field to be filled based on a preset feature recognition algorithm to obtain the field difference degree. The core is to quantify the degree of deviation between the two fields in terms of content, format and completeness. The field difference degree ranges from 0 to 100%. The higher the field difference degree, the greater the deviation between the target field to be filled and the original field to be filled; the lower the field difference degree, the stronger the consistency between the two. The specific preset feature recognition algorithm is not specifically limited in this application embodiment.

[0069] Step S240: Determine the audit type of the target field to be filled based on the verification credibility and field difference, and mark it for audit. The audit types include direct entry into the database, random inspection, and manual review.

[0070] Specifically, the verification credibility and field difference corresponding to the target field to be filled are normalized and quantified, and the review score of the target field to be filled is obtained based on the normalization and quantification. Different review scores correspond to different review types, which can be determined based on a preset review type mapping relationship. The preset review type mapping relationship is the correspondence between review scores and review types. The specific content is not specifically limited in this application embodiment. After determining the review type, the filling position of the target field to be filled in the target structured insurance policy information can be marked to remind relevant reviewers to conduct difference review of the target field to be filled. The mark can be in the form of text or symbols. The specific form is not specifically limited in this application embodiment, as long as the relevant reviewers can intuitively distinguish between direct entry, sampling inspection, and manual review.

[0071] In this embodiment of the application, by defining target logical blocks with a quality score lower than a preset threshold as blocks to be verified, it is convenient to achieve accurate targeted verification of low-quality blocks and avoid the waste of efficiency caused by indiscriminate review of all blocks. By matching the verification dimensions and weights with the quality score of the blocks, the verification content can be made more relevant to the quality defects of the blocks, thereby improving the accuracy of verification credibility. At the same time, based on the verification credibility and field difference, three types of review are divided into direct entry into the database, sampling inspection, and manual review. This ensures efficient entry of high credibility and low difference fields into the database, and controls risk fields through sampling inspection and manual review, which greatly reduces the probability of erroneous data flowing into the business system.

[0072] Furthermore, after obtaining the target structured policy information corresponding to the historical policy image, the method provided in this application embodiment may further include: The system detects the policy status, validity period, and near-payment period of structured insurance policy information; based on the policy status, validity period, and near-payment period, it generates a renewal reminder for the target structured insurance policy information and sends the renewal reminder to the policyholder corresponding to the structured insurance policy information.

[0073] Specifically, after the information is completed, core fields related to policy status and payment, such as the current date, effective date, payment method, due date, payment cycle, and last payment date, can be extracted from the structured policy information system based on a preset feature recognition algorithm. Then, combined with insurance business specifications, rules for determining the policy status validity period and the approaching payment period can be formulated, for example: The rules for determining the validity period of an insurance policy can be as follows: If the current date is greater than or equal to the effective date and the current date is less than the expiry date, the policy status can be determined as valid and in effect. If the current date is greater than or equal to the expiry date, the policy status can be determined as lapsed; If the current date is less than the effective date, the policy status is determined to be inactive.

[0074] The rules for determining when the payment period is approaching can be as follows: For policies with payment methods of annual / semi-annual / quarterly / monthly payments, calculate the due date (last payment date + payment period). If the due date minus the current date is less than or equal to the preset reminder threshold (e.g., 30 days) and the due date minus the current date is greater than 0, it can be determined that the payment period is approaching. If the current date is greater than the due date, it can be determined that the payment is overdue.

[0075] Based on the policy's validity period and the approaching payment period, a renewal reminder corresponding to the target structured policy information is generated, and a renewal reminder SMS / WeChat message is automatically sent to the policyholder. The message includes the policy number, the amount due, and the payment deadline. If the policyholder has authorized the process, a payment order can be automatically generated and an attempt can be made to deduct the payment from the linked payment channel. A success notification will be sent after the payment is successfully deducted.

[0076] By automating the detection of policy validity and payment period, the cumbersome process of manually checking and notifying each policy individually is replaced, greatly improving the timeliness and accuracy of renewal reminders and preventing policyholders from missing payment periods or policies from becoming invalid due to human oversight.

[0077] This application provides an electronic device, such as... Figure 3 As shown, Figure 3 The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this electronic device 300 does not constitute a limitation on the embodiments of this application.

[0078] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0079] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by only one line, but this does not mean that there is only one bus or one type of bus.

[0080] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0081] The memory 303 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0082] Electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Servers can also be included. Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0083] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0084] This application provides a computer program product including a computer program that, when executed by a processor, implements the methods described in any of the above embodiments.

[0085] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0086] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for structured filling of historical insurance policy information, characterized in that, include: Identify business data identifiers in historical policy images, and partition the historical policy images based on the business data identifiers to obtain multiple initial logical blocks, each with a different business data identifier. Based on preset evaluation dimensions, a multi-dimensional quality evaluation is performed on each initial logical block to obtain the block quality score corresponding to each initial logical block. Based on the block quality score of each initial logical block, image preprocessing is performed on each initial logical block to obtain the corresponding target logical block. The preset evaluation dimensions include character distinguishability, background interference degree, and field integrity. Based on the target business data identifier and segment quality score corresponding to each target logical segment, determine the associated logical segments corresponding to each target logical segment; Based on each target logical module and its corresponding associated logical module, the fields of the structured information template corresponding to each target logical module are filled to obtain the structured information module corresponding to each target logical module. By combining the structured information blocks corresponding to all target logical blocks, the target structured policy information corresponding to the historical policy image is obtained.

