A clustering analysis method for group insurance policies based on neural network encoder

The neural network encoder processed group insurance policy data, determined the initial outlier point and adjusted the clustering algorithm parameters, solved the problem of low clustering analysis efficiency caused by high dimensions and noise of group insurance policy data, and achieved more accurate and reliable clustering Class results.

CN118035768BActive Publication Date: 2024-11-05上海栈略数据技术有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410062451.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-16
Publication Date
2024-11-05
Estimated Expiration
2044-01-16

AI Technical Summary

Technical Problem

The high dimensions and noise of group insurance policy data lead to poor accuracy and low efficiency of cluster analysis.

Method used

The cluster analysis method of group insurance policy based on neural network encoder is adopted. By processing group insurance policy data to output semantic vectors, the initial outliers are determined, the neighborhood radius and minimum sample points of the clustering algorithm are adjusted, and the noise evaluation value is evaluated. Correct the adjustment coefficients to improve the accuracy and reliability of clustering analysis.

Benefits of technology

It improves the accuracy and efficiency of cluster analysis of group insurance policies, avoids noise interference, ensures the integrity of cluster structure, reduces the error of manual judgment, and improves the reliability of clustering results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118035768B_ABST
    Figure CN118035768B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of financial service technology, and in particular to a group insurance policy clustering analysis method based on a neural network encoder, comprising: processing group insurance policy data according to a neural network encoder to output a semantic vector of the group insurance policy data; determining a label for an initial outlier according to a data evaluation value of a single group insurance policy data; determining a neighborhood radius and a minimum number of sample points of a clustering algorithm; performing clustering analysis on the semantic vectors of a number of group insurance policy data according to a clustering algorithm; determining an adjustment to the neighborhood radius of the clustering algorithm according to a silhouette coefficient of a clustering analysis result; determining the eligibility of the clustering analysis result according to a noise evaluation value of the clustering analysis result; and determining a correction to an adjustment coefficient according to the number of outliers in the clustering analysis result. The present invention overcomes the problem in the prior art that the clustering analysis of the group insurance policy data has a high dimension and a large noise, resulting in a low accuracy and low efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of financial services, and in particular to a group insurance policy clustering analysis method based on a neural network encoder. Background Art

[0002] In group insurance (group insurance) business, cluster analysis of insurance policies is an important task, which can help insurance companies better understand the characteristics of customer groups, optimize product design, formulate differentiated pricing strategies, and improve the efficiency of risk management. However, group insurance policy data is usually high-dimensional and complex, which makes it difficult for traditional clustering methods to effectively handle. In addition, group insurance policy data may contain outliers, which will interfere with the clustering results, thereby affecting the accuracy and reliability of clustering.

[0003] Chinese Patent Publication No.: CN108460689A discloses a policy analysis method, device, terminal device and storage medium, including: obtaining policy information of each policy from a policy database, wherein the policy information includes policy identification information, attribute information of a policy object and identity information of an insurance salesperson, wherein the policy identification information is used to uniquely identify the policy, and the identity information is used to uniquely identify the insurance salesperson; analyzing the policy information, associating the policies corresponding to the policy information having the same value of the attribute information based on the attribute information, and constructing a policy relationship network; setting a unique sub-network number for each policy relationship network; writing the sub-network number of each policy relationship network into the personal information of the insurance salesperson involved in the policy relationship network according to the identity information, wherein the personal information includes the identity information; obtaining the personal information of each insurance salesperson from a salesperson database; analyzing the personal information, associating the insurance salespersons having the same value of the personal information based on the personal information, and constructing a salesperson relationship network, including: determining a second to-be-matched network according to the personal information matching parameters; for each of the second parameters to be matched, traversing the insurance salesperson, establishing a second direct relationship based on the second parameter to be matched between different insurance salespersons with the same second parameter value, the second parameter value is used to identify the second direct relationship; using the identity identification information of the insurance salesperson as the second network node, using the second parameter value of the second parameter to be matched as the second relationship node, associating the second network nodes having the second direct relationship with each other with the second relationship node establishing the second direct relationship, and constructing the salesperson relationship network; analyzing the insurance fraud risk of the insurance salesperson according to the salesperson relationship network, including: obtaining the second relationship node in the salesperson relationship network; if the second relationship node includes the sub-network number, calculating the number of nodes of the second network node associated with each sub-network number; if the number of nodes of the second network node associated with the sub-network number exceeds a preset number threshold, confirming the second network node associated with the sub-network number as a risk network node, and obtaining the risk policy relationship network corresponding to the sub-network number; outputting the personal information of the insurance salesperson corresponding to the risk network node and the risk policy relationship network;

