Insurance liability analysis method, device and computer-readable storage medium

Abstract

The present invention discloses a method for parsing insurance liability. The method for parsing insurance liability includes: obtaining an insurance policy image to be identified and an insurance clause document to be parsed; processing the insurance policy image to be identified to obtain a target standard policy field and a target policy field value; processing the insurance clause document to be parsed to obtain a target annotation label and its target information, wherein the target information includes an insurance liability parsing formula; according to the insurance liability parsing formula, obtaining a target parameter value from the target standard policy field and its target policy field value and the target annotation label and its target information, and substituting the target parameter value into the insurance liability parsing formula to obtain a liability parsing result. The present invention also discloses an insurance liability parsing device and a computer-readable storage medium. The present invention can solve the problem of limitations of existing insurance liability parsing methods.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an insurance liability analysis method, device and computer-readable storage medium. Background Art

[0002] In the process of insurance sales, sales staff usually prepare insurance proposals based on the customer's current situation and show them to the user so that the customer can understand the insurance plan and insurance benefits. Among them, the most core part of the insurance proposal is the insurance liability analysis part, which shows the insurance liability, including the liability payment conditions and payment methods.

[0003] At present, the analysis of insurance liability mainly relies on traditional search technology. The insurance benefit values ​​under different input parameters are calculated in advance to generate a benefit data table. After the customer enters the relevant parameters, the corresponding insurance benefit is queried from the benefit data table through search technology. The above method has certain limitations. It can only obtain the pre-entered insurance benefit. If it is not extracted and entered, it cannot be analyzed. Summary of the invention

[0004] The main purpose of the present invention is to provide an insurance liability analysis method, device and computer-readable storage medium, aiming to solve the problem of limitations of existing insurance liability analysis methods.

[0005] To achieve the above object, the present invention provides an insurance liability analysis method, the insurance liability analysis method comprising:

[0006] Obtain the insurance policy image to be identified and the insurance clause document to be parsed;

[0007] Processing the insurance policy image to be identified to obtain a target standard insurance policy field and a target insurance policy field value;

[0008] Processing the insurance clause document to be parsed to obtain a target label and target information thereof, wherein the target information includes an insurance liability parsing formula;

[0009] According to the insurance liability analysis formula, the target parameter value is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and the target parameter value is substituted into the insurance liability analysis formula to obtain the liability analysis result.

[0010] Optionally, the step of processing the to-be-identified insurance policy image to obtain a target standard insurance policy field and its target insurance policy field value comprises:

[0011] Identify the policy image to be identified to obtain initial policy information, wherein the initial policy information includes identification characters and position information thereof;

[0012] According to the preset mapping relationship between the non-standard policy field and the standard policy field, the initial policy field and its corresponding standard policy field are matched from the identification characters;

[0013] The field value characteristics corresponding to the standard policy field are obtained, and the recognition characters other than the initial policy field are matched according to the field value characteristics and the position information to obtain the target policy field value corresponding to the standard policy field.

[0014] Optionally, before the step of matching the initial policy field and its corresponding standard policy field from the recognition characters according to the preset mapping relationship between the non-standard policy field and the standard policy field, the step further includes:

[0015] Acquire a sample policy image and its corresponding first product type and insurance clause information, identify the sample policy image, and obtain sample policy field information;

[0016] Acquire a first target policy field corresponding to the first product type, and determine a first non-standard policy field and a first standard policy field corresponding to each first product type according to the insurance clause information;

[0017] Obtain a second standard policy field according to the first target policy field and the first standard policy field, and perform statistical analysis on the policy sample field information according to the first product type and the second standard policy field to obtain a statistical analysis result;

[0018] Based on the statistical analysis results, the first non-standard policy field and the first standard policy field, a mapping relationship between the preset non-standard policy fields and the standard policy fields is constructed; wherein the mapping relationship between the preset non-standard policy fields and the standard policy fields includes a sub-mapping relationship between the non-standard policy fields and the standard policy fields corresponding to each first product type.

[0019] Optionally, the step of matching the initial policy field and its corresponding standard policy field from the recognition characters according to the preset mapping relationship between the non-standard policy field and the standard policy field comprises:

[0020] According to the mapping relationship between the preset non-standard policy fields and the standard policy fields, the product field and its corresponding standard product field are matched from the identification characters;

[0021] Determine a target sub-mapping relationship corresponding to the standard product field from the mapping relationship between the preset non-standard policy field and the standard policy field;

[0022] According to the target sub-mapping relationship and the specific policy field value characteristics obtained based on the insurance terms information, the remaining policy field and its corresponding standard remaining policy field are matched from the recognition characters, wherein the standard policy field includes the standard product field and the standard remaining policy field.

[0023] Optionally, the step of processing the insurance clause to be parsed to obtain a target label and target information thereof includes:

[0024] Inputting the insurance clause document to be parsed into a pre-trained position annotation model to obtain a position annotation result;

[0025] According to the position marking result, the insurance clause document to be parsed is intercepted to obtain the target insurance clause content;

[0026] Using a pre-trained labeling model to label the target insurance clause content, obtaining a labeling result, wherein the labeling result includes the target label and its corresponding original information, wherein the original information includes the liability information to be parsed;

[0027] The original information is structured to obtain target information corresponding to each target annotation label, wherein the target information includes an insurance liability analysis formula, and the insurance liability analysis formula is obtained by structured processing of the liability information to be analyzed.

[0028] Optionally, before the step of labeling the target insurance clause content using a pre-trained labeling model to obtain a labeling result, the step further includes:

[0029] Acquire a first training sample set, wherein the first training sample set includes insurance clause content samples, real annotation labels and real information thereof, wherein the real annotation labels and real information thereof are obtained based on annotations of a preset corpus;

[0030] Training a preset label annotation model using the first training sample set to obtain a trained label annotation model;

[0031] Wherein, the preset label annotation model includes an information extraction layer and a classification layer;

[0032] The step of training the preset label annotation model by using the first training sample set to obtain a trained label annotation model comprises:

[0033] Inputting the insurance clause content samples into the information extraction layer for information extraction to obtain feature information corresponding to each insurance clause content sample;

[0034] Converting the feature information into a feature vector, inputting the feature vector into the classification layer to obtain a predicted label, and determining corresponding prediction information according to the predicted label and the feature information;

[0035] Calculate the loss value based on the predicted label, the predicted information, the real label of the insurance clause content sample and its real information;

[0036] The parameters of the preset label annotation model are updated by a gradient descent algorithm according to the loss value, and iterative training is performed based on the first training sample set to obtain a trained label annotation model.

[0037] Optionally, the insurance liability analysis method further includes:

[0038] Obtaining sample insurance clause documents, and classifying the sample insurance clause documents by product name;

[0039] Perform cluster analysis on the contents of each part of the classified insurance clause sample documents to obtain clustering results;

[0040] Labeling various types of insurance clauses according to the clustering results to obtain preset labels, and statistically analyzing the values ​​corresponding to the preset labels in the various types of insurance clauses to obtain value features;

[0041] The preset corpus is constructed according to the preset tags and the value features.

[0042] Optionally, the insurance liability analysis method further includes:

[0043] Acquire the product name of the to-be-identified policy image from the target standard policy field and its target policy field value, and determine the target display method according to the product name;

[0044] The responsibility analysis result is displayed based on the target display method.

[0045] In addition, to achieve the above-mentioned purpose, the present invention also provides an insurance liability analysis device, which includes: a memory, a processor, and an insurance liability analysis program stored on the memory and executable on the processor, and the insurance liability analysis program, when executed by the processor, implements the steps of the insurance liability analysis method described above.

[0046] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which an insurance liability analysis program is stored, and when the insurance liability analysis program is executed by a processor, the steps of the insurance liability analysis method as described above are implemented.

