CN117078434A - Method, system, equipment and storage medium for inquiring policy renewal difficulty - Google Patents
Method, system, equipment and storage medium for inquiring policy renewal difficulty Download PDFInfo
- Publication number
- CN117078434A CN117078434A CN202310945890.2A CN202310945890A CN117078434A CN 117078434 A CN117078434 A CN 117078434A CN 202310945890 A CN202310945890 A CN 202310945890A CN 117078434 A CN117078434 A CN 117078434A
- Authority
- CN
- China
- Prior art keywords
- policy
- difficulty
- policy file
- model
- files
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000003860 storage Methods 0.000 title claims abstract description 14
- 230000010354 integration Effects 0.000 claims abstract description 50
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 33
- 239000013598 vector Substances 0.000 claims abstract description 33
- 230000004044 response Effects 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 53
- 238000003066 decision tree Methods 0.000 claims description 30
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 238000009826 distribution Methods 0.000 claims description 13
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000003062 neural network model Methods 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 10
- 230000006870 function Effects 0.000 description 16
- 238000013515 script Methods 0.000 description 10
- 230000004913 activation Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000007774 longterm Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000013077 scoring method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000004806 packaging method and process Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/08—Insurance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/242—Query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2455—Query execution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Accounting & Taxation (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Finance (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention provides a method, a system, equipment and a storage medium for inquiring policy renewal difficulty, belonging to the technical field of financial insurance data processing. The method comprises the following steps: determining a queried policy file list in response to a query instruction of user equipment, wherein the query instruction carries a service personnel identifier and information for specifying a renewal time period; respectively inputting the policy files in the policy file list into a tree integration model to extract leaf node coding vectors of the policy files; the leaf node coding vectors of the policy files are respectively input into a sorting algorithm model to obtain difficulty grading values corresponding to the policy files; and returning the policy file list and the difficulty score value to the user equipment so that the user equipment presents each policy file in the policy file list based on the difficulty score value. The method and the device can be used for acquiring the policy renewal difficulty information.
Description
Technical Field
The invention relates to the technical field of financial insurance data processing, in particular to a method for inquiring policy renewal difficulty, a system for inquiring policy renewal difficulty, electronic equipment and a machine-readable storage medium.
Background
The insurance policy (policy) is a written proof that an insurance person and an applicant sign an insurance contract, and includes information such as insurance period, insurance fee, etc. In electronic applications, insurance companies often record policy files via a business database. The insurance business personnel need to check the insurance policy file and contact the customer regularly to remind the customer of the payment period of the insurance policy, so that the loss of the required guarantee of the customer caused by too late payment time is avoided, but the difficulty of completing the renewal of each insurance policy is different due to the difference of the insurance policy information and the customer requirement, and the insurance company often regulates and controls the business resources of the company according to the difference of the insurance policy renewal difficulty so as to improve the charge completion rate.
Currently, the renewal difficulty of insurance policy is usually manually determined and reported to insurance companies. Professional personnel check the insurance policy needing to be paid and judge the renewal difficulty level of the insurance policy according to personal experience, however, the manual judgment is difficult to cover various types of insurance policy files in a service database, key information can be omitted, the renewal difficulty level judgment results of different personnel on the insurance policy files are very different, when customer service is carried out, the manual judgment results of renewal difficulty levels are difficult to provide effective difficulty information for service personnel, and help is difficult to provide for improving customer service of insurance companies.
Disclosure of Invention
The invention aims to provide a method, a system, equipment and a storage medium for inquiring policy renewal difficulty, which avoid the problem that the accuracy and the effectiveness of information are difficult to ensure when the policy renewal difficulty level is judged manually, so that the information meeting the requirements of clients is difficult to provide for business personnel, thereby realizing the automatic judgment of the policy renewal difficulty level and providing the effective difficulty information of respective business associated policies for the business personnel.
In order to achieve the above object, the present specification adopts the following scheme:
in a first aspect, an embodiment of the present invention provides a method for querying a policy renewal difficulty, where the method includes:
determining a queried policy file list in response to a query instruction of user equipment, wherein the query instruction carries service personnel identification and information of a designated duration period, and the query instruction is used for indicating fields of a policy file to be queried and a data structure of the policy file list;
the policy files in the policy file list are respectively input into a tree integration model to extract leaf node coding vectors of the policy files, wherein the tree integration model comprises a plurality of decision trees formed based on the characteristics determined by the attributes of the business personnel and the characteristics determined by the attributes of the policy;
The leaf node coding vectors of the policy files are respectively input into a sorting algorithm model to obtain difficulty grading values corresponding to the policy files, wherein the difficulty grading values are used for indicating the duration payment difficulty between the policy files in the policy file list;
and returning the policy file list and the difficulty score value to the user equipment so that the user equipment presents each policy file in the policy file list based on the difficulty score value.
In a second aspect, an embodiment of the present invention provides a system for querying a policy renewal difficulty, where the system includes:
the inquiry module is used for responding to an inquiry instruction of the user equipment to determine an inquired policy file list, wherein the inquiry instruction carries service personnel identification and information of a designated duration period, and the inquiry instruction is used for indicating fields of a policy file to be inquired and a data structure of the policy file list;
the decision module is used for respectively inputting the policy files in the policy file list into a tree integration model to extract leaf node coding vectors of the policy files, and the tree integration model comprises a plurality of decision trees formed based on the characteristics determined by the business personnel attributes and the characteristics determined by the policy attributes;
The evaluation module is used for respectively inputting leaf node coding vectors of the policy files into the sorting algorithm model to obtain difficulty grading values corresponding to the policy files, wherein the difficulty grading values are used for indicating the duration payment difficulty between the policy files in the policy file list;
and the return module is used for returning the policy file list and the difficulty score value to the user equipment so that the user equipment presents each policy file in the policy file list based on the difficulty score value.
In a third aspect, an embodiment of the present invention provides an electronic device, including:
at least one processor;
a memory coupled to the at least one processor;
wherein the memory stores instructions executable by the at least one processor, the at least one processor implementing the aforementioned methods by executing the memory-stored instructions.
In a fourth aspect, embodiments of the present invention provide a machine-readable storage medium storing machine instructions that, when executed on a machine, cause the machine to perform the aforementioned method.