[0004] It can be seen that the existing technology has the following problems: due to the high dimension and high noise of group insurance policy data, the cluster analysis of group insurance policy data has poor accuracy and low efficiency. Summary of the invention

[0005] To this end, the present invention provides a group insurance policy clustering analysis method based on a neural network encoder to overcome the problem in the prior art that the high dimension and high noise of group insurance policy data lead to poor accuracy and low efficiency of clustering analysis of group insurance policy data.

[0006] To achieve the above object, the present invention provides a group insurance policy clustering analysis method based on a neural network encoder, comprising:

[0007] Processing the group insurance policy data according to the neural network encoder to output a semantic vector of the group insurance policy data;

[0008] Determine the marking of initial outliers according to the data evaluation value of a single group insurance policy data;

[0009] Determine the neighborhood radius and minimum number of sample points for the clustering algorithm;

[0010] Perform cluster analysis on the semantic vectors of several group insurance policy data based on clustering algorithms;

[0011] Determine the adjustment of the neighborhood radius of the clustering algorithm according to the silhouette coefficient of the clustering analysis result;

[0012] Determining a noise evaluation value of the cluster analysis result, and determining the eligibility of the cluster analysis result according to the noise evaluation value of the cluster analysis result;

[0013] The correction of the adjustment coefficient is determined according to the number of outliers in the cluster analysis results.

[0014] Furthermore, the data evaluation value of the single group insurance policy data is calculated according to the following formula:

[0015]

[0016] Among them, P represents the data evaluation value of a single group insurance policy data, A represents the distance between the semantic vector of a single group insurance policy data and the semantic vector of its most adjacent group insurance policy data, and A0 represents the average distance between the semantic vectors of several group insurance policy data.

[0017] Furthermore, determining the marking of the initial outlier based on the data evaluation value of the single group insurance policy data includes marking the single group insurance policy data as an initial outlier under the condition that the data evaluation value is greater than a preset data evaluation value.

[0018] Furthermore, determining the neighborhood radius and the minimum number of sample points of the clustering algorithm includes determining the distance from the semantic vector of each group insurance policy data except the initial outlier to the semantic vector closest to it, arranging the distances in order from large to small, and selecting the current distance with the largest absolute value of the difference between the current distance and the previous distance as the neighborhood radius of the clustering algorithm. The minimum number of sample points within the neighborhood radius of the clustering algorithm is the number of dimensions of the semantic vector of the group insurance policy data plus 1.

[0019] Further, determining the adjustment of the neighborhood radius of the clustering algorithm according to the silhouette coefficient of the cluster analysis result includes adjusting the neighborhood radius of the clustering algorithm under the condition that the silhouette coefficient is less than or equal to a preset silhouette coefficient.

[0020] Furthermore, the adjustment of the neighborhood radius of the clustering algorithm includes adjusting the neighborhood radius with a first adjustment coefficient under the condition that the relative difference is less than or equal to a preset relative difference, and adjusting the neighborhood radius with a second adjustment coefficient under the condition that the relative difference is greater than the preset relative difference, and the relative difference is determined by the silhouette coefficient and the preset silhouette coefficient.

[0021] Furthermore, the noise evaluation value is calculated according to the following formula:

[0022]

[0023] Among them, Z represents the noise evaluation value, L represents the number of outliers in the cluster analysis results, and L0 represents the initial number of outliers.

[0024] Further, determining the eligibility of the cluster analysis result according to the noise evaluation value of the cluster analysis result includes determining that the cluster analysis result is qualified under the condition that the noise evaluation value is less than or equal to a preset noise evaluation value.

[0025] Furthermore, the correction of the adjustment coefficient determined according to the number of outliers in the cluster analysis results includes correcting the adjustment coefficient under the condition that the number of outliers is less than or equal to the initial number of outliers, and correcting the minimum number of sample points under the condition that the number of outliers is greater than the initial number of outliers.