[0047] The present invention provides an insurance liability parsing method, device and computer-readable storage medium, which obtains a policy image to be identified and an insurance clause document to be parsed; then, the policy image to be identified is processed to obtain a target standard policy field and its target policy field value; the insurance clause document to be parsed is processed to obtain a target annotation label and its target information, wherein the target information includes an insurance liability parsing formula; according to the insurance liability parsing formula, a target parameter value is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and the target parameter value is substituted into the insurance liability parsing formula to obtain a liability parsing result. In the embodiment of the present invention, automatic identification of all types of policy information and automatic structural processing of the insurance clause document to be parsed can be realized, thereby obtaining the target standard policy field and its target policy field value in the policy, the target annotation label and its target information in the insurance clause document to be parsed, wherein the target information includes the insurance liability parsing formula, and then the target parameter value in the insurance liability parsing formula is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and then the liability parsing result is calculated. Through the above method, automatic analysis of various types of insurance liabilities can be achieved, thereby solving the problem of limitations of existing insurance liability analysis methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the terminal structure of the hardware operating environment involved in the embodiment of the present invention;

[0049] Figure 2 This is a flow chart of the first embodiment of the insurance liability analysis method of the present invention;

[0050] Figure 3 This is a flow chart of a second embodiment of the insurance liability analysis method of the present invention;

[0051] Figure 4 It is a flowchart diagram of the fifth embodiment of the insurance liability analysis method of the present invention.

[0052] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0053] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0054] Reference Figure 1 , Figure 1 The figure is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiment of the present invention.

[0055] The terminal in the embodiment of the present invention may be a PC (personal computer), or may be a terminal device such as a tablet computer, a portable computer, or a server.

[0056] like Figure 1 As shown, the terminal may include: a processor 1001, such as a CPU (Central Processing Unit), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity Wireless-Fidelity, Wi-Fi interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0057] Those skilled in the art will understand that Figure 1 The terminal structure shown in the figure does not constitute a limitation on the terminal, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0058] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, and an insurance liability analysis program.

[0059] exist Figure 1 In the terminal shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the client and communicate data with the client; and the processor 1001 can be used to call the insurance liability parsing program stored in the memory 1005 and execute the various steps of the following insurance liability parsing method.

[0060] Based on the above hardware structure, various embodiments of the insurance liability analysis method of the present invention are proposed.

[0061] The invention provides an insurance liability analysis method.

[0062] Reference Figure 2 , Figure 2 It is a flowchart of the first embodiment of the insurance liability analysis method of the present invention.

[0063] In this embodiment, the insurance liability analysis method includes:

[0064] Step S10, obtaining the insurance policy image to be identified and the insurance clause document to be parsed;

[0065] The terminal in the embodiment of the present invention may be a PC, or a terminal device such as a tablet computer, a portable computer, a server, etc. In this embodiment, a server is used as an example for description.

[0066] In this embodiment, the insurance data to be parsed is first obtained, wherein the insurance data to be parsed includes an insurance policy image to be identified and an insurance clause document to be parsed.

[0067] To facilitate subsequent processing, it is possible to first detect whether the format of the insurance terms document to be parsed is a preset document format, wherein the preset document format can be optionally a txt document or a doc document. If it is not a preset document format, the insurance terms document to be parsed is format-converted to obtain the insurance terms document to be parsed after format conversion. For example, if the insurance terms document to be parsed is a pdf document, the pdf can be converted into a txt document or a doc document; if the insurance terms document to be parsed is a picture, it is recognized by OCR (Optical Character Recognition) recognition technology to obtain its text information and its position information in the document, and then the corresponding text information is input into the txt document or the doc document according to the position information. The subsequent processing is based on the insurance terms document to be parsed after format conversion.

[0068] Step S20, processing the to-be-identified insurance policy image to obtain a target standard insurance policy field and its target insurance policy field value;

[0069] Then, the policy image to be identified is processed to obtain the target standard policy field and its target policy field value. Specifically, the policy image to be identified is first identified to obtain the initial policy information, wherein the initial policy information includes identification characters and their position information; then, according to the mapping relationship between the preset non-standard policy field and the standard policy field, the initial policy field and its corresponding standard policy field are matched from the identification characters; finally, the field value features corresponding to the standard policy field are obtained, and the identification characters except the initial policy field are matched according to the field value features and position information to obtain the target policy field value corresponding to the standard policy field. The specific execution process can refer to the second embodiment described below.

[0070] Step S30, processing the insurance clause document to be parsed to obtain a target label and target information thereof, wherein the target information includes an insurance liability parsing formula;

[0071] The insurance clause document to be parsed is processed to obtain a target annotation label and its target information, wherein the target information includes an insurance liability parsing formula. The process of obtaining the target annotation label and its target information is as follows: input the insurance clause document to be parsed into a pre-trained position annotation model to obtain a position annotation result; intercept the insurance clause document to be parsed according to the position annotation result to obtain the target insurance clause content; use the pre-trained label annotation model to label the target insurance clause content to obtain a label annotation result, wherein the label annotation result includes the target annotation label and its corresponding original information, wherein the original information includes the liability information to be parsed; perform structured processing on the original information to obtain the target information corresponding to each target annotation label, wherein the target information includes an insurance liability parsing formula, and the insurance liability parsing formula is obtained by performing structured processing on the liability information to be parsed. The specific execution process can refer to the fifth embodiment described below.

[0072] Among them, insurance liability analysis formulas can be divided into two categories: basic logic operation standard analysis formulas and special insurance liability standard analysis formulas. Basic logic operation standard analysis formulas may include but are not limited to: multiple formulas, MAX and MIN maximum and minimum value formulas, etc. Special insurance liability standard analysis formulas may include reimbursement medical liability standard analysis formulas, cash flow annuity standard analysis formulas, etc.

[0073] It should be noted that the execution of step S20 and step S30 is not in any particular order.

[0074] Step S40, according to the insurance liability analysis formula, obtain the target parameter value from the target standard policy field and its target policy field value and the target annotation label and its target information, and substitute the target parameter value into the insurance liability analysis formula to obtain the liability analysis result.

[0075] Finally, according to the insurance liability parsing formula, the target parameter value is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and the target parameter value is substituted into the insurance liability parsing formula to obtain the liability parsing result.

[0076] For example, if the insurance liability analysis formula is: basic insurance amount*payment period*1%*[1+3%*(number of policy years-3)], then the target parameter values ​​to be obtained include the basic insurance amount, payment period and number of policy years, and then they are substituted into the above formula to calculate the liability analysis result.

[0077] The present invention provides an insurance liability parsing method, which obtains a policy image to be identified and an insurance clause document to be parsed; then, the policy image to be identified is processed to obtain a target standard policy field and its target policy field value; the insurance clause document to be parsed is processed to obtain a target annotation label and its target information, wherein the target information includes an insurance liability parsing formula; according to the insurance liability parsing formula, a target parameter value is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and the target parameter value is substituted into the insurance liability parsing formula to obtain a liability parsing result. In an embodiment of the present invention, automatic identification of all types of policy information and automatic structural processing of the insurance clause document to be parsed can be realized, thereby obtaining the target standard policy field and its target policy field value in the policy, the target annotation label and its target information in the insurance clause document to be parsed, wherein the target information includes the insurance liability parsing formula, and then the target parameter value in the insurance liability parsing formula is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and then the liability parsing result is calculated. Through the above method, automatic analysis of various types of insurance liabilities can be achieved, thereby solving the problem of limitations of existing insurance liability analysis methods.

[0078] Further, based on the above first embodiment, a second embodiment of the insurance liability analysis method of the present invention is proposed. Figure 3 , Figure 3 It is a flowchart diagram of the second embodiment of the insurance liability analysis method of the present invention.

[0079] In this embodiment, the above step S20 includes:

[0080] Step S21, identifying the policy image to be identified to obtain initial policy information, wherein the initial policy information includes identification characters and position information thereof;

[0081] This embodiment provides a process for processing an insurance policy image to be identified, which is as follows:

[0082] First, the policy image to be identified is identified to obtain the initial policy information, wherein the initial policy information includes identification characters and position information, the identification characters include the initial policy field and the initial policy field value, and the position information is the position of the identification characters in the image, which can be represented in the form of coordinates. When performing image recognition, OCR (Optical Character Recognition) technology can be used for recognition. The specific recognition method can refer to the existing technology and will not be described here.