In the invention, the service personnel inquires and determines the associated policy file list of the service personnel through the user equipment, and each policy file in the policy file list is determined through the service personnel identification carried by the inquiry command and the information of the appointed duration period during inquiry, so that the server receiving the inquiry command carries out difficulty assessment on the service personnel and the policy files associated with the time range in the service database instead of carrying out difficulty assessment on all the policy files in the service database, thereby avoiding the influence of the unassociated service personnel on the policy duration difficulty and reducing the resource expense when the server uses a model to process data. And respectively inputting the policy files in the policy file list into a tree integration model, extracting leaf node coding vectors of the policy files, and judging whether the policy is successful or not according to the tree integration model, so that the attribute characteristics of service personnel and the importance association degree of the attribute characteristics of the policy files can be decided by the tree integration model, and different characteristic expressions of the policy files of the service personnel can be obtained. And then, leaf node coding vectors of the policy files are respectively input into a sorting algorithm model, the difficulty score values among the policy files in the policy file list can be represented by the probability value output by the sorting algorithm model, so that the policy file list and the difficulty score values of the business personnel in the appointed duration range are obtained, the renewal difficulty difference among the policy files associated with the business personnel in the appointed duration range can be represented, and the method has the characteristics of accuracy and effectiveness compared with the difficulty assessment of all the policy files. The user equipment presents the policy files in the policy file list based on the difficulty grading values, so that the policy file renewal difficulty information of the business personnel using the user equipment is relatively differentially represented in the policy file list and is consistent with the attribute characteristics of the business personnel and the policy attribute characteristics, the policy renewal difficulty information is effective, and the renewal difficulty information can be beneficial to the business personnel to perform business operation conforming to the customer requirements on the policy associated with the business personnel in the renewal time range so as to improve the customer service of the insurance business.
Additional features and advantages of embodiments of the invention will be set forth in the detailed description which follows.
Drawings
The accompanying drawings are included to provide a further understanding of embodiments of the invention, and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain, without limitation, the embodiments of the invention. In the drawings:
FIG. 1 is a schematic diagram of the main steps of the method according to the embodiment of the present invention;
FIG. 2 is a tree structure diagram of an exemplary tree integration model according to an embodiment of the invention;
FIG. 3 is a schematic layer structure of an exemplary RankNet model according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an exemplary interaction scenario between an application server and a database according to an embodiment of the present invention;
fig. 5 is a schematic block diagram of an exemplary electronic device according to an embodiment of the present invention.
Detailed Description
For the purposes of clarity, technical solutions and advantages of the present specification, the following detailed description of the embodiments of the present invention will be given with reference to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating and illustrating the invention, are not intended to limit the invention.
As described above, the policy renewal difficulty level is manually judged to be defective. The rules of the charging difficulty in the renewal period are set according to the human experience, so that the rules possibly deviate from the background of the actual business, and deviation often exists in the actual production and use process. Through the attribute analysis of the insurance policy, an artificial intelligent machine model can be adopted to predict the difficulty value of each insurance policy file, wherein the predicted difficulty value is determined by the attribute characteristics of the insurance policy and represents the information of the renewal difficulty among all insurance policies. However, the prediction of the difficulty value is affected by service personnel who do not associate the policy, the difficulty of the policy renewal is actually different from service personnel who do the customer service, the difficulty difference between the associated policies corresponding to the service personnel who do the customer service is not relatively expressed, and the policy renewal difficulty values obtained by the overall prediction of all the policy files are distributed to the service personnel who do the customer service, the policy renewal difficulty values do not accord with the attribute characteristics of the service personnel who do the customer service, the information of the predicted difficulty values is difficult to be effective when the customer service is performed, and the information support is difficult to be provided for the customer service of an insurance company.
In view of this, the present disclosure provides a scheme for querying policy renewal difficulty, which can provide effective policy renewal difficulty information for service personnel who need to perform customer service, without simultaneously evaluating difficulty for all policy files. The inquiry policy renewal difficulty may be implemented by a server, and the server may receive an inquiry instruction sent by the user device, where the inquiry instruction may be used to indicate a field of a policy file to be inquired and a data structure of a policy file list. The server may determine the queried policy file list by querying the information specifying the renewal time period and the service person identification in the instructions in response to the query instructions. After the policy file list is obtained, the policy files in the policy file list can be input into a tree integration model to obtain coding vectors with the property characteristic expression capability of service personnel and policies, the coding vectors are input into a sorting algorithm model to obtain the difficulty grading value of the policy files, the difficulty grading value and the policy file list can be returned to user equipment, so that the service personnel can obtain the policy files in the policy file list presented by the user equipment based on the difficulty grading value, obtain the difficulty information expression of the relative difference between the policy files, be beneficial to the customer service of the service personnel, improve the charging completion rate of insurance companies, and have economic values. It should be appreciated that the methods provided herein may be performed by a device having computing, instruction processing, and communication capabilities, such as a server or electronic device.
In a first aspect, referring to fig. 1, an embodiment of the present invention provides a method for querying a policy renewal difficulty, which may be applied to a server, and the server may be referred to as a difficulty assessment server, and the method may include:
s1) determining a queried policy file list in response to a query instruction of user equipment, wherein the query instruction carries service personnel identification and information of a designated duration period, and the query instruction is used for indicating fields of a policy file to be queried and a data structure of the policy file list.
In an embodiment of the present invention, policy renewal may refer to a fee that is paid for a long term (e.g., greater than 1 year) of the policy to complete a renewal (or renewal) and/or a short term (e.g., less than or equal to 1 year) of the policy to complete a renewal (or renewal) over a period of time, where the last time of the period of time may be before the end of the grace period or before a specified expiration time in the policy. The policy renewal difficulty is a measurement result, such as a difficulty grading value, obtained by the server calculating the policy file to estimate how easily the policy corresponding to the policy file completes renewal or renewal payment after the business personnel performs customer service.
In some possible implementations, the difficulty assessment server may be a physical server or a server instance or container instance, and the hardware of the instance may be a resource instance in the server cluster, which is composed of processor resources and memory resources, and has computing and instruction processing functions and communication capabilities, such as a cloud server. The difficulty assessment server is deployed with scripts and/or programs, and can realize the method for inquiring the policy renewal difficulty when executing the scripts and/or programs. The difficulty assessment server may be configured with or in communication with a business database that records policy files to obtain the policy files from the business database. The difficulty assessment server may communicate with the user device to receive query instructions sent by the user device and return a list of policy files and difficulty scoring values in response to the query instructions. The user equipment can be electronic equipment, such as mobile electronic equipment, a computer and the like, and can be used for responding to user operation of business personnel before customer service is carried out, sending a query instruction to the difficulty assessment server, presenting each policy file in a policy file list based on the difficulty grading value, and presenting the policy files according to relative difficulty differences, so that the business personnel can be helped to carry out customer service. The business personnel can carry out the work of customer service by themselves according to the regulations or can be regulated and assigned by an insurance company to carry out the work of customer service, the business personnel carrying out the customer service can be the signer recorded in the policy file list, the business personnel carrying out the customer service can also be the customer service assigned to carry out the policy file in the policy file list, and the customer service can comprise services such as payment period reminding, payment mode description, policy information description and the like.