[0026] Furthermore, the correction of the adjustment coefficient includes correcting the adjustment coefficient according to the following first correction coefficient, setting

[0027]

[0028] Wherein, G1 represents the first correction coefficient;

[0029] The correction of the minimum number of sample points includes correcting the minimum number of sample points according to the following second correction coefficient:

[0030]

[0031] Wherein, G2 represents the second correction coefficient.

[0032] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention accurately evaluates the spatial position relationship between the semantic vector of a single group insurance policy data and the semantic vectors of several group insurance policy data based on the distance between the semantic vector of a single group insurance policy data and the semantic vector of its most adjacent group insurance policy data and the average distance between the semantic vectors of several group insurance policy data.

[0033] Furthermore, the present invention determines whether the single group insurance policy data is marked as an initial outlier based on the comparison result between the data evaluation value and the preset data evaluation value, so as to screen out the noise data to help maintain the integrity of the clustering structure, make the clustering tighter, and improve the accuracy and reliability of the clustering results.

[0034] Furthermore, the present invention determines the neighborhood radius of the clustering algorithm based on the distance from the semantic vector of each group insurance policy data except the initial outlier to the Kth closest semantic vector, thereby avoiding the interference of the initial outlier and determining the neighborhood radius more accurately and efficiently.

[0035] Furthermore, the present invention determines whether to adjust the neighborhood radius of the clustering algorithm based on the comparison result of the silhouette coefficient and the preset silhouette coefficient, which helps to judge whether the current clustering has reached a reasonable level to determine whether to adjust the neighborhood radius and improve the accuracy of the clustering analysis results.

[0036] Furthermore, the present invention determines an adjustment coefficient for adjusting the neighborhood radius of the clustering algorithm based on a comparison result between the first relative difference and the first preset relative difference, so as to accurately select the adjustment coefficient according to the silhouette coefficient, avoid waste of computing resources due to excessive adjustment of the neighborhood radius, and improve the adjustment efficiency of the clustering algorithm.

[0037] Furthermore, the present invention accurately evaluates the noise situation of cluster analysis according to the number of outliers in the cluster analysis result and the number of initial outliers, which is helpful for the subsequent improvement of the clustering algorithm.

[0038] Furthermore, the present invention determines whether the cluster analysis result is qualified according to the comparison result of the noise evaluation value Z of the cluster analysis result and the preset noise evaluation value Z0, thereby avoiding the problem of poor accuracy and low efficiency caused by manual judgment of the cluster analysis result.

[0039] Furthermore, when the cluster analysis result of the present invention is unqualified, the correction of the adjustment coefficient or the correction of the minimum number of sample points is determined according to the comparison result of the number of outliers in the cluster analysis result and the initial number of outliers, and the appropriate correction method is accurately determined to improve the accuracy of the cluster analysis.

[0040] Furthermore, the present invention calculates the correction coefficient according to a specific formula to correct the adjustment coefficient and the minimum number of sample points, thereby avoiding the problem of over-correction or low correction efficiency caused by manual correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a group insurance policy clustering analysis method based on a neural network encoder according to an embodiment of the present invention;

[0042] Figure 2 It is a flow chart of the neighborhood radius adjustment process of the group insurance policy clustering analysis method based on the neural network encoder according to an embodiment of the present invention;

[0043] Figure 3 The present invention is a flowchart of a clustering algorithm correction process of a group insurance policy clustering analysis method based on a neural network encoder in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; 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.

[0045] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0046] See also Figure 1 As shown, Figure 1 It is a flow chart of a group insurance policy clustering analysis method based on a neural network encoder according to an embodiment of the present invention; Figure 2 It is a flow chart of the neighborhood radius adjustment process of the group insurance policy clustering analysis method based on the neural network encoder according to an embodiment of the present invention; Figure 3 The present invention is a flowchart of a clustering algorithm correction process of a group insurance policy clustering analysis method based on a neural network encoder in an embodiment of the present invention.