2. The method for structured filling of historical insurance policy information according to claim 1, characterized in that, Based on the target business data identifier and segment quality score corresponding to the target logical segment, the associated logical segments corresponding to the target logical segment are determined, including: Identify the multi-dimensional segment feature parameters corresponding to the target logical segment, and determine the association filtering value corresponding to the target logical segment based on the multi-dimensional segment feature parameters and the segment quality score; Based on the target business data identifier corresponding to the target logic module, the logic module to be examined is determined from other logic modules. The logic module to be examined is other logic modules whose corresponding business data identifier has an identifier similarity to the target business data identifier. Based on the association filtering value and the identifier similarity corresponding to each examined logic block, the associated logic block corresponding to the target logic block is determined from the examined logic blocks.

3. The method for structured filling of historical policy information according to claim 2, characterized in that, The step of determining the association filtering value corresponding to the target logical segment based on the multi-segment feature parameters and the segment quality score includes: Based on the multi-segment feature parameters, the segment position of the target logical segment in the historical policy image is determined, and based on the preset segment position mapping relationship and the segment position, the segment weight corresponding to the target logical segment is determined. The preset segment position mapping relationship is the correspondence between segment position and segment weight. Based on the multi-segment feature parameters, the segment data volume and segment data type contained in the target logical segment are determined. Based on the preset data mapping relationship, the segment data volume, and the segment data type, the data weight corresponding to the target logical segment is determined. The preset data mapping relationship is the correspondence between the parameter combination of segment data volume and segment data type and the data weight. Based on the multi-segment feature parameters, the fuzzy positions and fuzzy regions contained in the target logic segment are determined. Based on the preset fuzzy mapping relationship, the fuzzy positions and the fuzzy regions, the fuzzy weights corresponding to the target logic segment are determined. The preset fuzzy mapping relationship is the correspondence between the parameter combination of fuzzy positions and fuzzy regions and the fuzzy weights. Based on the sector weight, the data weight, the fuzzy weight, and the sector quality score, the association filtering value is determined.

4. The method for structured filling of historical insurance policy information according to claim 3, characterized in that, Based on the target logical module and its corresponding associated logical modules, the structured information template corresponding to the target logical module is populated with fields, including: Based on the fill identifier of the fill position in the structured information template, the corresponding original fill field and the original fuzzy parameter corresponding to the original fill field are identified from the target logical block. The original fuzzy parameter includes the character integrity of the original fill field and the degree of background interference of the original character. Based on the original fuzzy parameters corresponding to the original fill field, determine the original fuzzy score of the original fill field; Based on the original fuzzy score of the original fill field, the initial field to be filled corresponding to the position to be filled is determined. The initial field to be filled is then converted according to the format of the position to be filled to obtain the target field to be filled. Based on the target field to be filled, the field to be filled in the structured information template corresponding to the target logical block is filled.

5. The method for structured filling of historical policy information according to claim 4, characterized in that, The step of determining the initial field to be filled corresponding to the position to be filled based on the original fuzzy score of the original fill field includes: When the original fuzzy score of the original fill field is lower than the preset simulated score threshold, the original fill field is determined as the fill field to be examined, and based on the fill identifier corresponding to the fill field to be examined, the associated fill field corresponding to the fill field to be examined is identified from each associated logic block. Based on the associated section weight, associated data weight, and associated fuzzy weight corresponding to each associated logical section, the section priority of each associated logical section is determined; The similarity of all associated fields is fused to obtain the field similarity. When the field similarity is higher than the preset similarity threshold, the associated field corresponding to the associated logical section with the highest section priority is determined as the initial field to be filled.

6. The method for structured filling of historical policy information according to claim 4, characterized in that, Also includes: Target logical blocks with a quality score lower than a preset quality threshold are identified as blocks to be verified, and the target fields to be filled corresponding to the blocks to be verified are identified. The target verification dimension and the dimension weight of each target verification dimension are determined based on the quality score of the segment to be verified, and the verification credibility is calculated based on the verification information and dimension weight corresponding to each target verification dimension. The target field to be filled is compared with the corresponding original field to be filled to determine the field difference. Based on the verification credibility and the field difference, the audit type of the target field to be filled is determined and audited. The audit type includes direct entry into the database, random inspection, and manual review.

7. The method for structured filling of historical policy information according to claim 1, characterized in that, After obtaining the target structured policy information corresponding to the historical policy image, the process further includes: The policy status, validity period, and near-payment period of the structured policy information are detected. Based on the policy status validity period and the near-payment period, a renewal reminder corresponding to the target structured policy information is generated, and the renewal reminder is sent to the policyholder corresponding to the structured policy information.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, the at least one application being configured to: perform a historical policy information structured filling method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, include: The computer program is stored and can be loaded by a processor and executed as described in any one of claims 1-7, which is a method for structuring and filling historical policy information.

10. A computer program product, characterized in that, The system includes a computer program that, when executed by a processor, implements the steps of a method for structuring and filling historical policy information as described in any one of claims 1-7.

Images