In some possible applications, the query instruction may be a database operation instruction, or may be structured interface data for the difficulty assessment server to determine the database operation instruction. The service personnel identifier carried in the query instruction may be a unique identifier identifiable by a machine, for example, the service personnel identifier may be an employee number or an identity identifier of the service personnel in an insurance company, or may be an identifier of user equipment used by the service personnel. The designated duration period carried in the query instruction may be a default designated period or a period customized by the service personnel, the designated period may be 1 week, two weeks, 1 month, etc., the customized period may be 15 days, 45 days, etc., and the number of policy files associated with the service personnel in the designated duration period in the service database may be at least two, for example, 5, 10, 20, 50, etc. The designated duration time period and the business personnel identifier in the query instruction can be used as field information for indicating the field of the policy file to be queried. The query may further indicate a data structure of the policy file list, for example, a data column in the policy file list representing a field or attribute information of each policy file may include a sequence number, a policy file name, an access address of the policy file, policy file attribute information, service personnel attribute information, and the like, and in some exemplary applications, the data structure may further include a data column of a difficulty score value, so that the policy file list and the difficulty score value may be transmitted to the user device as a data packet of one list.
In the embodiment of the invention, the difficulty evaluation server can analyze the information carrying the service personnel identification and the appointed duration period in the query instruction, and generate the queried policy file list by querying the service database. The aforementioned step S1) may include:
s101) inquiring and determining service personnel attribute information corresponding to a field of the service personnel identifier and a full-quantity policy file corresponding to the service personnel identifier based on the service personnel identifier in the inquiry instruction;
s102) filtering the full-quantity policy file according to a time stamp field to reserve the policy file in a specified duration period of the query instruction;
s103) marking the attribute information of the service personnel to the reserved policy file to generate a policy file list.
In some possible implementations, the business database may record business person attribute information for the business person identification and a timestamp for each policy file, and may record the business person identification associated with each policy file. The difficulty assessment server may use the parsed business person identification to query from the business database for business person attribute information corresponding to fields of the business person identification and a full policy file corresponding to the business person identification, the full policy file including policy files for policy renewal at each time period (the policy files may be associated by an insurance company authorization assignment and/or the business person ticket). The difficulty evaluation server can obtain the policy file in the appointed duration time period through the database operation instruction and the filtering according to the time stamp field. The difficulty evaluation server may mark the business person attribute information into a data column of the business person attribute field corresponding to each reserved policy file, and may generate a policy file list through the queried policy file and data structure, for example, mark (access address of) the queried policy file attribute information and the policy file into a data column of the policy file attribute field and a field (access address field) containing a policy file electronic document path, respectively, so that the business person attribute characteristics and the attribute characteristics of the policy file may be used simultaneously when estimating the difficulty score value. It should be noted that, in the embodiment of the present invention, the label may be recording the attribute information to be labeled corresponding to the name attribute information or the unique identification attribute information of the policy file.
In an embodiment of the present invention, the difficulty assessment server may be configured with a file (script and/or program) of a trained tree integration model, which may include a plurality of decision trees formed based on features determined by business person attributes and features determined by policy attributes. In the training aspect of the tree integration model, the difficulty assessment server can be configured with training scripts, a computer language call library and an operating environment, wherein the call library is used for providing function packaging support of program methods and/or functions and the like, and the operating environment can provide dependency relations required by program and/or script execution. When the training script is executed, a training method of a tree integration model may be implemented, and the training method of the tree integration model may include:
t1) acquiring attribute information of a history policy file and attribute information of service personnel from a service database, and taking renewal information in the attribute information of the history policy file as a target variable, wherein the renewal information comprises information of whether policy renewal is carried out in a grace period or information of whether policy renewal is carried out in a policy-specified failure time;
t2) taking the history policy file as a training sample, and training through an XGBOOST model to obtain a tree integration model; wherein each layer of leaf nodes of each decision tree in the tree integration model is a feature selected based on policy attribute information and service personnel attribute information in the training sample; the selected feature is a feature determined by a business person attribute or a feature determined by a policy attribute.
In some possible implementations, the attributes of the policy file may include attributes of premium, insurance duration (long/short term), insurance category, etc., and the insurance category may include personal and group insurance, property insurance, personal insurance, etc. The business person attributes may include attributes such as marginal contribution rate in the last year, equity premium for the last three years, failure rate in the last three years, attendance rate in the last year, and business person job level. It will be appreciated that the specific attributes are not limiting implementations and may be selected and designed based on the actual circumstances of the test, the effect of use, and the needs of the company.
In some possible implementations, after each policy file is recorded in the service database, a text serialization tool and a word segmentation tool may be used to extract keywords of the foregoing attributes from text contents of each policy file, and the text contents corresponding to the foregoing attribute keywords may be extracted as attribute information of each policy file, so that attribute information of each policy file may be recorded in the service database corresponding to each policy file. Periodic information acquisition and statistics can be carried out on the business personnel, and business personnel attribute information corresponding to the business personnel identification is recorded in a business database. The attribute information of the policy file can be marked to the corresponding policy file, the attribute information of the business personnel can be marked (randomly or proportionally distributed) to the policy file, one policy file marked with the attribute information of the policy file and the attribute information of the business personnel can be taken as a sample, a sample set is formed, and as a training sample, the training sample is randomly split into a training set and a verification set according to a certain proportion in order to realize the characteristic distinguishing and stability of the policy file relative to each attribute determination, and meanwhile, the time period of expiration time or invalidation time of the policy file of the same batch training is kept consistent, for example, the time period is 1 day, 3 days, 1 week and the like.