[0047] The group insurance policy clustering analysis method based on a neural network encoder in an embodiment of the present invention includes:

[0048] Step S1, processing the group insurance policy data according to the neural network encoder to output a semantic vector of the group insurance policy data;

[0049] Step S2, determining the mark of the initial outlier according to the data evaluation value P of the single group insurance policy data;

[0050] Step S3, determining the neighborhood radius R and the minimum number of sample points M of the clustering algorithm;

[0051] Step S4, performing cluster analysis on the semantic vectors of a number of group insurance policy data according to a clustering algorithm;

[0052] Step S5, determining the adjustment of the neighborhood radius R of the clustering algorithm according to the silhouette coefficient X of the clustering analysis result;

[0053] Step S6, determining the eligibility of the cluster analysis result according to the noise evaluation value Z of the cluster analysis result;

[0054] Step S7, determining the correction of the adjustment coefficient Ti according to the number of outliers L in the cluster analysis result.

[0055] In an embodiment of the present invention, the neural network encoder is preferably a Seq2Seq model encoder, and the group insurance policy data includes but is not limited to the number of members, the industry of the group, the geographical location of the group, age distribution, gender ratio, occupational category, health status, number of claims, claim amount, and claim frequency.

[0056] Specifically, the data evaluation value P of a single group insurance policy data is calculated according to the following formula:

[0057]

[0058] Among them, A represents the distance between the semantic vector of a single group insurance policy data and the semantic vector of its most adjacent group insurance policy data, and A0 represents the average distance between the semantic vectors of several group insurance policy data.

[0059] Specifically, the present invention accurately evaluates the spatial position relationship between the semantic vector of a single group insurance policy data and the semantic vectors of several group insurance policy data based on the distance between the semantic vector of a single group insurance policy data and the semantic vector of its most adjacent group insurance policy data and the average distance between the semantic vectors of several group insurance policy data.

[0060] Specifically, when the data evaluation value P is calculated, it is determined whether the single group insurance policy data is marked as an initial outlier according to the comparison result between the data evaluation value P and the preset data evaluation value P0;

[0061] When P≤P0, it is determined that the single group insurance policy data is not marked as an initial outlier;

[0062] When P>P0, the single group insurance policy data is determined to be marked as an initial outlier;

[0063] In the embodiment of the present invention, the preset data evaluation value P0 is set to 0.98, and those skilled in the art may adjust the preset data evaluation value P0 according to specific circumstances.

[0064] Specifically, the present invention determines whether the single group insurance policy data is marked as an initial outlier based on the comparison result between the data evaluation value and the preset data evaluation value, so as to screen out the noise data to help maintain the integrity of the clustering structure, make the clustering tighter, and improve the accuracy and reliability of the clustering results.

[0065] Specifically, when the initial outlier is determined, the distance from the semantic vector of each group insurance policy data except the initial outlier to the Kth closest semantic vector is determined, and the distances are arranged in order from large to small. The current distance with the largest absolute value of the difference between the current distance and the previous distance is selected as the neighborhood radius R of the clustering algorithm. The minimum number of sample points within the neighborhood radius of the clustering algorithm is M.

[0066] In the embodiment of the present invention, K=2×W-1, M=W+1, where W is the number of dimensions of the semantic vector of the group insurance policy data.

[0067] Specifically, the present invention determines the neighborhood radius of the clustering algorithm based on the distance from the semantic vector of each group insurance policy data except the initial outlier to the Kth closest semantic vector, thereby avoiding the interference of the initial outlier and determining the neighborhood radius more accurately and efficiently.

[0068] Specifically, whether to adjust the neighborhood radius R of the clustering algorithm is determined according to the comparison result of the silhouette coefficient X of the clustering analysis result and the preset silhouette coefficient X0;

[0069] When X≤X0, it is determined to adjust the neighborhood radius R of the clustering algorithm;

[0070] When X>X0, it is determined that the neighborhood radius R of the clustering algorithm will not be adjusted;

[0071] In the embodiment of the present invention, the preset contour coefficient X0 is set to 0.6, and those skilled in the art may adjust the preset contour coefficient X0 according to specific circumstances.

[0072] Specifically, the present invention determines whether to adjust the neighborhood radius of the clustering algorithm based on the comparison result of the silhouette coefficient and the preset silhouette coefficient, which helps to judge whether the current clustering has reached a reasonable level to determine whether to adjust the neighborhood radius and improve the accuracy of the clustering analysis results.