[0083] Step S22, according to the preset mapping relationship between the non-standard policy field and the standard policy field, matching the initial policy field and its corresponding standard policy field from the identification characters;

[0084] Then, according to the mapping relationship between the preset non-standard policy fields and the standard policy fields, the initial policy fields and their corresponding standard policy fields are matched from the recognition characters. The construction of the mapping relationship between the preset non-standard policy fields and the standard policy fields can refer to the third embodiment described below, which will not be described in detail this time.

[0085] Step S23, obtaining the field value characteristics corresponding to the standard policy field, matching the recognition characters except the initial policy field according to the field value characteristics and the position information, and obtaining the target policy field value corresponding to the standard policy field.

[0086] Finally, the field value features corresponding to the standard policy fields are obtained, where the field value features are the features of the values ​​corresponding to the standard policy fields, and may include one or more. For example, for the standard policy field "policy number", its field value feature is a combination of numbers and letters, and the characters are usually more than 5 digits; for another example, for the standard policy field "ID number", its field value feature is 18 digits.

[0087] Furthermore, the identification characters except the initial policy field are matched according to the field value characteristics and position information to obtain the target policy field value corresponding to the standard policy field.

[0088] Specifically, the step of "matching the identification characters except the initial policy field according to the field value characteristics and position information to obtain the target policy field value corresponding to the standard policy field" includes:

[0089] Step a231, recording the identification characters except the initial policy field as the initial policy field value, and obtaining the first position information of the initial policy field and the second position information of the initial policy field value from the position information;

[0090] Step a232, calculating the relative distance between each initial policy field and each initial policy field value according to the first position information and the second position information;

[0091] Step a233, filtering out suspected field values ​​corresponding to the initial policy fields from the initial policy field values ​​according to the relative distance and the preset range;

[0092] Step a234, matching the target policy field value corresponding to the standard policy field from the suspected field value according to the field value feature.

[0093] In this embodiment, the matching process of the target policy field value corresponding to the standard policy field is as follows:

[0094] For ease of explanation, the identification characters except the initial policy field are recorded as the initial policy field value, and the first position information of the initial policy field and the second position information of the initial policy field value are first obtained from the position information.

[0095] Then, the relative distance between each initial policy field and each initial policy field value is calculated based on the first position information and the second position information. The relative distance calculation method may include but is not limited to: 1) directly calculating the distance between the first position information and the second position information. For example, when the first position information and the second position information are represented in the form of coordinates, the coordinates of an initial policy field are (x1, y1), and the coordinates of an initial policy field value are (x2, y2), then the distance d=[(x2-x1) 2 +(y2-y1) 2 ] 1 / 2 ; 2) Determine the relative distance based on the distance and direction. For example, the distance between the first location information and the second location information can be calculated first, and the distance score can be determined based on the mapping relationship between the distance and the preset distance and the score; at the same time, the direction of the initial policy field value relative to the initial policy field can be determined based on the first location information and the second location information, and then the direction score can be determined based on the mapping relationship between the direction and the preset direction and the score, and the distance score and the direction score can be added, and the sum can be used as the relative distance.

[0096] Then, according to the relative distance and the preset range, the suspected field value corresponding to the initial policy field is screened out from the initial policy field value. That is, the initial policy field value whose relative distance to the initial policy field is within the preset range is taken as the suspected field value to narrow the subsequent matching range and improve the matching efficiency.

[0097] Finally, the target policy field value corresponding to the standard policy field is matched from the suspected field values ​​according to the field value characteristics. When matching, the value corresponding to the field value characteristics of the currently matched standard policy field can be filtered out from the suspected field values ​​as the target policy field value.

[0098] It should be noted that after the matching is completed, it is checked whether there are standard policy fields that do not match the target policy field value (recorded as unmatched standard policy fields). If so, the recognition characters other than the target policy field value and the standard policy field are matched according to the field value characteristics of the unmatched standard policy field to expand the matching range and perform further matching.

[0099] In addition, since the policy field values ​​of different standard policy fields may be the same, if there is a standard policy field that does not match the target policy field value (recorded as an unmatched standard policy field), it is possible to detect whether there is a standard policy field with the same policy field value in the unmatched standard policy field. If so, its policy field value can be directly obtained. For example, if the "Insurance Period" field value of an insurance product is "Same as Main Insurance", the insurance period field value of the main insurance product in the policy is copied to the insurance period of the insurance product.

[0100] Furthermore, after obtaining the target policy field value, it is possible to further detect whether the target policy field value meets the standardized format requirements. If not, the target policy field value is standardized to obtain a standard policy field value. Among them, the step of "standardizing the target policy field value" includes: detecting whether there are preset redundant characters in the target policy field value; if there are preset redundant characters, deleting the preset redundant characters in the target policy field value; and / or, detecting whether there are abbreviation characters in the target policy field value; if there are abbreviation characters, replacing the abbreviation characters in the target policy field value with the corresponding Chinese full name; and / or, detecting whether the target policy field value meets the output format; if not, format conversion is performed on the target policy field value; and / or, detecting whether there is a latest expression for the target policy field value; if so, replacing the target policy field value with the latest expression. Through the above method, the standardized output of the standard policy field value can be achieved, which is convenient for subsequent further insurance liability parsing and processing.

[0101] In an embodiment of the present invention, all characters and their position information in the insurance policy image are first identified, and then the standard policy field is obtained by matching according to the mapping relationship between the preset non-standard policy field and the standard policy field, and then the target policy field value corresponding to the standard policy field is obtained by matching according to the field value characteristics and position information of the standard policy field. Through the above method, there is no need to set corresponding policy templates for various types of policies for identification. Only simple image recognition technology is needed to realize information recognition of all types of policies, and intelligent recognition of information of various types of policies can be realized, thereby facilitating subsequent analysis of insurance liability.

[0102] Furthermore, based on the above-mentioned second embodiment, a third embodiment of the insurance liability analysis method of the present invention is proposed.

[0103] In this embodiment, before the above step S22, the insurance liability analysis method further includes:

[0104] Step A, obtaining a sample policy image and its corresponding first product type and insurance clause information, identifying the sample policy image, and obtaining sample policy field information;

[0105] In this embodiment, a sample policy image and its corresponding product type (which is recorded as the first product type to distinguish it from the product type of the subsequent policy image to be identified) and insurance clause information are obtained, wherein the sample policy image can be a selected policy image of different product types, so as to be used to count the expression of different policy fields of policies of different companies and different product types. The insurance clause information is the information of the insurance clause part corresponding to the policy, which is obtained through structured processing and can include clause labels and their label values.

[0106] Then, the sample policy image is identified to obtain the sample policy field information. The sample policy field information is the policy field in the sample policy image corresponding to the standard policy field. When performing image recognition, OCR technology can be used for recognition. The specific recognition method can refer to the existing technology and will not be described here.

[0107] Step B, obtaining the first target policy field corresponding to the first product type, and determining the first non-standard policy field and the first standard policy field corresponding to each first product type according to the insurance clause information;

[0108] Then, the target policy field corresponding to the first product type is obtained (to distinguish it from the target policy field of the subsequent policy image to be identified, it is recorded as the first target policy field), wherein the first target policy field refers to the policy field existing in the insurance policy corresponding to each first product type, and is also the key field that needs to be parsed for the insurance policy corresponding to the first product type for policy entry or user viewing.