In some possible implementations, the difficulty assessment server may perform data preprocessing in order to guarantee model training effects. The difficulty evaluation server may perform data exploration and outlier processing on the obtained historical policy file by calling a program method or function in the library to obtain information such as the size, the dimension (the number of attributes) and the like of the training sample data through statistics, so as to determine whether an outlier exists (for example, when the expiration time of the grace period of the policy file as a sample exceeds the time period of the sample set of the batch, the abnormal value is removed). The difficulty evaluation server may also perform data cleaning and selection by calling a program method or a function in the library, for example, a missing value processing function (which may interpolate samples), a continuous feature box processing method (program method) for dividing a numerical range in attribute information, a class feature encoding method (program method) for obtaining an encoded value representation of text, data feature screening (for example, feature selection using a packaged classifier, and alternative use of the same class of features), etc., so that the sample set may have a numerical representation of a policy file and attribute information, various identifiers, numerical range information, and encoded values corresponding to the attribute information, etc., and the attribute information and the policy file in step T1) may each be represented by encoded values formed by values 0 and/or 1, for example, using a one-hot encoding method.
In some possible implementations, after the data preprocessing is completed, the renewal information in the attribute information of the history policy file may be used as a target variable, for example, whether the target variable is a policy renewal in a grace period or not, and multiple decision trees, that is, a tree integration model, may be obtained through training by using the foregoing training samples and the encapsulated code representation of the XGBOOST model, where each decision tree may be a binary tree structure and each decision tree may include multiple layers of leaf nodes. The foregoing attributes may be described as a plurality of features in a decision condition, and a greedy algorithm or weighted score method may be used to determine the selected feature at each leaf node location during the training process, e.g., for greedy algorithms, traversing the feature at each leaf node location, determining that the objective function value is the smallest, and selecting whether the current feature is suitable as the current leaf node. The text corresponding to the coded value on each leaf node may be a decision condition (i.e., description of the feature) determined by the business person attributes such as "whether the attendance rate is greater than 90%", "whether the attendance rate is greater than 50%", "whether the marginal contribution rate is greater than 10%", "whether the job level is greater than the job level threshold", and the attribute of the insurance file (i.e., description of the feature) "whether the premium is greater than 5000 yuan", "whether the policy category is a group", "whether the policy is a long-term insurance greater than 1 year", etc. The XGBOOST model is a gradient ascending algorithm model of a training tree integrated model, the difference between a predicted value and a true value is reduced by continuously increasing a new tree fitting residual, and the predicted value (output by a leaf node) of any decision tree can represent the influence estimation size of the feature serving as the leaf node relative to a target variable when the current policy file is taken as an input sample.
In some possible examples, determining decision trees in multiple iterations, please refer to fig. 2, for the first decision tree (subtree 1), with respect to the guaranteed renewal in the grace period, when the i (i is a positive integer) th policy file is taken as an input sample (may be the name of the policy file/the unique identification attribute information and the coded value of the tag and the extracted attribute information), the decision condition (represented by the circle filled with oblique lines) in the first decision tree (subtree 1) is that the "attendance rate is greater than 90%" and the estimated size of the effect output by the leaf nodes (represented by the circle filled with the lattice) is the predicted value y, for example, the attribute information of the sample may determine that the attendance rate is greater than 90%, y may be 0.50 (the predicted value classified as "yes" side), and the real value corresponding to the attribute information of the i th policy file is 0.27, and the output value y 1 (residual value) may be 0.23, wherein the predicted value may also be the predicted value of a multi-layered leaf node, e.g. the firstThe decision condition corresponding to the layer leaf node is that whether the attendance rate is more than 90%, and the decision condition corresponding to the second layer leaf node is that whether the job class is higher than a job class threshold when the attendance rate is more than 90%; for the second decision tree (subtree 2), when the ith policy file is taken as an input sample, the estimated magnitude of the influence of the leaf node output under the decision condition of "whether the policy is higher than 5000 yuan" in the second decision tree (subtree 2), i.e., the predicted value (y-y) 1 =0.27), and the real value corresponding to the attribute information of the i-th policy file is 0.20, the output value y 2 (residual value) may be 0.07, and so on until subtree n is determined, predicted value (y-y 1 -y 2 ……-y n-1 ) Output value y n (n is a positive integer) in the sum of the predicted valuesAnd (2) taking the decision trees at the moment as the tree integration model trained in the step T2) when the difference between the true values corresponding to the 'guaranteed single renewal in the grace period' is minimum.
In the embodiment of the invention, the trained tree integration model and the list of the policy files can be used to obtain the coding vector of the policy file. The method for inquiring the policy renewal difficulty can comprise the following steps:
s2) respectively inputting the policy files in the policy file list into a tree integration model to extract leaf node coding vectors of the policy files, wherein the tree integration model comprises a plurality of decision trees formed based on the characteristics determined by the attributes of the service personnel and the characteristics determined by the attributes of the policy.
In some possible implementations, the encoding vector may be a vector composed of values 0/1, and may represent, as input, a distribution in the tree integration model of features corresponding to each attribute with respect to the target variable "guaranteed renewal", when the policy file (name attribute information) and the marked and extracted attribute information are included, the distribution being a distribution based on a degree of association of the features of the policy file with the importance of the target variable "guaranteed renewal", and may be an index position distribution of "yes" or "not" in a decision condition corresponding to leaf nodes in the tree integration model. Step S2) may include:
S201), encoding the policy file (name attribute information and tag and extracted attribute information), and inputting the encoded value into a tree integration model, wherein one encoded value is a vector (composed of 0/1) of one numerical feature corresponding to one attribute information of the policy file;
s202) determining leaf nodes allocated to the current coding value in the tree integration model, namely determining the leaf nodes corresponding to the decision condition when the current coding value accords with the decision condition, and obtaining the position index value of the allocated leaf nodes in the tree integration model;
s203) converts the position index value into a code vector composed of numerical identification values 0/1, i.e., a code vector extracted from the tree integration model.