[0073] Specifically, when it is determined to adjust the neighborhood radius R of the clustering algorithm, the relative difference △X between the contour coefficient X and the preset contour coefficient X0 is calculated, and the adjustment coefficient Ti for adjusting the neighborhood radius R of the clustering algorithm is determined according to the comparison result of the relative difference △X and the preset relative difference △X0, and △X=(X0-X) / X0 is set;

[0074] When △X≤△X0, it is determined that the neighborhood radius R is adjusted by the first adjustment coefficient T1;

[0075] When △X>△X0, it is determined that the neighborhood radius R is adjusted by the second adjustment coefficient T2;

[0076] The adjusted neighborhood radius is set to Rr=T1×R, or Rr=T2×R.

[0077] In the embodiment of the present invention, 0<T2<T1<1, the value of the first adjustment coefficient T1 is preferably 0.9, and the value of the second adjustment coefficient T2 is preferably 0.8. Those skilled in the art can adjust the first preset relative difference △X0, the first adjustment coefficient T1 and the second adjustment coefficient T2 according to specific circumstances.

[0078] Specifically, the present invention determines an adjustment coefficient for adjusting the neighborhood radius of the clustering algorithm based on a comparison result between the first relative difference and the first preset relative difference, so as to accurately select the adjustment coefficient according to the silhouette coefficient, avoid waste of computing resources due to excessive adjustment of the neighborhood radius, and improve the adjustment efficiency of the clustering algorithm.

[0079] Specifically, when the neighborhood radius R of the clustering algorithm is adjusted, the noise evaluation value Z of the clustering analysis result is calculated according to the following formula, and is set:

[0080]

[0081] Among them, L represents the number of outliers in the cluster analysis results, and L0 represents the initial number of outliers.

[0082] Specifically, the present invention accurately evaluates the noise situation of cluster analysis according to the number of outliers in the cluster analysis result and the number of initial outliers, which is helpful for the subsequent improvement of the clustering algorithm.

[0083] Specifically, when the noise evaluation value Z is calculated, whether the cluster analysis result is qualified is determined according to the comparison result between the noise evaluation value Z of the cluster analysis result and the preset noise evaluation value Z0;

[0084] When Z≤Z0, the cluster analysis result is determined to be qualified;

[0085] When Z>Z0, the cluster analysis result is determined to be unqualified;

[0086] In the embodiment of the present invention, the preset noise evaluation value Z0 is 0.22. The preset noise evaluation value Z0 is obtained when the difference between the number of actual outliers and the number of initial outliers is 10 or -10. Those skilled in the art can adjust the preset noise evaluation value Z0 according to the specific situation.

[0087] Specifically, the present invention determines whether the cluster analysis result is qualified according to the comparison result of the noise evaluation value Z of the cluster analysis result and the preset noise evaluation value Z0, thereby avoiding the problem of poor accuracy and low efficiency caused by manual judgment of the cluster analysis result.

[0088] Specifically, when it is determined that the cluster analysis result is unqualified, it is determined whether to modify the adjustment coefficient Ti according to the comparison result of the number of outliers L in the cluster analysis result and the initial number of outliers L0;

[0089] When L≤L0, it is determined to correct the adjustment coefficient Ti;

[0090] When L>L0, it is determined that the minimum number of sample points M is to be corrected.

[0091] Specifically, when the cluster analysis result of the present invention is unqualified, the correction of the adjustment coefficient or the correction of the minimum number of sample points is determined according to the comparison result of the number of outliers in the cluster analysis result and the initial number of outliers, and the appropriate correction method is accurately determined to improve the accuracy of the cluster analysis.

[0092] Specifically, when it is determined to correct the adjustment coefficient Ti, the first correction coefficient G1 for correcting the adjustment coefficient Ti is calculated according to the following formula and is set:

[0093]

[0094] The corrected adjustment coefficient is set to TT1 = T1 × G1, or TT2 = T2 × G1.

[0095] Specifically, when it is determined to correct the minimum number of sample points M, the second correction coefficient G2 for correcting the minimum number of sample points M is calculated according to the following formula and is set:

[0096]

[0097] The corrected minimum number of sample points M is set to Mm=M×G2.

[0098] Specifically, the present invention calculates the correction coefficient according to a specific formula to correct the adjustment coefficient and the minimum number of sample points, thereby avoiding the problem of over-correction or low correction efficiency caused by manual correction.