[0109] At the same time, the first non-standard policy field and the first standard policy field corresponding to each first product type are determined according to the insurance clause information. That is, the standard policy fields and their expressions included in the insurance clauses are determined according to the insurance clause information. This is because the insurance clauses often specify corresponding expressions for some standard policy fields. For example, in some insurance products, the standard policy field "payment period" is expressed as "payment period" in the insurance clauses. Correspondingly, the insurance policy will generally use "payment period". Therefore, the expression of the standard policy fields involved in the insurance clauses can assist in subsequent matching, which can be matched faster and more accurately, thereby improving the efficiency and accuracy of information standardization processing.

[0110] Step C, obtaining a second standard policy field according to the first target policy field and the first standard policy field, and performing statistical analysis on the policy sample field information according to the first product type and the second standard policy field to obtain a statistical analysis result;

[0111] Next, a second standard policy field is obtained according to the first target policy field and the first standard policy field, wherein the second standard policy field is the policy field in the first target policy field except the first standard policy field.

[0112] Then, the policy sample field information is statistically analyzed according to the first product type and the second standard policy field to obtain the statistical analysis result. During the statistical analysis, the policy fields corresponding to the second standard policy fields of the same first product type are statistically analyzed according to the policy sample field information. Correspondingly, the statistical analysis result includes the non-standard policy fields corresponding to the second standard policy fields.

[0113] Step D, constructing a mapping relationship between the preset non-standard policy fields and the standard policy fields based on the statistical analysis results, the first non-standard policy fields and the first standard policy fields; wherein the mapping relationship between the preset non-standard policy fields and the standard policy fields includes a sub-mapping relationship between the non-standard policy fields and the standard policy fields corresponding to each first product type.

[0114] Finally, a mapping relationship between the preset non-standard policy field and the standard policy field is constructed based on the statistical analysis results, the first non-standard policy field and its first standard policy field. The mapping relationship between the non-standard policy field and the standard policy field can be in a many-to-one or one-to-one form.

[0115] Among them, the statistical analysis results include the second standard policy field and its corresponding second non-standard policy field. When constructing the mapping relationship, for different first product types, a first sub-mapping relationship is constructed based on its first non-standard policy field and its first standard policy field, and a second sub-mapping relationship is constructed based on the second non-standard policy field and its second standard policy field. The mapping relationship between the preset non-standard policy field and the standard policy field is composed of the above-mentioned first sub-mapping relationship and the second sub-mapping relationship. That is, the mapping relationship actually includes multiple sub-mapping relationships, each of which is a mapping relationship between non-standard policy fields and standard policy fields of different product types, wherein the standard policy field in each sub-mapping relationship is the first target policy field of each first product type. Specifically, each sub-mapping relationship is further divided into a sub-mapping relationship between the first non-standard policy field and the first standard policy field obtained from the insurance terms information and a sub-mapping relationship between the second non-standard policy field and the second standard policy field obtained from the policy sample field information statistics.

[0116] Furthermore, the statistical analysis result includes the second standard policy field and its corresponding non-standard policy field, and step C may further include:

[0117] Step C1, obtaining synonyms of the second standard policy field and the second non-standard policy field;

[0118] Step C2, constructing a mapping relationship between the preset non-standard policy field and the standard policy field based on the statistical analysis result, the synonym, the first non-standard policy field and the first standard policy field.

[0119] Furthermore, in order to expand the matching scope, improve the matching efficiency and the accuracy of the matching results, synonyms of the second standard policy field and the second non-standard policy field can also be obtained, and then a mapping relationship between the preset non-standard policy field and the standard policy field can be constructed based on the statistical analysis results and the synonyms. That is, the synonyms are added to the second non-standard policy field.

[0120] In this embodiment, by constructing non-standard policy fields corresponding to preset standard policy fields, subsequent identification and matching of policy fields can be facilitated.

[0121] Furthermore, based on the above third embodiment, a fourth embodiment of the insurance liability analysis method of the present invention is proposed.

[0122] In this embodiment, the above step S22 includes:

[0123] Step a221, according to the preset mapping relationship between the non-standard policy field and the standard policy field, match the product field and its corresponding standard product field from the identification characters;

[0124] In this embodiment, when the mapping relationship between the preset standard policy field and the standard policy field is constructed in combination with the insurance clause information, that is, the mapping relationship between the preset standard policy field and the standard policy field includes a sub-mapping relationship between non-standard policy fields of different product types and standard policy fields, and each sub-mapping relationship is further divided into a sub-mapping relationship between the first non-standard policy field obtained from the insurance clause information and the first standard policy field and a sub-mapping relationship between the second non-standard policy field and the second standard policy field obtained by statistics of the policy sample field information. At this time, the product field and its corresponding standard product field can be matched from the recognition characters according to the mapping relationship between the preset non-standard policy field and the standard policy field. Among them, the product field, recorded as the field corresponding to the product type, can be obtained by matching the target fields such as the product name, product type or policy name, and then determining its corresponding standard product field, that is, the standardized product type name.

[0125] Step a222, determining a target sub-mapping relationship corresponding to the standard product field from the mapping relationship between the preset non-standard policy field and the standard policy field;

[0126] Then, the target sub-mapping relationship corresponding to the standard product field is determined from the mapping relationship between the preset non-standard policy field and the standard policy field. That is, the sub-mapping relationship corresponding to the matched product type is determined and recorded as the target sub-mapping relationship.

[0127] Step a223, according to the target sub-mapping relationship and the specific policy field value characteristics obtained based on the insurance terms information, match the remaining policy field and its corresponding standard remaining policy field from the recognition characters, wherein the standard policy field includes the standard product field and the standard remaining policy field.

[0128] As an implementation method, the remaining policy field and its corresponding standard remaining policy field can be matched from the recognition characters according to the target sub-mapping relationship, wherein the standard policy field includes the standard product field and the standard remaining policy field, and the standard remaining policy field is other policy fields included in the policy except the standard product field.

[0129] When matching, a part of the remaining policy fields and their standard remaining policy fields can be determined first based on the sub-mapping relationship between the first non-standard policy field obtained from the insurance terms information and the first standard policy field in the target sub-mapping relationship, and then another part of the remaining policy fields and their standard remaining policy fields can be determined based on the sub-mapping relationship between the second non-standard policy field obtained from the policy sample field information and the second standard policy field in the target sub-mapping relationship.

[0130] As another implementation, field recognition can also be performed in combination with the specific policy field value features obtained based on the insurance clause information. Specifically, according to the target sub-mapping relationship and the specific policy field value features obtained based on the insurance clause information, the remaining policy field and its corresponding standard remaining policy field are matched from the recognition characters, wherein the standard policy field includes a standard product field and a standard remaining policy field. Among them, the specific policy field value features obtained based on the insurance clause information are the field value features corresponding to certain specific policy fields obtained by analyzing the insurance clause information. During recognition, the field value of the corresponding field can be preferentially identified according to the specific policy field value features, and then the corresponding part of the remaining policy field can be identified when its position information is within a certain range, and then the remaining standard policy field corresponding to the part of the remaining policy field is determined according to the target sub-mapping relationship, and then the remaining remaining policy field and its standard remaining policy field are determined according to the target sub-mapping relationship. The specific recognition process can refer to the previous implementation.

[0131] Through the above method, the matching efficiency of the insurance policy fields can be improved, and the matching accuracy of the insurance policy fields can also be improved.

[0132] Further, based on the above first embodiment, a fifth embodiment of the insurance liability analysis method of the present invention is proposed. Figure 4 , Figure 4 It is a flowchart diagram of the fifth embodiment of the insurance liability analysis method of the present invention.

[0133] In this embodiment, the above step S30 includes:

[0134] Step S31, inputting the insurance clause document to be parsed into a pre-trained position annotation model to obtain a position annotation result;

[0135] This embodiment provides a labeling method for insurance clauses and a method for standardizing information thereof, which is as follows:

[0136] First, input the insurance clause document to be parsed into the pre-trained position annotation model to obtain the position annotation result. Among them, the position annotation model is used to annotate the positions of the various parts of the insurance clause content (such as basic information of insurance clauses, basic information of liability, etc., which can be subdivided as needed) in the insurance clause document to be parsed, so as to help to take screenshots of the contents of each part of the insurance clause in the subsequent process, so as to narrow the processing scope of the insurance clause, improve the processing efficiency, and avoid the interference of useless parts, so as to improve the processing accuracy to a certain extent.