In some possible implementations, illustratively, when the current policy file is taken as input, 4 encoding values (x 1, x2, x3, x 4) may be obtained, i.e., 4 features, each consisting of a 0/1 sequence, the tree integration model may include two decision trees of the same hierarchy, each tree having three leaf nodes and each tree may have two decision conditions, e.g., a first decision tree decision condition of "job level is above a job level threshold" and "job rate is above 90% when job level is above a job level threshold", a second decision tree decision condition of "whether premium is above 5000 yuan" and "policy is long-term insurance greater than 1 year when premium is above 5000 yuan". The position index values of the first layer of leaf nodes of each decision tree are all 0 (namely, the first leaf node in each decision tree indicates that the first decision condition is not achieved), the second layer of leaf nodes are distributed left and right (the left leaf node can be made to be the second leaf node, the right leaf node is made to be the third leaf node, and the first decision condition is achieved and the second decision condition is not achieved respectively), and the position index values are respectively 1 and 2. After being input to the tree integration model, 4 code values will be assigned to one leaf node position of each decision tree, e.g., 4 code values assigned to the left leaf node of the second level of the first decision tree (the second leaf node of the first decision tree) and the right leaf node of the second level of the second decision tree (the third leaf node of the second decision tree). At this time, a method encapsulated in the XGBoost model, for example, an apply () method, may be called, the position index values of the two leaf nodes in the tree integration model are obtained as [1,2], the two index values are converted into encoding vectors, for example, may be converted into [0,1,0,0,0,1], where 1 represents "yes" in the current policy file achievement decision condition, that is, the decision condition belonging to the leaf node, 0 represents "no" in the current policy file non-achievement decision condition, that is, the decision condition not belonging to the leaf node, and the encoding vector [0,1,0,0,0,1] may represent that the attendance rate is greater than 90% when the job level is higher than the job level threshold, and simultaneously represents that the policy is not a long-term insurance greater than 1 year when the policy is higher than 5000 yuan.
In the embodiment of the invention, the diversity between the paired policy files can be determined by using a pairing method, so that when the method is used, the difficulty scoring value is obtained by representing the score (probability value) of the relative ordering. The pairing contrast scoring method can comprise a scoring method of support vector machines such as a Ranking SVM and an IR SVM, and a model scoring method of a Ranking boost distribution model, a GBrank regression model, a rank Net neural network model and the like. The difficulty assessment server may be configured with a file (script and/or program) of a trained ranking algorithm model, which is a neural network model obtained by training a neural network double-tower model; the neural network model is a single tower model, and the layer structure of the neural network model is the same as the layer structure of any one of the neural network double tower models. The neural network double-tower model is the RankNet model. Referring to FIG. 3, in the training side, the dual-tower structure in the RankNet model may include an input layer, a hidden layer, and an output layer, the input layer may be used to receive the encoded vectors of policy files i and j, and the hidden layer may include a data layer z i And z j Two full connection layers 1 and two full connection layers 2, an input layer and a hidden layer of a double-tower layer structureThe layers do not cross, in the output layer, the difference value between the output values of the full connection layer 2 can be input into an activation function, and the probability value representing the difference of the paired policy files can be obtained through the output value of the activation function.
In terms of training the ranking algorithm model, the difficulty assessment server may be configured with a training script, a computer language call library for providing functional packaging support for program methods and/or functions, and a runtime environment that may provide dependencies required for program and/or script execution. When the training script is executed, a ranking algorithm model may be obtained. In the actual production process, because of the numerous insurance policies, a great deal of time is often required to be spent on making sample pairs for all insurance policy files in a business database, and the situation of insufficient memory of a server can occur. The method for obtaining the sorting algorithm model can comprise the following steps:
r1) forming a plurality of pairs of policy files based on the policy files corresponding to the same business personnel identifier and the same duration period in the history policy files;
R2) taking the identification values of the paired policy files as target values, wherein the identification values are used for indicating the difficulty of continuous payment among the paired policy files within a specified time range;
r3) extracting leaf node coding vectors of paired policy files from the tree integrated model to serve as a training set;
r4) training a neural network double-tower model based on the training set and the target value, and taking any one of the trained neural network double-tower models as a sequencing algorithm model.
In some possible implementations, pairs of policy files may be extracted from the tree integration model as training sets, where the number of training sets may be 1/2 of the product of the number of policy files for training, C, and (C-1), where C is a positive integer, i.e., C× (C-1)/2 file pairs. In some possible examples, the encoding vectors of the first policy file and the second policy file of the pair of policy files may be extractedb i ,b j And can configure the target value f (b) i ,b j )：
In this equation, the identification value may include a first flag value 1, a second flag value 1/2, and a third flag value 0. When the difficulty score value of the first policy file for the continuous payment is higher than that of the second policy file for the continuous payment, the target value is 1; when the difficulty score value of the first policy file continuous payment is equal to the difficulty score value of the second policy file continuous payment, the target value is 1/2; and when the difficulty score value of the first policy file continuous payment is smaller than that of the second policy file continuous payment, the target value is 0. The loss value between the probability value output by the RankNet model and the target value can be determined by calling a cross entropy function or a loss function, and further iteration training is carried out based on the magnitude of the loss value. After the RankNet model training is completed, any one of the double-tower models can be selected as a layer structure of the ordering algorithm model. The test set of the policy file may be used to score and call an AUC index evaluation function to calculate an AUC index score of the ranking algorithm model, and the ranking algorithm model is evaluated according to the AUC index score to determine whether further sample training is required.
In the embodiment of the invention, the obtained sorting algorithm model can be used for receiving the coding vector of the policy file extracted from the trained tree integration model so as to obtain the difficulty grading value of the policy file. The method for inquiring the policy renewal difficulty further comprises the following steps:
s3) respectively inputting leaf node coding vectors of the policy files into a sorting algorithm model to obtain difficulty grading values corresponding to the policy files, wherein the difficulty grading values are used for indicating the duration payment difficulty between the policy files in the policy file list.
In the embodiment of the invention, the layer structure of the sorting algorithm model may include an input layer, a data layer, a full connection layer 1 and a full connection layer 2 for receiving the encoding vectors of the single policy file, and an activation function as an output layer, wherein the input value of the activation function may be the output value of the full connection layer 2, the output value of the activation function may be a probability value, the probability value is the difficulty score value output by the difficulty assessment server, the greater the probability value is, the greater the difficulty score value is, the higher the difficulty of realizing the policy renewal of the policy file is, otherwise, the smaller the probability value is, the smaller the difficulty score is, and the lower the difficulty of realizing the policy renewal of the policy file is.
In the embodiment of the invention, the difficulty evaluation server can respond to the query instruction of the user equipment to perform data return operation. The method for inquiring the policy renewal difficulty further comprises the following steps:
s4) returning the policy file list and the difficulty score value to the user equipment, so that the user equipment presents each policy file in the policy file list based on the difficulty score value.
In some possible implementations, the difficulty assessment server may generate a data column for recording a difficulty score value in the policy file list in response to an instruction of a query instruction of the user equipment, where the method for querying the policy renewal difficulty before returning the policy file list and the difficulty score value to the user equipment may further include:
m1) based on the difficulty grading value, adjusting the ordering of all the policy files in the policy file list.