[0099] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A group insurance policy clustering analysis method based on a neural network encoder, characterized in that: include: Processing the group insurance policy data according to the neural network encoder to output a semantic vector of the group insurance policy data; Determine the marking of initial outliers according to the data evaluation value of a single group insurance policy data; Determine the neighborhood radius and minimum number of sample points for the clustering algorithm; Perform cluster analysis on the semantic vectors of several group insurance policy data based on clustering algorithms; Determine the adjustment of the neighborhood radius of the clustering algorithm according to the silhouette coefficient of the clustering analysis result; Determining a noise evaluation value of the cluster analysis result, and determining the eligibility of the cluster analysis result according to the noise evaluation value of the cluster analysis result; When it is determined that the cluster analysis result is unqualified, determining whether to modify the adjustment coefficient of the neighborhood radius according to the number of outliers in the cluster analysis result; Determining the eligibility of the cluster analysis result according to the noise evaluation value of the cluster analysis result includes determining that the cluster analysis result is qualified under the condition that the noise evaluation value is less than or equal to a preset noise evaluation value; Determining whether to correct the adjustment coefficient of the neighborhood radius according to the number of outliers in the cluster analysis result includes correcting the adjustment coefficient under the condition that the number of outliers is less than or equal to the initial number of outliers, and correcting the minimum number of sample points under the condition that the number of outliers is greater than the initial number of outliers; The correction of the adjustment coefficient includes correcting the adjustment coefficient according to the following first correction coefficient: , Among them, G1 represents the first correction coefficient, L represents the number of outliers in the cluster analysis results, and L0 represents the initial number of outliers; The correction of the minimum number of sample points includes correcting the minimum number of sample points according to the following second correction coefficient: , Wherein, G2 represents the second correction coefficient, L represents the number of outliers in the cluster analysis results, and L0 represents the initial number of outliers.

2. The method for clustering analysis of group insurance policies based on a neural network encoder according to claim 1, characterized in that: The data evaluation value of the single group insurance policy data is calculated according to the following formula: ; Among them, P represents the data evaluation value of a single group insurance policy data, A represents the distance between the semantic vector of a single group insurance policy data and the semantic vector of its most adjacent group insurance policy data, and A0 represents the average distance between the semantic vectors of several group insurance policy data.

3. The group insurance policy clustering analysis method based on a neural network encoder according to claim 2 is characterized in that: Determining the marking of the initial outlier based on the data evaluation value of the single group insurance policy data includes marking the single group insurance policy data as an initial outlier under the condition that the data evaluation value is greater than a preset data evaluation value.

4. The method for clustering analysis of group insurance policies based on a neural network encoder according to claim 3, characterized in that: The method of determining the neighborhood radius and the minimum number of sample points of the clustering algorithm includes determining the distance from the semantic vector of each group insurance policy data except the initial outlier to the semantic vector closest to it, arranging the distances in ascending order, and selecting the distance with the largest difference between the current distance and the previous distance as the neighborhood radius of the clustering algorithm. The minimum number of sample points within the neighborhood radius of the clustering algorithm is the number of dimensions of the semantic vector of the group insurance policy data plus 1.

5. The method for clustering analysis of group insurance policies based on a neural network encoder according to claim 4, characterized in that: Determining the adjustment of the neighborhood radius of the clustering algorithm according to the silhouette coefficient of the clustering analysis result includes adjusting the neighborhood radius of the clustering algorithm under the condition that the silhouette coefficient is less than or equal to a preset silhouette coefficient.

6. The method for clustering analysis of group insurance policies based on a neural network encoder according to claim 5, characterized in that: The adjustment of the neighborhood radius of the clustering algorithm includes adjusting the neighborhood radius with a first adjustment coefficient under the condition that the relative difference is less than or equal to a preset relative difference, and adjusting the neighborhood radius with a second adjustment coefficient under the condition that the relative difference is greater than the preset relative difference, and the relative difference is determined by the silhouette coefficient and the preset silhouette coefficient.

7. The method for clustering analysis of group insurance policies based on a neural network encoder according to claim 6, characterized in that: The noise evaluation value is calculated according to the following formula: , Wherein, Z represents the noise evaluation value.