[0137] The training process of the position annotation model is as follows: first obtain the second training sample set, wherein the second training sample set includes the insurance clause sample document and the annotation boxes of each part of the insurance clause content. The contents of each part of the insurance clause may include but are not limited to the basic information of the insurance clause, the basic information of the liability, etc., which can be divided according to actual needs, and the annotation box is used to mark the corresponding position of each part of the clause content in the insurance clause sample document. Then, the preset position annotation model is trained through the second training sample set to obtain a trained position annotation model. Among them, the preset position annotation model can be optionally a multi-classification model, such as a neural network model, a random forest model, a logistic regression model, etc. The specific training process can refer to the prior art. By pre-training the position annotation model, it is used to annotate the positions of each part of the insurance clause content in the insurance clause sample document, which is helpful for subsequent screenshots of each part of the insurance clause content, so as to narrow the processing scope of the insurance clause, improve the processing efficiency, and avoid the interference of useless parts, so as to improve the accuracy of processing to a certain extent.

[0138] Step S32, intercepting the insurance clause document to be parsed according to the position marking result to obtain the target insurance clause content;

[0139] Then, the insurance clause document to be parsed is intercepted according to the position labeling results to obtain the target insurance clause content. That is, the part of the information that needs to be structured is intercepted to facilitate subsequent labeling and structured processing. For example, the important information of the insurance clause mainly includes basic insurance information (such as file name, full name of insurance company, product name, design type, page number and position of liability), liability payment information (such as basic liability information, liability payment conditions and liability payment formula, etc.), limited information and some other special information (such as dividend information).

[0140] Step S33, labeling the target insurance clause content using a pre-trained label labeling model to obtain a label labeling result, wherein the label labeling result includes the target label and its corresponding original information, wherein the original information includes the liability information to be parsed;

[0141] After intercepting the target insurance clause content from the insurance clause document to be parsed, the target insurance clause content is labeled using a pre-trained labeling model to obtain a labeling result, wherein the labeling result includes the target label and its corresponding original information, the original information is the value corresponding to the target label, and the original information includes the liability information to be parsed, and the liability information to be parsed includes the liability payment condition description and the liability payment formula description. The training process of the labeling model can refer to the fifth embodiment described below.

[0142] As one of the labeling methods, a labeling model that can be used to identify and label various labels can be pre-trained. Then, the target insurance clause content is input into the labeling model to obtain the labeling result.

[0143] As another labeling method, multiple labeling models can be pre-trained, each of which is for different types of insurance clause content, so as to label different types of insurance clause content. Then, according to the type of the target insurance clause content, each type of target insurance clause content is input into the corresponding labeling model to obtain the labeling result. Compared with the above labeling method, this embodiment can make each type of target insurance clause content processed in parallel, which can improve the processing efficiency, and the labeling model is trained for each type of insurance clause content, and its processing is more targeted, thereby improving the accuracy of the processing result.

[0144] Here, it should be noted that the target labeling labels can be divided into 17 categories, including basic information of insurance terms, basic information of liability, detailed analysis of liability, proportional coefficient, restriction labels, other labels, annuity labels, universal or investment-linked labels, dividend product labels, disease labels, medical labels, medical expense labels, medical allowances, disability labels, continuous bonuses, and pension security entrustment labels. Each major category has specific detailed labels. For example, the basic information of insurance terms can include 5 detailed labels: file name, full name of insurance company, product name, design type, page number and position of liability. The basic information of liability can include 4 detailed labels: liability name, liability layer 1, liability layer 2, and liability clause overview. The detailed analysis of liability can include 5 detailed labels: waiting period, text description of liability condition payment, liability payment formula, description of liability association with this contract, and description of liability association with other contracts. The restriction labels can include 6 detailed labels: payment form, payment frequency limit, payment period upper limit, interval between the same accident, payment limit description, and payment amount limit. Tags, other tags can include optional liability core words, policy loan ratio, common word complex situations, guaranteed renewal period, no claims discount, these five detailed tags, annuity tags can include starting age for receiving, ending age for receiving, conversion of annuity options, annuity into universal account, annuity guaranteed payment, these five detailed tags, universal or investment-linked tags can include initial fee, risk premium, policy management fee, buy-sell spread, minimum guaranteed interest rate, these five detailed tags, dividend product tags can include dividend information location, dividend mode, dividend utilization The disease label can include two detailed labels: disease type and disease group. The medical label can include one detailed label: the number of days of liability extension. The medical expense label can include one detailed label: deductible. The medical allowance label can include four detailed labels: deductible period, daily payment allowance, base of each payment allowance, and insurance score. The disability label can include two detailed labels: re-disability and change of disability status. The continuous bonus can include two detailed labels: description of continuous bonus conditions and description of continuous bonus payment. Of course, it can be understood that other sub-labels can also be appropriately set under each detailed label. The setting of the above-mentioned identification labels is determined based on the identification and statistical analysis of the existing insurance terms information, and can cover various important information of the insurance terms.

[0145] For example (remember Example 1), for the liability clause part "From the corresponding date of the fourth year of effectiveness of this contract, if the insured survives to the corresponding date of the annual effectiveness of this contract, the company will pay the care annuity according to the following provisions each year: / / Care annuity = basic insurance amount * payment period (number of years) * 1% * [1 + 3% * (the number of policy years that have passed under this contract - 3)]", the liability payment condition description can be "From the corresponding date of the fourth year of effectiveness of this contract, if the insured survives to the corresponding date of the annual effectiveness of this contract", and the liability payment formula description is "The company will pay the care annuity according to the following provisions each year: / / Care annuity = basic insurance amount * payment period (number of years) * 1% * [1 + 3% * (the number of policy years that have passed under this contract - 3)]."

[0146] For another example (Example 2), for the liability clause part "If the insured survives to the corresponding date of the year when he / she reaches the age of seventy-five, the company shall pay the maturity insurance benefit in accordance with the following provisions, and this contract shall terminate. / / Maturity insurance benefit = basic insurance amount × payment period (number of years)", the liability payment condition description can be obtained as "If the insured survives to the corresponding date of the year when he / she reaches the age of seventy-five", and the liability payment formula description is "The company shall pay the maturity insurance benefit in accordance with the following provisions, and this contract shall terminate. / / Maturity insurance benefit = basic insurance amount × payment period (number of years)".

[0147] Step S34, structured processing is performed on the original information to obtain target information corresponding to each target annotation label, wherein the target information includes an insurance liability analysis formula, and the insurance liability analysis formula is obtained by structured processing of the liability information to be analyzed.

[0148] Finally, the original information is structured, that is, it is represented in a unified format to obtain the target information corresponding to each target label, so that the insurance clause document to be parsed is finally output in a structured manner with target label and target information, which is convenient for subsequent parsing of insurance liability. Among them, the target information includes the insurance liability parsing formula, which is obtained by structured processing of the liability information to be parsed.

[0149] When performing structured processing, it can be processed based on a pre-built corpus, which specifies the standardized expression of the corresponding values ​​of each annotation tag, including the unified expression of units, the unified expression of characters, the standardized expression of words, the formula expression, etc. For example, for units, the term units are unified as d, m, and y, representing days (days), months, and years respectively; for characters, it can include the expression of conformity, the unified expression of Chinese and English characters, the formula symbols are unified as: max, min, +, -, *, / , and the Chinese and English characters are unified to be modified to the corresponding full Chinese names or English abbreviations; for words, standardized expressions are uniformly adopted; for formula expressions, on the one hand, it is also the expression of formula symbols and words, and on the other hand, it is the expression in the form of various types of formulas, such as the expression "the greater of the fifth insurance contract anniversary and the first insurance contract anniversary of reaching the age of sixty-five", which can be expressed by the formula "max (insured age + 5, 65)", and for another example, the expression "from the first policy anniversary", which can be expressed by the formula "insured age + 1". For another example, the expression of the liability payment formula can be converted according to the unified regulations of the above-mentioned formula symbols and sentence analysis. For example, the text description of the liability payment part is "The company will pay the death benefit according to the balance of the accidental injury insurance amount agreed in this contract after deducting the disability insurance benefit already paid, and this contract will terminate." Then the liability payment formula can be converted to "accidental injury insurance amount - paid insurance amount".

[0150] For another example, in Example 1 above, based on the description of the above liability payment conditions and liability payment formula, we can get the preliminary insurance liability payment formula: basic insurance amount*payment period (number of years)*1%*[1+3%*(number of policy years this contract has passed - 3)], and then, standardize the words used therein to get the standard insurance liability payment formula: basic insurance amount*payment period*1%*[1+3%*(number of policy years - 3)].

[0151] For another example, in Example 2 above, based on the description of the above liability payment conditions and liability payment formula, a preliminary insurance liability payment formula can be obtained: basic insurance amount * payment period (number of years). Then, the words used therein are standardized to obtain the standard insurance liability payment formula: basic insurance amount * payment period.

[0152] The embodiment of the present invention automatically structures the unstructured insurance clause information by means of labels, and obtains corresponding target annotation labels and target information, thereby labeling and structuring the insurance clauses. Compared with manual processing, the processing efficiency and accuracy can be greatly improved, and the subsequent analysis of insurance liability can be facilitated.

[0153] Furthermore, based on the fifth embodiment described above, a sixth embodiment of the insurance liability analysis method of the present invention is proposed.

[0154] In this embodiment, before the above step S33, the insurance liability analysis method further includes:

[0155] Step E, obtaining a first training sample set, wherein the first training sample set includes insurance clause content samples, real annotation labels and real information thereof, wherein the real annotation labels and real information thereof are obtained based on annotations of a preset corpus;

[0156] In this embodiment, a process of obtaining a label annotation model is provided, which is specifically as follows:

[0157] First, a first training sample set is obtained. The first training sample set includes insurance clause content samples, true annotation labels and their true information, wherein the true annotation labels and their true information are obtained based on the annotation of the preset corpus, the true annotation labels are obtained by manually judging the labels to which the text in the insurance clause content samples belongs based on the preset labels in the preset corpus, and the true information is the value corresponding to the true annotation label in the insurance clause content sample. The construction method of the preset corpus can refer to the sixth embodiment described below.

[0158] Step F: training the preset label annotation model using the first training sample set to obtain a trained label annotation model.

[0159] Then, the preset label annotation model is trained by the first training sample set to obtain a trained label annotation model. The preset label annotation model may include one or more. When one is included, the first training sample set is directly input into the preset label annotation model for iterative training. When multiple are included, the insurance clause content samples may be input into each preset label annotation model by type for iterative training to obtain a label annotation model for labeling different types of insurance clause content samples.

[0160] Specifically, the preset label annotation model includes an information extraction layer and a classification layer, and step F includes:

[0161] Step F1, inputting the insurance clause content samples into the information extraction layer for information extraction, and obtaining feature information corresponding to each insurance clause content sample;

[0162] Step F2, converting the feature information into a feature vector, inputting the feature vector into the classification layer to obtain a predicted label, and determining corresponding prediction information according to the predicted label and the feature information;

[0163] Step F3, calculating the loss value according to the predicted label, the predicted information, the real label of the insurance clause content sample and its real information;

[0164] Step F4, updating the parameters of the preset label annotation model by using a gradient descent algorithm according to the loss value, and performing iterative training based on the first training sample set to obtain a trained label annotation model.

[0165] In this embodiment, the label annotation model includes an information extraction layer and a classification layer, wherein the information extraction layer is used to extract feature information, and the classification layer is used to classify the feature information to determine the label corresponding to each word or sentence. The training process of the label annotation model is as follows:

[0166] First, the insurance clause content samples are input into the information extraction layer for information extraction to obtain the feature information corresponding to each insurance clause content sample, where the feature information may include but is not limited to part-of-speech information (subject, predicate, and object information), entity information (entity name), position information (position information in the entire sentence), word vector, sentence vector, etc. The extraction of various features can be performed using corresponding modules, and the information extraction layer can be constructed according to actual needs.

[0167] Then, the feature information is converted into a feature vector, and the feature vector is input into the classification layer to obtain a preliminary label, which is recorded as a predicted label, and then the corresponding prediction information is determined according to the predicted label and the feature information, wherein the prediction information is the predicted value corresponding to the predicted label. When determining the prediction information, the value feature corresponding to the preset value in the preset corpus can be matched with the extracted feature information.

[0168] After obtaining the predicted label and the corresponding predicted information, the loss value is calculated based on the predicted label, the predicted information, the real label of the insurance clause content sample and its real information. Specifically, the first loss value can be calculated based on the predicted label and the real label, and then the second loss value can be calculated based on the predicted information and the real information. The first loss value and the second loss value are added to obtain the final loss value. The calculation of the first loss value and the second loss value can be calculated using loss functions such as Mean Square Error (MSE), Mean Absolute Error (MAE), Multi-classification SVM loss (Hinge Loss) function, and cross entropy loss function.

[0169] Finally, the parameters of the preset label annotation model are updated through the gradient descent algorithm according to the loss value, and iterative training is performed based on the first training sample set, that is, the gradients of the nodes in each layer of the label annotation model are updated according to the total loss, and then the weight parameters of each node are updated, and the above steps F1 to F3 are repeated, and the iteration is continued until the network converges, that is, the total loss steadily decreases to a smaller range (lower than the preset threshold or reaches the minimum value). At this time, a trained label annotation model can be obtained. The optimization problem of large-scale sample data can be solved by the gradient descent algorithm. The specific gradient descent algorithm can refer to the existing technology and will not be described here.

[0170] In this embodiment, a label annotation model is pre-trained to annotate the labels and their values ​​(original information) contained in the content of various types of insurance clauses, and then the original information is structured to obtain the target annotation labels and their target information corresponding to the insurance clause document to be parsed, so as to label and structure the insurance clauses. Compared with manual processing, the processing efficiency and accuracy can be greatly improved, and the subsequent analysis of insurance liability can be facilitated.

[0171] Furthermore, based on the sixth embodiment described above, a seventh embodiment of the insurance liability analysis method of the present invention is proposed.

[0172] In this embodiment, before step E, the insurance liability analysis method further includes:

[0173] Step G, obtaining sample insurance clause documents, and classifying the sample insurance clause documents by product name;

[0174] In this embodiment, a process of obtaining a preset corpus is provided, which is specifically as follows:

[0175] First, obtain the sample insurance clause documents, and classify the sample insurance clause documents by product name. When classifying, you can first obtain the product name of the sample insurance clause documents, and then classify the sample insurance clause documents by product name, so as to classify the same product into one category.

[0176] Step H, performing cluster analysis on the contents of each part of the classified insurance clause sample document to obtain a clustering result;

[0177] Then, cluster analysis is performed on the various parts of the classified insurance clause sample documents to obtain clustering results. When performing cluster analysis, clustering algorithms based on partitioning (such as k-means clustering algorithm), hierarchical clustering algorithms (such as Chameleon Floyd algorithm, AGNES (Agglomerative NE Sting, bottom-up aggregation algorithm), CURE (Clustering Using RE Presentatives), etc.), density-based clustering algorithms, grid-based clustering algorithms, model-based clustering algorithms, and fuzzy-based clustering algorithms can be used. Of course, to facilitate clustering, feature information of each part of the insurance clause content of each type of product can be extracted separately, such as word vectors or sentence vectors, and then cluster analysis can be performed based on the extracted feature information to obtain clustering results.

[0178] Step I: labeling various types of insurance clauses according to the clustering results to obtain preset labels, and statistically analyzing the values ​​corresponding to the preset labels in the various types of insurance clauses to obtain value features;

[0179] After obtaining the clustering results, the contents of various insurance clauses are labeled according to the clustering results to obtain preset labels. When labeling, the labeling methods include but are not limited to: 1) manual labeling, that is, sending the clustering results and the contents of various insurance clauses to the work end, so that the staff can set the corresponding labels according to the clustering results; 2) machine labeling, that is, matching the clustering results with the preset multi-type labels, and selecting the corresponding labels as preset labels according to the matching results; 3) combining manual labeling with machine labeling. While labeling, the values ​​corresponding to the preset labels in the contents of various insurance clauses are statistically analyzed to obtain value features.

[0180] For example, the subdivision label of upper limit of payment period may include the upper limit of single payment days, the upper limit of payment days in a single policy year, the upper limit of cumulative payment days during the validity period of the contract, the upper limit of payment days in each guaranteed renewal period, the upper limit of cumulative payment days for multiple insurance benefits in a single policy year, the upper limit of cumulative payment days for multiple insurance benefits during the validity period of the contract, the upper limit of cumulative payment days for the same reason during the validity period of the contract, and the upper limit of cumulative payment days for the same reason in a single policy year. The value features of these are all combinations of numbers and letters, and the letters are year / month / day.

[0181] As an implementation method, various types of insurance clauses may be labeled directly according to the clustering results, and the labels of all types of products may be used as preset labels.

[0182] Furthermore, the step of “labeling various insurance clauses according to the clustering results to obtain preset labels” includes:

[0183] Step I1, labeling various insurance clauses according to the clustering results to obtain initial labels;

[0184] Step I2, counting the initial tags according to the product types corresponding to the initial tags, and clustering and de-duplication processing the initial tags according to the statistical results to obtain preset tags.

[0185] As another implementation method, the content of various insurance clauses can be labeled according to the clustering results to obtain initial labels, which are the collection of the contents of various insurance clauses of various product types; then, the initial labels are counted according to the product types corresponding to the initial labels, that is, the labels of different product types are counted separately to obtain labels of various types of products, and then the initial labels are clustered and deduplicated according to the statistical results to merge labels with the same meaning but different names in various types of products to avoid multiple expressions of the same thing, and then the final preset labels can be obtained. In the above manner, the preset labels obtained can be made more refined to avoid duplication and similar situations.

[0186] Step J: constructing the preset corpus according to the preset tags and the value features.

[0187] After obtaining the preset labels and their corresponding features, a preset corpus is constructed based on the preset labels and value features.

[0188] In this embodiment, by counting and refining a large number of insurance clauses, a corpus is finally constructed, that is, a label indicator system of multi-dimensional insurance clause characteristics, so as to facilitate the subsequent labeling of insurance clauses and their information, labeling and structuring the insurance clauses, and facilitating the subsequent analysis of insurance liability.

[0189] Furthermore, based on the fifth to seventh embodiments described above, an eighth embodiment of the insurance liability analysis method of the present invention is proposed.

[0190] In this embodiment, before step S33, the method further includes:

[0191] Step K, detecting whether there is a table in the target insurance clause content;

[0192] In this embodiment, since the table is not conducive to the subsequent labeling process, it is possible to first detect whether there is a table in the target insurance clause content before inputting the target insurance clause content into the labeling model for processing.

[0193] If it exists, execute step L to obtain the row, column and dimension information of the table in the target insurance clause content;

[0194] Step M1: if the dimension of the table is one-dimensional, the row and column information are connected according to a first preset expression to obtain the processed target insurance clause content;

[0195] Step M2: if the dimension of the table is multi-dimensional, the row and column information are connected according to a second preset expression to obtain the processed target insurance clause content;

[0196] If a table exists in the target insurance clause content, the row and column information and dimensions of the table in the target insurance clause content are obtained, wherein the row and column information is recorded as the information displayed in each row and each table.

[0197] If the dimension of the table is one-dimensional, the row and column information is connected and processed according to the first preset expression to obtain the processed table content, and then the processed table and other content are used as the processed target insurance clause content. Among them, the first preset expression can be (illustrated as including 2 fields): table identifier, field 1! [dimension field corresponding to field 1] = field 2! [dimension field corresponding to field 2], where "!" only serves as a separator, and the table identifier can be determined according to the type of the table, which is used to distinguish that the expression here is a table. For example, the assignment type table can be set to table, and the display type table can be set to gentable. Fields 1 and 2 correspond to the table fields of the first row or first column, and the dimension field corresponding to field 1 or 2 is the value of the row or column corresponding to field 1 or 2. In addition, in order to further facilitate the subsequent labeling processing, each field can be standardized and converted. Specifically, it can be converted according to the mapping relationship between the preset non-standard field and the standardized field. Correspondingly, the first preset expression can be (illustrated as including 2 fields): table identifier, field 1! Standardized field 1! [dimension field corresponding to field 1] = field 2! Standardized field 2! [Dimension field corresponding to field 2].

[0198] For example, if the table is as shown in Table 1 below, the conversion can be obtained: gentable insurance policy year (see Explanation 2)! (Policy year) [1st year, 2nd year, 3rd year, 4th year and later] = percentage of basic insurance amount! Proportional coefficient 1! [10%, 20%, 40%, 100%].

[0199]

[0200] Table 1 One-dimensional table

[0201] If the dimension of the table is multi-dimensional, the row and column information are connected according to the second preset expression to obtain the processed table content, and then the processed table and other contents are used as the processed target insurance clause content.

[0202] The second preset expression may be (illustrated as including three fields): table identifier, field 1! [dimension field corresponding to field 1] # field 2! [dimension field corresponding to field 2] = field 3! [value of field 3 corresponding to fields 1 and 2], where "#" only serves as a separator, and the table identifier may be determined according to the type of the table, and is used to distinguish that the expression is a table. For example, an assignment type table may be set to table, and a display type table may be set to gentable. Fields 1, 2, and 3 correspond to the table fields of the first row and first column, and the value of field 3 corresponding to fields 1 and 2 is the value of the row or column corresponding to field 1 or 2, and the dimension field corresponding to field 3 is the value corresponding to field 3 under fields 1 and 2. In addition, to further facilitate the subsequent labeling processing, each field may be standardized and converted. Specifically, the conversion may be performed according to the mapping relationship between the preset non-standard field and the standardized field. Correspondingly, the second preset expression may be (illustrated as including three fields): table identifier, field 1! standardized field 1! [dimension field corresponding to field 1] # field 2! standardized field 2! [Dimension field corresponding to field 2] = field 3! Standardized field 3! [Values ​​of field 3 corresponding to fields 1 and 2] In addition, the second preset expression can also represent the dimension fields corresponding to each field in the form of a matrix based on the above.

[0203] For example, if the table is as shown in Table 2 below, it can be converted to: table Insurance period! Insurance period! [10 years, 15 years] # Payment period! Payment period! [3 years, 5 years, 10 years] = Annual survival benefit payment ratio! Proportional coefficient 1! [9%, 14%, 18%, 10%, 15%, 25%]. It can also be converted to: Insurance period ×Payment period =Annual survival benefit payment ratio [9%, 14%, 18%, 10%, 15%, 25%].

[0204]

[0205] Step S33 includes:

[0206] The processed target insurance clause content is labeled using a pre-trained labeling model to obtain a labeling result.

[0207] After obtaining the processed target insurance clause content, the processed target insurance clause content is labeled using a pre-trained labeling model to obtain a labeling result. The specific execution process can refer to the above embodiment and will not be described in detail here.

[0208] In this embodiment, by performing dimensionality reduction processing on the table, it is converted into corresponding expressions according to the table dimensions to achieve standardized expression of the table, thereby facilitating subsequent processing.

[0209] Furthermore, based on the above embodiments, a ninth embodiment of the insurance liability analysis method of the present invention is proposed.

[0210] In this embodiment, after the above step S40, the insurance liability analysis method further includes:

[0211] Step N, obtaining the product name of the to-be-identified policy image from the target standard policy field and its target policy field value, and determining a target display method according to the product name;

[0212] Step O, displaying the responsibility analysis result based on the target display method.

[0213] In this embodiment, after obtaining the insurance liability analysis result, the product name of the policy image to be identified is obtained from the target standard policy field and its target policy field value, and the target display method is determined according to the product name. The target display method can be determined according to the preset mapping relationship between the product name and the display method, and the display method can include but is not limited to hierarchical display, display by cash flow method, and display according to a preset template.

[0214] Then, the liability analysis results are displayed based on the target display method. For example, for the display of non-annuity products, it can be divided into three layers for display. The first is the liability basis display, which only generates a compensation phrase containing the protection liability data; the second is the display of the compensation conditions, liability payment and liability analysis results, which generates a statement containing the compensation conditions, protection liability data and compensation restrictions; the third is to consult the database and extract the original text of the terms of the liability. For the display of annuity products, if the liability type is annuity, the liability analysis results are displayed in a cash flow manner. The starting point of the cash flow is the policy effective date, and the end point is the policy termination date. The cash flow nodes and amounts can be queried from the above-mentioned target standard policy fields and their target policy field values, the target annotation labels and their target information to obtain the corresponding values ​​for filling.

[0215] The present invention also provides a computer-readable storage medium, on which an insurance liability analysis program is stored. When the insurance liability analysis program is executed by a processor, the steps of the insurance liability analysis method described in any of the above embodiments are implemented.

[0216] The specific embodiments of the computer-readable storage medium of the present invention are basically the same as the embodiments of the above-mentioned insurance liability analysis method, and will not be described in detail here.

[0217] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0218] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0219] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0220] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for analyzing insurance liability, It is characterized in that The insurance liability analysis method comprises the following steps: Obtain the insurance policy image to be identified and the insurance clause document to be parsed; Processing the insurance policy image to be identified to obtain a target standard insurance policy field and a target insurance policy field value; Processing the insurance clause document to be parsed based on a pre-trained position annotation model and a pre-trained label annotation model to obtain a target annotation label and target information thereof, wherein the target information includes an insurance liability parsing formula; According to the insurance liability analysis formula, the target parameter value is obtained from the target standard policy field and its target policy field value and the target annotation label and its target information, and the target parameter value is substituted into the insurance liability analysis formula to obtain the liability analysis result, wherein, if the insurance liability analysis formula is: basic insurance amount*payment period*1%*[1+3%*(number of policy years-3)], the target parameter value to be obtained includes the basic insurance amount, payment period and number of policy years; The step of processing the insurance clause document to be parsed based on the pre-trained position annotation model and the pre-trained label annotation model to obtain the target annotation label and its target information includes: Inputting the insurance clause document to be parsed into a pre-trained position annotation model to obtain a position annotation result; According to the position marking result, the insurance clause document to be parsed is intercepted to obtain the target insurance clause content; Using a pre-trained labeling model to label the target insurance clause content, obtaining a labeling result, wherein the labeling result includes the target label and its corresponding original information, wherein the original information includes the liability information to be parsed; The original information is structured to obtain target information corresponding to each target annotation label, wherein the target information includes an insurance liability analysis formula, and the insurance liability analysis formula is obtained by structured processing of the liability information to be analyzed.

2. The insurance liability analysis method according to claim 1, It is characterized in that The step of processing the to-be-identified insurance policy image to obtain the target standard insurance policy field and its target insurance policy field value comprises: Identify the policy image to be identified to obtain initial policy information, wherein the initial policy information includes identification characters and position information thereof; According to the preset mapping relationship between the non-standard policy field and the standard policy field, the initial policy field and its corresponding standard policy field are matched from the identification characters; The field value characteristics corresponding to the standard policy field are obtained, and the recognition characters other than the initial policy field are matched according to the field value characteristics and the position information to obtain the target policy field value corresponding to the standard policy field.

3. The insurance liability analysis method according to claim 2, It is characterized in that Before the step of matching the initial policy field and its corresponding standard policy field from the identification characters according to the preset mapping relationship between the non-standard policy field and the standard policy field, the method further includes: Acquire a sample policy image and its corresponding first product type and insurance clause information, identify the sample policy image, and obtain sample policy field information; Acquire a first target policy field corresponding to the first product type, and determine a first non-standard policy field and a first standard policy field corresponding to each first product type according to the insurance clause information; Obtain a second standard policy field according to the first target policy field and the first standard policy field, and perform statistical analysis on the policy sample field information according to the first product type and the second standard policy field to obtain a statistical analysis result; Based on the statistical analysis results, the first non-standard policy field and the first standard policy field, a mapping relationship between the preset non-standard policy fields and the standard policy fields is constructed; wherein the mapping relationship between the preset non-standard policy fields and the standard policy fields includes a sub-mapping relationship between the non-standard policy fields and the standard policy fields corresponding to each first product type.

4. The insurance liability analysis method according to claim 3, It is characterized in that The step of matching the initial policy field and its corresponding standard policy field from the recognition characters according to the preset mapping relationship between the non-standard policy field and the standard policy field comprises: According to the mapping relationship between the preset non-standard policy fields and the standard policy fields, the product field and its corresponding standard product field are matched from the identification characters; Determine a target sub-mapping relationship corresponding to the standard product field from the mapping relationship between the preset non-standard policy field and the standard policy field; According to the target sub-mapping relationship and the specific policy field value characteristics obtained based on the insurance terms information, the remaining policy field and its corresponding standard remaining policy field are matched from the recognition characters, wherein the standard policy field includes the standard product field and the standard remaining policy field.

5. The insurance liability analysis method according to claim 1, It is characterized in that Before the step of labeling the target insurance clause content using the pre-trained labeling model to obtain the labeling result, the step further includes: Acquire a first training sample set, wherein the first training sample set includes insurance clause content samples, real annotation labels and real information thereof, wherein the real annotation labels and real information thereof are obtained based on annotations of a preset corpus; Training a preset label annotation model using the first training sample set to obtain a trained label annotation model; Wherein, the preset label annotation model includes an information extraction layer and a classification layer; The step of training the preset label annotation model by using the first training sample set to obtain a trained label annotation model comprises: Inputting the insurance clause content samples into the information extraction layer for information extraction to obtain feature information corresponding to each insurance clause content sample; Converting the feature information into a feature vector, inputting the feature vector into the classification layer to obtain a predicted label, and determining corresponding prediction information according to the predicted label and the feature information; Calculate the loss value based on the predicted label, the predicted information, the real label of the insurance clause content sample and its real information; The parameters of the preset label annotation model are updated by a gradient descent algorithm according to the loss value, and iterative training is performed based on the first training sample set to obtain a trained label annotation model.

6. The insurance liability analysis method according to claim 5, It is characterized in that The insurance liability analysis method further includes: Obtaining sample insurance clause documents, and classifying the sample insurance clause documents by product name; Perform cluster analysis on the contents of each part of the classified insurance clause sample documents to obtain clustering results; Labeling various types of insurance clauses according to the clustering results to obtain preset labels, and statistically analyzing the values ​​corresponding to the preset labels in the various types of insurance clauses to obtain value features; The preset corpus is constructed according to the preset tags and the value features.

7. The insurance liability analysis method according to any one of claims 1 to 6, It is characterized in that The insurance liability analysis method further includes: Acquire the product name of the to-be-identified policy image from the target standard policy field and its target policy field value, and determine the target display method according to the product name; The responsibility analysis result is displayed based on the target display method.

8. An insurance liability analysis device, It is characterized in that The insurance liability analysis device includes: a memory, a processor, and an insurance liability analysis program stored in the memory and executable on the processor. When the insurance liability analysis program is executed by the processor, the steps of the insurance liability analysis method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores an insurance liability analysis program, and when the insurance liability analysis program is executed by the processor, the steps of the insurance liability analysis method according to any one of claims 1 to 7 are implemented.

Images