The sequence numbers of the policy files corresponding to the difficulty scoring values in the policy file list can be adjusted according to the order of the difficulty scoring values from large to small, so that the user equipment can display the policy files with larger difficulty in the policy file list before the policy files with smaller difficulty without further processing the difficulty scoring values and the policy file list. Therefore, the server can automatically provide effective difficulty information for business personnel, and provide support for improving the difficulty information for the customer service of the insurance company. Referring to fig. 4, a business person a performing customer service in a first period of time may send a query instruction to a difficulty assessment server through user operation, where the difficulty assessment server may obtain a policy file list 1 through database instruction operation and model processing, in the policy file list 1, the difficulty assessment server may adjust each policy file in the policy file list 1 based on the size of the difficulty score, and the difficulty assessment server may return the policy file list 1 (which includes the difficulty score at this time, for easy observation, and may be converted into a tenth difficulty score for presentation, and it may be understood that this may be an alternative implementation manner, for example, may also be converted into a percentile) to the user device 1, so as to provide policy renewal difficulty information required for performing customer service to the business person a. And a second business person for customer service in a second time period can send a query instruction to the difficulty assessment server through user operation, the difficulty assessment server can obtain a policy file list 2 through database instruction operation and model processing, in the policy file list 2, the difficulty assessment server adjusts each policy file in the policy file list 2 based on the difficulty grading value, and the difficulty assessment server can return the policy file list 2 (containing the difficulty grading value at the moment) to the user equipment 2 so as to provide policy renewal difficulty information required by customer service for the second business person. In contrast, it is found that the policy file list 1 includes the policy file 0035 (the four digits may be the unique identifier of the file), the policy file 2035, the policy file 1197, and the policy file 1042, and the policy file list 2 includes the policy file 1115, the policy file 1455, the policy file 2417, and the policy file 2035, both of which have the same policy file 2035 (bold in fig. 4), but the difficulty score of the policy file 2035 is changed with respect to different business persons and different time periods (for example, the first time period is within the grace period of the second renewal and the second time period is within the grace period of the third renewal), and the difficulty score 1.4 of the policy file 2035 in the policy file list 2 is lower than the difficulty score 6.5 in the policy file 1, so that the associated policy renewal difficulty information of the policy file can be differentially expressed with respect to the business persons currently performing customer service within the current time period.
In some possible implementations, the user device may present the policy files in the policy file list differently on the basis of step M1) or separately (without adjustment by the difficulty assessment server). In a first example, the user device may determine a range of difficulty score values to which the difficulty score values corresponding to the policy files pertain, each of which may correspond to a text color. For example, when the difficulty score value corresponding to the first policy file belongs to the first difficulty score value range, the file name and the unique identifier corresponding to the first policy file can be presented in red text; when the difficulty scoring value corresponding to the second policy file belongs to the second difficulty scoring value range, the file name and the unique identifier corresponding to the second policy file can be presented in blue text; when the difficulty score value corresponding to the third policy file belongs to a third difficulty score value range, the file name and the unique identifier corresponding to the third policy file can be presented in a green text, wherein the minimum difficulty score value in the first difficulty score value range is higher than the maximum difficulty score value in the second difficulty score value range, and the minimum difficulty score value in the second difficulty score value range is higher than the maximum difficulty score value in the third difficulty score value range. In a second example, the user device may determine a difficulty score value range to which a difficulty score value corresponding to each policy file belongs, where each difficulty score value range may correspond to an operation text of a client service proposed for execution, and the operation text may include texts such as a notification reminder of a phone/sms/mail/application, a number of reminders, and no reminders at present, where the client difficulty score value is less than (under ten minutes) 2 hours, which may indicate that the client policy has a strong guarantee requirement, that policy payment is easy to occur, and that the user may concentrate on other clients having a requirement, so that help information for executing the client service may be provided to a service person.
It should be noted that, in order to ensure the stability of the application of the model in the difficulty assessment server in actual production, the model of the difficulty assessment server may be scored for stability monitoring. The method for inquiring the policy renewal difficulty further comprises the following steps:
s5) determining a group stability index value of an overall model formed by the tree integration model and the sorting algorithm model based on the difficulty score value distribution corresponding to the training sample and the difficulty score value distribution corresponding to the policy file list;
s6) determining the stability level of the overall model based on the population stability index value and a specified threshold range.
The numerical value of the group stability index (Population Stability Index, PSI) can be used for observing whether the overall model of the difficulty evaluation server needs to be subjected to model iteration and updating, and the group stability index can measure the distribution change of the model result scores on training samples and recent samples and reflects the stability of score distribution. If the PSI value is smaller than 0.1, the overall model of the difficulty evaluation server does not need to be adjusted, and the overall model can be indicated to be very stable; if the PSI value is between 0.1 and 0.2, the overall model stability of the difficulty evaluation server is changed, and the overall model stability can be reduced; if the group stability index PSI value is greater than 0.2, the overall model stability of the difficulty assessment server is problematic, which may indicate that an optimization model needs to be adjusted, for example, a new sample set is formed using a recent (e.g., within a year) policy file, and training is performed.
According to the embodiment of the invention, the tree integration model and the sorting algorithm model are deployed on the difficulty assessment server, and the policy file list and the difficulty score value can be returned to the user equipment simultaneously in response to the query instruction sent by the user equipment, so that service personnel using the user equipment can obtain the policy file list presented by the user equipment, the policy files in the policy file list are presented relatively differently based on the difficulty score value, the automatic query of the difficulty score value can be performed, and the policy renewal difficulty information of the relative difference between the associated policy files is provided for the service personnel performing the customer service, thereby realizing the provision of effective difficulty information which is beneficial to the customer service and providing support of the difficulty information for improving the customer service of insurance companies.
In a second aspect, the embodiment of the present invention further provides a system for inquiring about policy renewal difficulty under the same inventive concept as the previous embodiment, where the system may include:
the inquiry module is used for responding to an inquiry instruction of the user equipment to determine an inquired policy file list, wherein the inquiry instruction carries service personnel identification and information of a designated duration period, and the inquiry instruction is used for indicating fields of a policy file to be inquired and a data structure of the policy file list;
The decision module is used for respectively inputting the policy files in the policy file list into a tree integration model to extract leaf node coding vectors of the policy files, and the tree integration model comprises a plurality of decision trees formed based on the characteristics determined by the business personnel attributes and the characteristics determined by the policy attributes;
the evaluation module is used for respectively inputting leaf node coding vectors of the policy files into the sorting algorithm model to obtain difficulty grading values corresponding to the policy files;
and the return module is used for returning the policy file list and the difficulty score value to the user equipment so that the user equipment presents each policy file in the policy file list based on the difficulty score value.
Specifically, determining the queried policy file list in response to the query instruction of the user equipment includes:
based on the service personnel identification in the query instruction, querying and determining service personnel attribute information corresponding to a field of the service personnel identification and a full-quantity policy file corresponding to the service personnel identification;
filtering the full-quantity policy file according to a time stamp field to reserve the policy file in a specified duration period of the query instruction;
And marking the attribute information of the service personnel to the reserved policy file to generate a policy file list.
Specifically, the system may further include: the training module of the tree integration model is used for:
acquiring attribute information of a history policy file and attribute information of service personnel from a service database, and taking renewal information in the attribute information of the history policy file as a target variable, wherein the renewal information comprises information of whether to renew payment in a grace period;
taking the history policy file as a training sample, and training through an XGBOOST model to obtain a tree integration model;
each layer of leaf nodes of each decision tree in the tree integration model is a feature selected based on the policy attribute information and the business person attribute information in the training sample;
the selected feature is a feature determined by a business person attribute or a feature determined by a policy attribute.
Specifically, the sorting algorithm model is a neural network model obtained by training a neural network double-tower model;
the neural network model is a single tower model, and the layer structure of the neural network model is the same as the layer structure of any one of the neural network double tower models.
Specifically, the system may further include: the obtaining module of the sorting algorithm model is used for:
based on the policy files corresponding to the same business personnel identifier and the same duration time period in the history policy files, forming a plurality of pairs of policy files;
the identification values of the paired policy files are used as target values, and the identification values are used for representing the difficulty level of continuous payment among the paired policy files in a specified time range;
extracting leaf node coding vectors of paired policy files from the tree integration model to serve as a training set;
and training a neural network double-tower model based on the training set and the target value, and taking any one of the trained neural network double-tower models as a sorting algorithm model.
Specifically, the identification value comprises a first marked value, a second marked value and a third marked value;
the first identification value, the second identification value and the third identification value are respectively used for indicating that the difficulty score value of the first policy file continuous payment in the grace period is higher than, equal to and smaller than the difficulty score value of the second policy file continuous payment;
the first policy file and the second policy file are the paired policy files.
Specifically, the neural network double-tower model is a RankNet model.
Specifically, before returning to the user equipment, the method further includes:
and adjusting the ordering of all the policy files in the policy file list based on the difficulty grading value.
Specifically, the system may further include: the index determining module is used for:
determining a group stability index value of an overall model formed by the tree integration model and the sorting algorithm model based on the difficulty score value distribution corresponding to the training sample and the difficulty score value distribution corresponding to the policy file list;
a stability level of the overall model is determined based on the population stability index value and a specified threshold range.
In a third aspect, an embodiment of the present invention further provides an electronic device under the same inventive concept as the previous embodiment, including: at least one processor; a memory coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the at least one processor implementing the methods of the previous embodiments by executing the instructions stored by the memory. Referring to fig. 5, an exemplary electronic device is provided, and the internal structure of the electronic device may be a server, an industrial personal computer, a terminal device, a microcontroller, etc. as shown in fig. 5. The electronic device comprises a processor A01, a network interface A02 and a memory which are connected through a bus. Wherein the processor a01 of the electronic device is adapted to provide computing, instruction processing and control capabilities. The memory of the electronic device includes a memory storage a03 and a nonvolatile storage medium a04. The nonvolatile storage medium a04 stores an operating system B01 and a computer program B02. The memory a03 provides an environment for the operation of the operating system B01 and the computer program B02 on the nonvolatile storage medium a04. The network interface a02 of the electronic device is used for communication with a network. The computer program B02, when executed by the processor a01, implements the method in the foregoing embodiments.
In a fourth aspect, embodiments of the present application also provide a machine-readable storage medium having stored thereon machine instructions which, when executed on a machine, cause the machine to perform the method of the previous embodiments.
The foregoing details of the optional implementation of the embodiment of the present application have been described in detail with reference to the accompanying drawings, but the embodiment of the present application is not limited to the specific details of the foregoing implementation, and various simple modifications may be made to the technical solution of the embodiment of the present application within the scope of the technical concept of the embodiment of the present application, and these simple modifications all fall within the protection scope of the embodiment of the present application.
In addition, the specific features described in the above embodiments may be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, various possible combinations of embodiments of the present application are not described in detail.
Those skilled in the art will appreciate that all or part of the steps in implementing the methods of the embodiments described above may be implemented by a program stored in a storage medium, including instructions for causing a single-chip microcomputer, chip or processor (processor) to perform all or part of the steps of the methods of the embodiments described herein. While the aforementioned storage medium may be non-transitory, the storage medium may include: a U-disk, a hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a Flash Memory (Flash Memory), a magnetic Memory, an optical Memory, or other various media capable of storing program codes.
In addition, any combination of various embodiments of the present invention may be performed, so long as the concept of the embodiments of the present invention is not violated, and the disclosure of the embodiments of the present invention should also be considered.
Claims (12)
1. A method of querying policy renewal difficulty, the method comprising:
determining a queried policy file list in response to a query instruction of user equipment, wherein the query instruction carries service personnel identification and information of a designated duration period, and the query instruction is used for indicating fields of a policy file to be queried and a data structure of the policy file list;
the policy files in the policy file list are respectively input into a tree integration model to extract leaf node coding vectors of the policy files, wherein the tree integration model comprises a plurality of decision trees formed based on the characteristics determined by the attributes of the business personnel and the characteristics determined by the attributes of the policy;
the leaf node coding vectors of the policy files are respectively input into a sorting algorithm model to obtain difficulty grading values corresponding to the policy files;
and returning the policy file list and the difficulty score value to the user equipment so that the user equipment presents each policy file in the policy file list based on the difficulty score value.
2. The method of claim 1, wherein the determining the queried policy file list in response to the query instruction of the user device comprises:
based on the service personnel identification in the query instruction, querying and determining service personnel attribute information corresponding to a field of the service personnel identification and a full-quantity policy file corresponding to the service personnel identification;
filtering the full-quantity policy file according to a time stamp field to reserve the policy file in a specified duration period of the query instruction;
and marking the attribute information of the service personnel to the reserved policy file to generate a policy file list.
3. The method of claim 1, wherein the training method of the tree integration model comprises:
acquiring attribute information of a history policy file and attribute information of service personnel from a service database, and taking renewal information in the attribute information of the history policy file as a target variable, wherein the renewal information comprises information of whether to renew payment in a grace period;
taking the history policy file as a training sample, and training through an XGBOOST model to obtain a tree integration model;
Each layer of leaf nodes of each decision tree in the tree integration model is a feature selected based on the policy attribute information and the business person attribute information in the training sample;
the selected feature is a feature determined by a business person attribute or a feature determined by a policy attribute.
4. The method of claim 3, wherein,
the sorting algorithm model is a neural network model obtained by training a neural network double-tower model;
the neural network model is a single tower model, and the layer structure of the neural network model is the same as the layer structure of any one of the neural network double tower models.
5. The method of claim 4, wherein,
the method for obtaining the sorting algorithm model comprises the following steps:
based on the policy files corresponding to the same business personnel identifier and the same duration time period in the history policy files, forming a plurality of pairs of policy files;
the identification values of the paired policy files are used as target values, and the identification values are used for representing the difficulty level of continuous payment among the paired policy files in a specified time range;
Extracting leaf node coding vectors of paired policy files from the tree integration model to serve as a training set;
and training a neural network double-tower model based on the training set and the target value, and taking any one of the trained neural network double-tower models as a sorting algorithm model.
6. The method of claim 5, wherein,
the identification value comprises a first marked value, a second marked value and a third marked value;
the first identification value, the second identification value and the third identification value are respectively used for indicating that the difficulty score value of the first policy file continuous payment in the grace period is higher than, equal to and smaller than the difficulty score value of the second policy file continuous payment;
the first policy file and the second policy file are the paired policy files.
7. The method of claim 4, wherein,
the neural network double-tower model is a RankNet model.
8. The method of querying policy renewal difficulty according to claim 1, wherein prior to returning to the user device, the method further comprises:
and adjusting the ordering of all the policy files in the policy file list based on the difficulty grading value.
9. The method of querying policy renewal difficulty according to claim 3, further comprising:
determining a group stability index value of an overall model formed by the tree integration model and the sorting algorithm model based on the difficulty score value distribution corresponding to the training sample and the difficulty score value distribution corresponding to the policy file list;
a stability level of the overall model is determined based on the population stability index value and a specified threshold range.
10. A system for querying a policy renewal difficulty, the system comprising:
the inquiry module is used for responding to an inquiry instruction of the user equipment to determine an inquired policy file list, wherein the inquiry instruction carries service personnel identification and information of a designated duration period, and the inquiry instruction is used for indicating fields of a policy file to be inquired and a data structure of the policy file list;
the decision module is used for respectively inputting the policy files in the policy file list into a tree integration model to extract leaf node coding vectors of the policy files, and the tree integration model comprises a plurality of decision trees formed based on the characteristics determined by the business personnel attributes and the characteristics determined by the policy attributes;
The evaluation module is used for respectively inputting leaf node coding vectors of the policy files into the sorting algorithm model to obtain difficulty grading values corresponding to the policy files;
and the return module is used for returning the policy file list and the difficulty score value to the user equipment so that the user equipment presents each policy file in the policy file list based on the difficulty score value.
11. An electronic device, comprising:
at least one processor;
a memory coupled to the at least one processor;
wherein the memory stores instructions executable by the at least one processor, the at least one processor implementing the method of any one of claims 1 to 9 by executing the instructions stored by the memory.
12. A machine readable storage medium storing machine instructions which, when run on a machine, cause the machine to perform the method of any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310945890.2A CN117078434A (en) | 2023-07-28 | 2023-07-28 | Method, system, equipment and storage medium for inquiring policy renewal difficulty |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310945890.2A CN117078434A (en) | 2023-07-28 | 2023-07-28 | Method, system, equipment and storage medium for inquiring policy renewal difficulty |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117078434A true CN117078434A (en) | 2023-11-17 |
Family
ID=88701413
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310945890.2A Pending CN117078434A (en) | 2023-07-28 | 2023-07-28 | Method, system, equipment and storage medium for inquiring policy renewal difficulty |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117078434A (en) |
-
2023
- 2023-07-28 CN CN202310945890.2A patent/CN117078434A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112633962B (en) | Service recommendation method and device, computer equipment and storage medium | |
| CN111915366B (en) | User portrait construction method, device, computer equipment and storage medium | |
| US20150286949A1 (en) | Problem analysis and priority determination based on fuzzy expert systems | |
| CN107798341A (en) | User view Forecasting Methodology, electronic equipment and computer-readable recording medium | |
| CN111950889A (en) | Client risk assessment method and device, readable storage medium and terminal equipment | |
| CN108038655A (en) | Recommendation method, application server and the computer-readable recording medium of department's demand | |
| CN112734161A (en) | Method, equipment and storage medium for accurately identifying empty-shell enterprises | |
| CN113807940B (en) | Information processing and fraud recognition method, device, equipment and storage medium | |
| CN111127185A (en) | Credit fraud identification model construction method and device | |
| CN115668265A (en) | Risk transfer configurator and simulation engine providing look-ahead and look-behind metrics for risk driven combinations of manipulating and adjusting underwriting objectives and method thereof | |
| CN109345371A (en) | Personal reference report backtracking method and system | |
| CN113947476A (en) | Method, device, equipment and storage medium for identifying lost-message client | |
| CN113535848A (en) | Block chain-based credit investigation grade determination method, device, equipment and storage medium | |
| CN111353728A (en) | Risk analysis method and system | |
| CN111652471A (en) | List distribution control method and device, electronic equipment and storage medium | |
| CN117078434A (en) | Method, system, equipment and storage medium for inquiring policy renewal difficulty | |
| CN116701584A (en) | Intelligent question-answering method and device based on electricity user portrait and electronic equipment | |
| CN115293867A (en) | Financial reimbursement user portrait optimization method, device, equipment and storage medium | |
| CN112712270B (en) | Information processing method, device, equipment and storage medium | |
| CN112182320B (en) | Cluster data processing method, device, computer equipment and storage medium | |
| CN111008204B (en) | Quotation library processing method and device, storage medium and server | |
| CN113076317B (en) | Big data-based data processing method, device, equipment and readable storage medium | |
| CN113722602A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN116308807A (en) | Model updating method and device, nonvolatile storage medium and electronic equipment | |
| CN114331595A (en) | Artificial intelligence based product recommendation method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |