Motorcycle insurance product personalized recommendation method and device and computer program product

Abstract

The invention belongs to the technical field of personalized recommendation of the insurance industry, and aims to solve the problems of low efficiency, difficulty in meeting increasing personalized requirements of consumers and reduction of customer satisfaction in the prior art. The invention discloses a motorcycle insurance product personalized recommendation method and device and a computer program product, and the method comprises the steps: training a neural network model in advance based on the service life of a vehicle, a historical accident record of the vehicle, driving behavior data, personnel data and other information, thereby obtaining a classification model, and carrying out the classification of users according to the classification model. The user category of the user is obtained; and finally, according to the user category of the user, matching the insurance product corresponding to the user category of the user. Therefore, according to the method, the multi-dimensional data such as the historical accident record and the driver behavior data are integrated, and the machine learning technology is combined to match insurance products, so that compared with the traditional technology, the efficiency is improved, and the personalized insurance buying demand of the user can be met.

Description

Technical Field

[0001] The present invention belongs to the technical field of personalized recommendation of insurance products, and specifically relates to a method, device and computer program product for personalized recommendation of motorcycle insurance products. Background Art

[0002] At present, the recommendation of motorcycle insurance often relies on manual sales, standardized product design and a wide-ranging marketing strategy, that is, insurance agents recommend motorcycle insurance products for the current year based on the motorcycle owner's insurance purchase record and the previous year's accident record. This method is not only inefficient and difficult to meet the growing personalized needs of consumers, but may also lead to waste of resources and reduced customer satisfaction.

[0003] At the same time, with the rapid development of big data, cloud computing and artificial intelligence technology, the insurance industry has gradually shifted from manual recommendation to intelligent recommendation, especially the application of technologies such as machine learning and deep learning, which makes it possible to extract valuable information from massive data. Therefore, based on the aforementioned shortcomings of manual recommendation of insurance products, how to provide a personalized recommendation method for motorcycle insurance products based on artificial intelligence has become an urgent problem to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device and computer program product for personalized recommendation of motorcycle insurance products, so as to solve the problems of low efficiency, difficulty in meeting the growing personalized needs of consumers and decreased customer satisfaction in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, a method for personalized recommendation of motorcycle insurance products is provided, comprising: Obtaining user information of a target user and vehicle information of a motorcycle corresponding to the target user, wherein the vehicle information includes the age of the vehicle, the vehicle model, and the historical accident record of the vehicle, and the user information includes driving behavior data and personnel data; generating insurance characteristic data of the target user based on the user information and the vehicle information; Inputting the insurance feature data into a classification model to obtain a user category of the target user, wherein the classification model is trained based on a parameter optimization algorithm and takes sample insurance feature data of several sample users as input and the user category of each sample user as output; According to the user category of the target user, a motorcycle insurance product corresponding to the user category is matched from an insurance product library as a recommended insurance product; The recommended insurance product is sent to the client corresponding to the target user to complete the personalized recommendation of motorcycle insurance products for the target user.

[0006] Based on the above disclosed content, the present invention generates sample insurance feature data of each sample user in advance through information such as the vehicle age, vehicle model, historical vehicle accident record of the motorcycles corresponding to several sample users, as well as their own driving behavior data and personnel data, and then trains the neural network model with the sample insurance feature data of each sample user as input and the user category of each sample user as output, thereby obtaining a classification model; in this way, the present invention is equivalent to generating insurance features based on data related to user and motorcycle insurance, and pre-training a classification model based on this; therefore, in actual use, it is only necessary to obtain the user information of the user and the vehicle information of the corresponding motorcycle, and then, based on the aforementioned information, generate the insurance feature data of the target user; then, the insurance feature data is input into the aforementioned trained classification model to obtain the user category of the target user; finally, according to the user category, a motorcycle insurance product corresponding to the user category can be matched from the insurance product library, thereby recommending it to the target user as a recommended insurance product.

[0007] Through the above design, the present invention pre-trains a neural network model based on information related to the user's motorcycle insurance business, such as vehicle age, vehicle model, historical vehicle accident record, driving behavior data and personnel data, to obtain a classification model, and thereby classifies insurance purchasing users, thereby obtaining a user category of insurance purchasing users; finally, the insurance product corresponding to the user category of the insurance purchasing user can be matched according to the user category of the insurance purchasing user; thus, the present invention integrates multi-dimensional data such as vehicle age, model, historical accident record, driver behavior data, etc., and combines machine learning technology to match insurance products. In this way, compared with traditional technologies, not only the efficiency is improved, but also the personalized insurance needs of users can be met. Therefore, the present invention is very suitable for large-scale application and promotion.

[0008] In a possible design, the classification model is trained in the following manner: Obtaining sample user information of several sample users and sample vehicle information of motorcycles corresponding to each sample user; Determining, based on the sample user information and the sample vehicle information of each sample user, a plurality of initial insurance characteristic indicators of each sample user and characteristic data corresponding to each initial insurance characteristic indicator; Based on the characteristic data of each initial insurance characteristic indicator, each initial insurance characteristic indicator is screened for characteristic importance, so as to screen out a plurality of actual insurance characteristic indicators from a plurality of initial insurance characteristic indicators; Using characteristic data corresponding to a plurality of actual insurance characteristic indicators of each sample user, sample insurance characteristic data of each sample user is generated, and the sample insurance characteristic data of each sample user is used to form a training data set; Based on a parameter optimization algorithm, the sample insurance feature data of each sample user in the training data set is used as input, and the user category of each sample user is used as output to train the neural network model, so as to obtain the classification model after the training is completed.

[0009] In a possible design, based on the characteristic data of each initial insurance characteristic indicator, each initial insurance characteristic indicator is subjected to a feature importance screening process, so as to screen out a plurality of actual insurance characteristic indicators from a plurality of initial insurance characteristic indicators, including: Performing feature mapping processing on the feature data corresponding to each initial insurance feature indicator to obtain mapped feature data corresponding to each initial insurance feature indicator, wherein the dimension of the mapped feature data corresponding to any initial insurance feature indicator is higher than the feature data corresponding to any initial insurance feature indicator; Using all the mapped feature data, a feature space set is formed, and a first information entropy of the feature space set is calculated; Deleting the i-th mapping feature data from the feature space set to obtain an updated feature space set; Calculating a second information entropy of the updated feature space set; Determine whether the difference between the second information entropy and the first information entropy is greater than 0; If yes, then the initial insurance characteristic index corresponding to the i-th mapped characteristic data is added to the insurance characteristic index set; Add 1 to i and delete the i-th mapped feature data from the feature space set again until i is equal to n, to obtain the final insurance feature index set, wherein the initial value of i is 1 and n is the total number of mapped feature data; Using the final insurance characteristic indicator set, a plurality of actual insurance characteristic indicators are determined.

[0010] In a possible design, the first information entropy of the feature space set is calculated, including: For the i-th mapping feature data in the feature space set, calculating the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set, wherein the target feature data set contains all the mapping feature data in the feature space set except the i-th mapping feature data; Calculating the entropy value between the i-th mapping feature data and each target feature data according to the similarity between the i-th mapping feature data and each target feature data to obtain a plurality of entropy values ​​corresponding to the i-th mapping feature data; Add 1 to i, and recalculate the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set until i equals n, and obtain multiple entropy values ​​corresponding to each mapping feature data; A plurality of entropy values ​​corresponding to each mapped feature data are summed, and a ratio between the summation result and a target parameter is used as a first information entropy of the feature space set, wherein the target parameter is the square of the total number of mapped feature data.

[0011] In a possible design, calculating the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set includes: For the mth target feature data in the target feature data set, the similarity between the i-th mapping feature data and the m-th target feature data is calculated using the following formula (1); (1) In the above formula (1), represents the similarity between the i-th mapping feature data and the m-th target feature data, Represents the k-th dimension data in the i-th mapping feature data, represents the data of the kth dimension in the target feature data, represents the standard deviation of the target sequence, A control factor represented as a constant, wherein the target sequence contains the k-th dimension data of each mapped feature data in the feature space set, and is the total dimension of the i-th mapping feature data; Correspondingly, according to the similarity between the i-th mapping feature data and each target feature data, calculating the entropy value between the i-th mapping feature data and each target feature data includes: The entropy value between the i-th mapping feature data and the m-th target feature data is calculated using the following formula (2); (2) In the above formula (2), Represents the entropy value between the i-th mapping feature data and the m-th target feature data.

[0012] In a possible design, based on a parameter optimization algorithm, the sample insurance feature data of each sample user in the training data set is used as input, and the user category of each sample user is used as output to train the neural network model, including: Initializing the model parameters of the neural network model to obtain several groups of model parameters; Using several sets of model parameters, an initial population is generated, wherein the initial population includes several individuals, and the position vector of each individual corresponds to a set of model parameters; Initialize the number of iterations t, and obtain the individual population at the t-th iteration, wherein the initial value of t is 1, and when t is 1, the individual population at the t-th iteration is the initial population; Generate a neural network model corresponding to each individual in the individual population at the t-th iteration according to the position vector of each individual in the individual population at the t-th iteration; Using the training data set to train the neural network model corresponding to each individual at the t-th iteration, and based on the loss function of each neural network model, calculating the fitness of each individual at the t-th iteration, wherein the smaller the loss function value of the neural network model of any individual at the t-th iteration, the greater the fitness value of the any individual; Based on the fitness of each individual at the tth iteration, the global optimal position vector is determined; Determine whether the iteration stop condition is met; If not, a nonlinear iteration factor at the t-th iteration is generated, and a first search coefficient and a second search coefficient at the t-th iteration are determined according to the nonlinear iteration factor; Determining whether the absolute value of the first search coefficient is less than a first threshold; If yes, generate the first random number at the t-th iteration, and perform a local position search on each individual at the t-th iteration based on the first random number at the t-th iteration, the first search coefficient, and the second search coefficient to obtain each updated individual, so as to form a new individual population using each updated individual, wherein the value interval of the first random number is [0,1]; Increment t by 1, and update the individual population at the t-th iteration to the new individual population, and regenerate the neural network model corresponding to each individual at the t-th iteration based on the position vector of each individual in the individual population at the t-th iteration, until the iteration stopping condition is met, and use the neural network model of the individual corresponding to the global optimal position vector that meets the iteration stopping condition as the classification model.

[0013] In one possible design, if the absolute value of the first search coefficient is greater than or equal to a first threshold, the method further includes: For any individual at the t-th iteration, randomly select the position vector of an individual from all individuals at the t-th iteration as the random position vector; A global position update factor is generated, and based on the global position update factor and the random position vector, a global position search is performed on any individual to obtain the updated individual, and after all individuals at the t-th iteration are polled, each updated individual is obtained.

[0014] In one possible design, the nonlinear iteration factor at the tth iteration is generated, including: The nonlinear iteration factor at the tth iteration is generated using the following formula (3); (3) In the above formula (3), represents the nonlinear iteration factor at the tth iteration, is the maximum number of iterations, Indicates a positive number; Correspondingly, based on the first random number, the first search coefficient and the second search coefficient at the t-th iteration, a local position search is performed on each individual at the t-th iteration to obtain each updated individual, which includes: For any individual at the tth iteration, the following formula (4) is used to perform local position search; (4) In the above formula (4), represents the position vector of any individual at the tth iteration, represents the updated position vector of any individual, denote the first search coefficient and the second search coefficient respectively, represents the global optimal position vector, represents the first random number, represents the second random number, represents the spiral enclosure coefficient, Indicates the second threshold.

[0015] In a second aspect, a personalized recommendation device for motorcycle insurance products is provided, comprising: An acquisition unit, wherein the user acquires user information of a target user and vehicle information of a motorcycle corresponding to the target user, wherein the vehicle information includes the age of the vehicle, the vehicle model, and the historical accident record of the vehicle, and the user information includes driving behavior data and personnel data; A feature extraction unit, used to generate insurance feature data of a target user based on the user information and the vehicle information; A classification unit, used for inputting the insurance feature data into a classification model to obtain a user category of the target user, wherein the classification model is trained based on a parameter optimization algorithm and takes sample insurance feature data of a number of sample users as input and the user category of each sample user as output; An insurance product matching unit, configured to match a motorcycle insurance product corresponding to the user category from an insurance product library according to the user category of the target user, so as to serve as a recommended insurance product; The recommendation unit is used to send the recommended insurance product to the client corresponding to the target user to complete the personalized recommendation of the motorcycle insurance product for the target user.

[0016] According to a third aspect, another device for personalized recommendation of motorcycle insurance products is provided. Taking the device as an electronic device as an example, the device includes a memory, a processor and a transceiver which are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the method for personalized recommendation of motorcycle insurance products as described in the first aspect or any possible design of the first aspect.

[0017] In a fourth aspect, a storage medium is provided, on which instructions are stored, and when the instructions are executed on a computer, the method for personalized recommendation of motorcycle insurance products as in the first aspect or any possible design of the first aspect is executed.

[0018] In a fifth aspect, a computer program product comprising instructions is provided, which, when executed on a computer, causes the computer to execute the method for personalized recommendation of motorcycle insurance products as described in the first aspect or any possible design of the first aspect.

[0019] Beneficial effects: (1) The present invention pre-trains a neural network model based on information related to the user's motorcycle insurance business, such as vehicle age, vehicle model, historical vehicle accident record, driving behavior data and personnel data, to obtain a classification model, and thereby classifies insurance purchasing users, thereby obtaining user categories of insurance purchasing users; finally, the insurance products corresponding to the user categories of the insurance purchasing users can be matched according to the user categories of the insurance purchasing users; thus, the present invention integrates multi-dimensional data such as vehicle age, model, historical accident record, driver behavior data, etc., and combines machine learning technology to match insurance products. In this way, compared with traditional technologies, not only the efficiency is improved, but also the personalized insurance needs of users can be met. Therefore, the present invention is very suitable for large-scale application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1A schematic diagram of the steps of the method for personalized recommendation of motorcycle insurance products provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a personalized recommendation device for motorcycle insurance products provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0022] It should be understood that although the terms first, second, etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another unit. For example, a first unit can be referred to as a second unit, and similarly, a second unit can be referred to as a first unit without departing from the scope of the exemplary embodiments of the present invention.

[0023] It should be understood that the term "and / or" that may appear in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" that may appear in this article describes another type of association object relationship, indicating that two relationships may exist. For example, A / and B can represent two situations: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0024] Example: See also Figure 1As shown, the personalized recommendation method for motorcycle insurance products provided in this embodiment integrates multi-dimensional data such as vehicle age, model, historical accident record, driver behavior data, etc., and combines machine learning technology to train a user classification model; finally, the classification model can be used to obtain the user category of the user, and insurance products are matched based on the user category. In this way, compared with traditional manual recommendations, not only the efficiency is improved, but also the personalized insurance needs of users can be met. Therefore, this method is very suitable for large-scale application and promotion; among them, for example, this method can be but not limited to running on the insurance server side. It can be understood that the aforementioned execution subject does not constitute a limitation on the embodiments of the present application. Accordingly, the operation steps of this method can be but not limited to the following steps S1 to S5.

[0025] S1. Obtain the user information of the target user and the vehicle information of the motorcycle corresponding to the target user, wherein the vehicle information includes the age of the vehicle, the vehicle model and the historical accident record of the vehicle, and the user information includes driving behavior data and personnel data; in specific applications, the historical accident record of the vehicle can be obtained by capturing the historical insurance record of the motorcycle corresponding to the target user, which can include but is not limited to the number of historical accidents, the amount of compensation, the type of accident, etc.; at the same time, for example, the driving behavior data can be obtained through but is not limited to the on-board computer of the motorcycle, and can include data such as the number of emergency brakes, the number of emergency accelerations, and the average speed; and the personnel information can include but is not limited to data such as age, gender, driving experience, and violation records; in this way, by obtaining the aforementioned user information and vehicle information related to the insurance business, feature extraction can be performed to screen out the features most relevant to the insurance business, and the process is shown in the following step S2.

[0026] S2. Generate insurance characteristic data of the target user based on the user information and the vehicle information; in specific applications, during training, this embodiment will filter out several actual insurance characteristic indicators related to the insurance business from the sample user information and sample vehicle information of each sample user. Therefore, in actual use, it is only necessary to filter out the data corresponding to the actual insurance characteristic indicators from the user information and vehicle information of the target user, and then use the data corresponding to the actual insurance characteristic indicators to form the insurance characteristic data of the target user; for example, assuming that the actual insurance characteristic indicators are; number of accidents, amount of compensation, type of accident, number of sudden accelerations, number of sudden brakes, average speed, violation record, driving age, etc., then, filter out the data corresponding to the aforementioned actual insurance characteristic indicators from the aforementioned user information and vehicle insurance information, and then, the filtered data can be used to form the insurance characteristic data of the target user; of course, the detailed process of determining the actual insurance characteristic indicators from the user information and vehicle information will be described in detail in the model training process of the following step S3.

[0027] After obtaining the insurance feature data of the target user, it is input into the pre-trained classification model to obtain the user category of the target user (such as whether it is a customer of the first category, the second category or the third category, and different categories correspond to different risk levels); then, insurance products can be matched according to the user category; wherein, the user category generation process can be but is not limited to as shown in the following step S3.

[0028] S3. Input the insurance feature data into the classification model to obtain the user category of the target user, wherein the classification model is based on a parameter optimization algorithm, and is trained with the sample insurance feature data of several sample users as input and the user category of each sample user as output; in actual application, this embodiment uses the sample insurance feature data of several sample users and the label data of each sample user (the label data is the real user category of each sample user) to train the neural network model; in this way, after training, the classification model can be used to perform user classification.

[0029] Optionally, the following discloses one of the training methods of the aforementioned classification model, as shown in the following steps S31 to S35.

[0030] S31. Obtain sample user information of several sample users and sample vehicle information of motorcycles corresponding to each sample user; in this embodiment, the sample user information and sample vehicle information of any sample user are the same as the user information and vehicle information of the target user in the aforementioned step S1, and the data types contained therein are not repeated one by one.

[0031] After the original data is obtained based on step S31, the insurance initial characteristic index and the characteristic data corresponding to each index can be determined, and the process is shown in the following step S32.

[0032] S32. Based on the sample user information and sample vehicle information of each sample user, determine multiple initial insurance characteristic indicators of each sample user and characteristic data corresponding to each initial insurance characteristic indicator; in this embodiment, each type of data in the sample user information and the sample vehicle information is used as an initial insurance characteristic indicator, such as the number of historical accidents, the amount of compensation, the type of accident, the number of sudden brakes, the number of sudden accelerations, the average speed, the age, the gender, the driving experience, the violation record, etc., are all used as an initial insurance characteristic indicator, and the data value corresponding to each of the aforementioned data is used as the characteristic data corresponding to the indicator; in this way, after obtaining multiple initial insurance characteristic indicators of each sample user and the characteristic data corresponding to each indicator, feature screening can be performed to screen out the indicators most relevant to the motorcycle insurance business; wherein, the feature screening process is shown in the following step S33.

[0033] S33. Based on the feature data of each initial insurance feature indicator, each initial insurance feature indicator is screened for feature importance, so as to screen out multiple actual insurance feature indicators from a number of initial insurance feature indicators; in specific applications, this embodiment adopts an entropy-based feature screening method to perform feature screening; specifically, the entropy value between each initial insurance feature indicator is calculated by calculating the similarity between each initial insurance feature indicator, and then, the entropy value is used to perform feature importance screening to obtain important features; wherein, its specific implementation process can be but is not limited to the following steps S33a to S33h.

[0034] S33a. Perform feature mapping processing on the feature data corresponding to each initial insurance feature indicator to obtain mapped feature data corresponding to each initial insurance feature indicator, wherein the dimension of the mapped feature data corresponding to any initial insurance feature indicator is higher than the feature data corresponding to any initial insurance feature indicator; in this embodiment, it is equivalent to mapping the features in the original data set to a higher dimensional space through mathematical transformation, so as to increase the nonlinear expression ability of the data, so that the model can capture more complex relationships, and at the same time, it can also make the data that was originally inseparable in the low-dimensional space become separable in the high-dimensional space, so as to facilitate subsequent model classification.

[0035] Optionally, examples may include but are not limited to feature mapping using kernel functions, that is, kernel functions may implicitly map data to a high-dimensional space without explicitly calculating high-dimensional feature vectors, and examples may include but are not limited to using kernel functions such as linear kernels, polynomial kernels, or radial basis functions (RBF); of course, feature mapping is one of the commonly used processing methods before feature screening, and its principle will not be repeated here.

[0036] After completing the feature mapping of each feature data, all the mapped feature data can be used to form a feature space set, and the process is shown in the following step S33b.

[0037] S33b. All the mapping feature data are used to form a feature space set, and the first information entropy of the feature space set is calculated. In this embodiment, the main steps of feature screening are to first calculate the similarity between each mapping feature data in the feature space set and the remaining mapping feature data, and then, through the similarity, to obtain the entropy value between each mapping feature data and the remaining mapping feature data, and then, based on the entropy value between each mapping feature data, to calculate the first information entropy of the feature space set; then, remove a mapping feature data from the feature space set, and then recalculate the second information entropy of the feature space set; finally, compare the second information entropy with the first information entropy to determine whether the removed mapping feature data is important data.

[0038] Specifically, the principle of feature screening based on the entropy principle is: introduce entropy based on the similarity between features to characterize the uncertainty of data distribution in the feature space; among them, different features contribute differently to the information entropy of the feature space, and important features that help to present the clustering structure should have the effect of reducing entropy, while unimportant or noise features will confuse the data distribution and lead to an increase in the overall information entropy; therefore, by calculating the contribution of each feature to the overall information entropy, important features can be screened out and the ranking of important features can be obtained.

[0039] In specific implementation, the first information entropy can be calculated by, for example but not limited to, using the following steps S33b1 to S33b4.

[0040] S33b1. For the i-th mapping feature data in the feature space set, calculate the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set, wherein the target feature data set contains all mapping feature data in the feature space set except the i-th mapping feature data; in specific applications, for the m-th target feature data in the target feature data set, the following formula (1) may be used, for example but not limited to, to calculate the similarity between the i-th mapping feature data and the m-th target feature data.

[0041] (1) In the above formula (1), represents the similarity between the i-th mapping feature data and the m-th target feature data, Represents the k-th dimension data in the i-th mapping feature data, Represents the data of the kth dimension in the target feature data, represents the standard deviation of the target sequence, A control factor represented as a constant, wherein the target sequence contains the k-th dimension data of each mapped feature data in the feature space set, and is the total dimension of the i-th mapping feature data; where m=1,2,3,...,M, and M is the total number of target feature data.

[0042] In this way, through the above formula (1), the similarity between the i-th mapping feature data and its corresponding target feature data can be calculated, and then, based on this, the entropy value between the i-th mapping feature data and each target feature data can be calculated. The calculation process is shown in the following step S33b2.

[0043] S33b2. According to the similarity between the i-th mapping feature data and each target feature data, the entropy value between the i-th mapping feature data and each target feature data is calculated to obtain multiple entropy values ​​corresponding to the i-th mapping feature data; in this embodiment, the entropy value between the i-th mapping feature data and the m-th target feature data can be calculated by using the following formula (2) for example but not limited thereto.

[0044] (2) In the above formula (2), Represents the entropy value between the i-th mapping feature data and the m-th target feature data.

[0045] In this way, through the above formula (2), the entropy value of the i-th mapping feature data and each target feature data can be calculated; finally, according to the above principle, the entropy value between the next mapping feature data and the remaining mapping feature data in the feature space set can be calculated, until all the mapping feature data in the feature space set are polled, and then multiple entropy values ​​corresponding to each mapping feature data can be obtained; wherein, the cyclic calculation process is shown in the following step S33b3.

[0046] S33b3. Increment i by 1, and recalculate the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set until i equals n, and obtain multiple entropy values ​​corresponding to each mapping feature data.

[0047] After calculating multiple entropy values ​​corresponding to each mapping feature data based on the aforementioned step S33b3, the first information entropy of the feature space set can be calculated based on this, and the process is shown in the following step S33b4.

[0048] S33b4. Sum multiple entropy values ​​corresponding to each mapped feature data, and use the ratio between the summation result and the target parameter as the first information entropy of the feature space set, wherein the target parameter is the square of the total number of mapped feature data.

[0049] In this way, through the aforementioned steps S33b1 to S33b4, the first information entropy of the original feature space set can be calculated; then, the i-th mapping feature data can be deleted from the original feature space set, and then the information entropy of the feature space set can be calculated again; finally, by comparing the first information entropy and the second information entropy, it can be determined whether the i-th mapping feature data is important data, wherein the aforementioned process is shown in the following steps S33c to S33f.

[0050] S33c. Deleting the i-th mapping feature data from the feature space set to obtain an updated feature space set; after deleting the i-th mapping feature data from the feature space set, the information entropy of the feature space set can be recalculated, and the process is shown in step S33d below.

[0051] S33d. Calculate the second information entropy of the updated feature space set; in this embodiment, the calculation process of the second information entropy is the same as the first information entropy, which will not be repeated here.

[0052] After calculating and removing the i-th mapping feature data, the second information entropy of the feature space set can be compared with the first information entropy to determine whether the i-th mapping feature data is important data. The judgment is constructed as shown in the following steps S33e and S33f.

[0053] S33e. Determine whether the difference between the second information entropy and the first information entropy is greater than 0.

[0054] S33f. If so, the initial insurance feature indicator corresponding to the i-th mapping feature data is added to the insurance feature indicator set; in this embodiment, if the second information entropy is greater than the first information entropy, it means that the initial insurance feature indicator corresponding to the i-th mapping feature data is an important feature and can be added to the insurance feature indicator set; otherwise, it is an invalid feature.

[0055] After completing the feature importance judgment of the indicator corresponding to the i-th mapping feature data, the judgment of the next mapping feature data can be carried out until all the feature mapping data in the feature space set are polled, and the final insurance feature indicator set can be obtained; wherein, the cyclic judgment process is shown in the following step S33g.

[0056] S33g. Increment i by 1 and delete the i-th mapped feature data from the feature space set again until i equals n, to obtain the final insurance feature index set, wherein the initial value of i is 1 and n is the total number of mapped feature data.

[0057] In this way, through the aforementioned steps S33c to S33g, important features can be screened out to form a final insurance feature indicator set, and then, the final insurance feature indicator set can be used to determine multiple actual insurance feature indicators, and the process is shown in the following step S33h.

[0058] S33h. Using the final insurance feature indicator set, multiple actual insurance feature indicators are determined. In this embodiment, for example, each insurance feature indicator in the final insurance feature indicator set is sorted in descending order according to the difference between the corresponding second information entropy and the first information entropy, so that the insurance feature indicators in the first several positions in the sorting are used as actual insurance feature indicators. Of course, all indicators in the final insurance feature indicator set can also be used as actual insurance feature indicators. Either of the two methods can be selected, and no specific limitation is made here.

[0059] After completing the feature screening through the above steps, the feature data corresponding to each actual insurance feature indicator can be used to form the sample insurance feature data of each sample user. For example, if multiple actual insurance feature indicators are; number of accidents, compensation amount, accident type, number of sudden accelerations, number of sudden brakes, average speed, violation record, driving experience, etc., then the data corresponding to the above-mentioned actual insurance feature indicators are screened out from the above-mentioned sample user information and sample auto insurance information, and then the screened data constitute the sample insurance feature data of the sample user.

[0060] After obtaining the sample insurance feature data of each sample user, a training data set can be formed, and then the neural network model can be trained based on the training data set, wherein the generation process of the training data set is shown in the following step S34.

[0061] S34. Generate sample insurance feature data of each sample user using feature data corresponding to multiple actual insurance feature indicators of each sample user, and use the sample insurance feature data of each sample user to form a training data set; in this embodiment, each sample user is also provided with label data to represent the real user category of each sample user; then, the neural network model can be trained based on the parameter algorithm and the aforementioned training data set to obtain a classification model; wherein, the training process can be but is not limited to as shown in the following step S35.

[0062] S35. Based on the parameter optimization algorithm, the sample insurance feature data of each sample user in the training data set is used as input, and the user category of each sample user is used as output to train the neural network model, so as to obtain the classification model after the training is completed; in this embodiment, the neural network model can be, but is not limited to, a BP neural network.

[0063] Optionally, the following discloses a method of training a BP neural network model based on a parameter optimization algorithm, and the process is shown in the following steps S35a to S35k.

[0064] S35a. Initialize the model parameters of the neural network model to obtain several groups of model parameters. In this embodiment, the model parameters may be, but are not limited to, the connection weights between each neuron in the BP neural network and the neurons in its adjacent layers, the bias term corresponding to each neuron, the number of network layers, the number of neurons in each layer in the hidden layer, etc. In this way, by initializing the model parameters multiple times to obtain model parameters of different combinations, several groups of model parameters can be obtained.

[0065] After obtaining several sets of model parameters, an initial population can be generated based on them, so that parameter optimization and model training can be performed based on them later. The process is shown in the following step S35b.

[0066] S35b. Generate an initial population using several groups of model parameters, wherein the initial population includes several individuals, and the position vector of each individual corresponds to a group of model parameters; in the present embodiment, each group of model parameters is equivalent to the position vector of each individual, that is, the position vector of an individual is actually a group of model parameters; in this way, after the initial population is generated, the position of each individual in the initial population can be searched, and in the search process, the model is trained by the training data set to obtain the fitness of each individual, so as to determine the optimal position according to the fitness, and then obtain the classification model.

[0067] The location search process of each individual may be, but is not limited to, as shown in the following steps S35c to S35k.

[0068] S35c. Initialize the number of iterations t, and obtain the individual population at the t-th iteration, wherein the initial value of t is 1, and when t is 1, the individual population at the t-th iteration is the initial population.

[0069] After obtaining the individual population at the t-th iteration, the position vector of each individual in the individual population at the t-th iteration can be used to generate the neural network model corresponding to each individual at the t-th iteration, that is, different neural network models are constructed using model parameters corresponding to different position vectors; wherein, the generation process of the neural network models of different individuals can be but is not limited to as shown in the following step S35d.

[0070] S35d. Generate a neural network model corresponding to each individual at the t-th iteration based on the position vector of each individual in the individual population at the t-th iteration; in this embodiment, after constructing the BP neural network model corresponding to each individual using the position vector of each individual, the training data set can be used to train each BP neural network model, thereby obtaining the fitness of each individual at the t-th iteration, wherein the fitness determination process is shown in the following step S35e.

[0071] S35e. The neural network model corresponding to each individual at the t-th iteration is trained using the training data set, and the fitness of each individual at the t-th iteration is calculated based on the loss function of each neural network model, wherein the smaller the loss function value of the neural network model of any individual at the t-th iteration, the greater the fitness value of any individual; in this embodiment, the loss function of the neural network model of any individual (i.e., the BP neural network model) is taken as the cross-entropy loss function; and the fitness is the inverse of the cross-entropy loss function, i.e., the smaller the cross-entropy loss function, the greater the fitness.

[0072] Based on this, after using the training data set to train the neural network model of each individual at the tth iteration, and calculating the fitness of each individual at the tth iteration based on the corresponding loss function, the global optimal position vector can be updated, and the process is shown in the following step S35f.

[0073] S35f. Based on the fitness of each individual at the t-th iteration, the global optimal position vector is determined; in the present embodiment, the maximum fitness is first selected from the fitness of each individual at the t-th iteration; then, it is determined whether the maximum fitness of this iteration is greater than the maximum fitness of the previous iteration; if so, the position vector of the individual corresponding to the maximum fitness of this iteration is used as the global optimal position vector; otherwise, the global optimal position vector is not updated, that is, the position vector of the individual corresponding to the maximum fitness of the previous iteration is used as the global optimal position vector.

[0074] After the global optimal position vector is obtained, the determination of stopping the iteration can be made, and the process is shown in the following step S35g.

[0075] S35g. Determine whether the iteration stop condition is met; in a specific application, the iteration stop condition may be, but is not limited to, the number of iterations reaching the maximum number of iterations, or the fitness of the individual corresponding to the global optimal position vector is greater than or equal to the fitness threshold; wherein, when the aforementioned iteration stop condition is not met, it is necessary to search the position of each individual, and the process is shown in the following steps S35h~S35j.

[0076] S35h. If not, a nonlinear iteration factor is generated for the t-th iteration, and the first search coefficient and the second search coefficient are determined based on the nonlinear iteration factor; in specific implementation, the value of the first search coefficient of the control algorithm for global position search or local position search depends on the iteration factor, wherein the traditional way of changing the iteration factor is not sufficient to adapt to the complex nonlinear search process, resulting in a limited position search range and easy to fall into the local optimum; therefore, this embodiment improves the traditional iteration factor so that its change method can meet the complex nonlinear search process.

[0077] Among them, the following discloses one generation process of the nonlinear iteration factor: Optionally, for example, but not limited to, the following formula (3) may be used to generate a nonlinear iteration factor at the t-th iteration.

[0078] (3) In the above formula (3), represents the nonlinear iteration factor at the tth iteration, is the maximum number of iterations, In this embodiment, for example The value is 8.

[0079] Thus, from the above formula (3), it can be seen that the improved iteration factor shows the characteristics of first slowly decreasing and then rapidly increasing. Therefore, the optimization process can enter the local position search earlier. As the number of iterations increases, its value increases, which can enable the individual to enter the global position search, thereby expanding the search range and avoiding the problem of falling into the local optimum too early.

[0080] After obtaining the nonlinear iteration factor at the t-th iteration, the first search coefficient and the second search coefficient can be calculated based on this; optionally, a third random number can be generated first, and then the nonlinear iteration factor at the t-th iteration is used to multiply the third random number and the constant 2 in sequence (that is, first multiply the third random number, and then multiply the constant 2) to obtain the product result; finally, the nonlinear iteration factor at the t-th iteration is subtracted from the product result, at this time, the first search coefficient at the t-th iteration can be obtained; at the same time, the nonlinear iteration factor at the t-th iteration is multiplied by the constant 2 to obtain the aforementioned second search coefficient.

[0081] After obtaining the first search coefficient and the second search coefficient, it is possible to determine whether to enter a local position search or a global position search based on the value of the first search coefficient; wherein the determination process is shown in the following step S35i.

[0082] S35i. Determine whether the absolute value of the first search coefficient is less than a first threshold value. In a specific application, for example, when the first threshold value is 1, that is, when the absolute value of the first search coefficient is less than 1, the local position search process is entered, and the process is shown in the following step S35j.

[0083] S35j. If so, generate a first random number at the t-th iteration, and based on the first random number at the t-th iteration, the first search coefficient and the second search coefficient, perform a local position search on each individual at the t-th iteration to obtain each updated individual, so as to form a new individual population using each updated individual, wherein the value interval of the first random number is [0,1]. In specific applications, take any individual at the t-th iteration as an example to illustrate the local position search process. Optionally, the following formula (4) may be used as an example but is not limited to perform the local position search.

[0084] (4) In the above formula (4), represents the position vector of any individual at the tth iteration, represents the updated position vector of any individual, denote the first search coefficient and the second search coefficient respectively, represents the global optimal position vector, represents the first random number, represents the second random number, represents the spiral enclosing coefficient, represents the second threshold (value is 0.5); in this embodiment, the value interval of the second random number is also [0,1]; wherein, the spiral enclosure coefficient adopts a decreasing function, that is, , that is, the spiral enclosing coefficient is different in each iteration.

[0085] Thus, based on the above formula (4), when the first random number is less than the second threshold, the first formula in formula (4) is used to search for the position, and when the first random number is greater than or equal to the second threshold, the second formula in formula (4) is used to search for the position; based on this, after each individual in the tth iteration uses the above formula (4) to perform a local position search, the updated position vector can be obtained, thereby completing the position update.

[0086] Further, if the absolute value of the first search coefficient is greater than or equal to the first threshold, a global position search method needs to be used to perform position update, and the process is as follows: Step 1: For any individual at the t-th iteration, randomly select the position vector of an individual from all individuals at the t-th iteration as the random position vector.

[0087] Step 2: Generate a global position update factor, and based on the global position update factor and the random position vector, perform a global position search on any individual to obtain the updated individual, and after polling all individuals at the tth iteration, obtain each updated individual.

[0088] In specific applications, this embodiment discloses one method of generating a global location update factor: Optionally, for example, but not limited to, first generating a first random variable and a second random variable, wherein the first random variable obeys a normal distribution and the second random variable obeys a Cauchy distribution; then, based on the first random variable and the second random variable, generating the global position update factor; in this embodiment, Python software can be used to generate the aforementioned first random variable and second random variable.

[0089] In a specific implementation, the following formula (5) may be used, for example but not limited to, to calculate the global position update factor at the t-th iteration.

[0090] (5) In the above formula (5), represents the global position update factor, represents the first random variable, represents the second random variable, represents the step index (value is (0,2)), Represents the step size parameter.

[0091] After calculating the global position update factor at the t-th iteration based on the above formula (5), the global position search of each individual at the t-th iteration can be performed in combination with the above random position vector. The process is as follows: The following formula (6) is used to perform a global position search on the position of any individual to obtain an updated position vector of any individual, so as to obtain an updated position vector of any individual.

[0092] (6) In the above formula (6), represents the updated position vector of any individual, represents the position vector of any individual at the tth iteration, represents the scale factor, Represents the random position vector; in this embodiment, the scale factor is taken as 0.01.

[0093] Therefore, after obtaining each updated individual through the aforementioned local position search or global position search, each updated individual can be used to form a new individual population; then, t can be incremented by 1, and the individual population at the tth iteration can be updated to the aforementioned new individual population, and then, the neural network model of each individual at the tth iteration can be regenerated based on the position vector of each individual in the individual population at the tth iteration, and retrained; finally, the fitness of each individual is obtained so as to re-judge the iteration stop condition; in this way, the aforementioned process is continuously repeated to complete the continuous updating of the position vector of each individual, and when the iteration is to meet the iteration stop condition, the neural network model of the individual corresponding to the global optimal position vector when the iteration stop condition is met can be used as the classification model; wherein, the cyclic iteration process is shown in the following step S35k.

[0094] S35k. Add 1 to t and update the individual population at the t-th iteration to the new individual population, and regenerate the neural network model corresponding to each individual at the t-th iteration based on the position vector of each individual in the individual population at the t-th iteration, until the iteration stopping condition is met, and use the neural network model of the individual corresponding to the global optimal position vector that meets the iteration stopping condition as the classification model.

[0095] Therefore, through the aforementioned steps S31 to S35 and their sub-steps, the parameter optimization algorithm can be used, and based on the training data set, the parameter optimization and training of the BP neural network can be completed, thereby obtaining a classification model; wherein, the BP neural network is used to output the probability that the target user is of different user categories, and this embodiment uses the example of dividing the user categories into a first category, a second category and a third category; further, the first category is used to characterize the target user as a low-risk user, the second category is used to characterize the target user as a medium-risk user, and the third category is used to characterize the target user as a high-risk user.

[0096] In this way, after obtaining the user category of the target user based on the classification model, insurance matching can be performed based on this, and the process is shown in the following step S4.

[0097] S4. According to the user category of the target user, a motorcycle insurance product corresponding to the user category is matched from the insurance product library as a recommended insurance product. In this embodiment, the insurance product library stores insurance products corresponding to different user categories, such as the first category is associated with a plurality of motorcycle insurance products whose premiums are lower than a first premium threshold and whose number of insurance items is higher than the first insurance item threshold; the third category is associated with a plurality of motorcycle insurance products whose premiums are higher than a second premium threshold and whose number of insurance items is lower than the second insurance item threshold; and the second category is associated with a plurality of motorcycle products whose premiums are between the first premium threshold and the second premium threshold and whose insurance items are between the second insurance item number threshold and the first insurance item number threshold.

[0098] In this way, motorcycle insurance products corresponding to the user category can be matched from the insurance product library according to the user category, and thus used as recommended insurance products; for example, if the user category of the target user is the first category (i.e., a low-risk user), then each motorcycle insurance product corresponding to the first category will be used as a recommended product and sent to the client of the target user.

[0099] The process of sending the recommended products is shown in the following step S5.

[0100] S5. Sending the recommended insurance product to the client corresponding to the target user to complete the personalized recommendation of motorcycle insurance products for the target user.

[0101] Thus, through the personalized recommendation method for motorcycle insurance products described in detail in the aforementioned steps S1 to S5, the present invention pre-trains a neural network model based on information related to the user's motorcycle insurance business, such as vehicle age, vehicle model, historical vehicle accident record, driving behavior data and personnel data, to obtain a classification model, and thereby classifies insurance purchasing users, thereby obtaining user categories of insurance purchasing users; finally, insurance products corresponding to the user categories of insurance purchasing users can be matched according to their user categories; thus, the present invention integrates multi-dimensional data such as vehicle age, model, historical accident record, driver behavior data, and combines machine learning technology to match insurance products. In this way, compared with traditional technologies, not only the efficiency is improved, but also the personalized insurance needs of users can be met. Therefore, the present invention is very suitable for large-scale application and promotion.

[0102] like Figure 2 As shown, the second aspect of this embodiment provides a hardware device for implementing the personalized recommendation method for motorcycle insurance products described in the first aspect of the embodiment, including: The acquisition unit acquires user information of a target user and vehicle information of a motorcycle corresponding to the target user, wherein the vehicle information includes the age of the vehicle, the vehicle model, and the historical accident record of the vehicle, and the user information includes driving behavior data and personnel data.

[0103] The feature extraction unit is used to generate insurance feature data of a target user based on the user information and the vehicle information.

[0104] A classification unit is used to input the insurance feature data into a classification model to obtain the user category of the target user, wherein the classification model is trained based on a parameter optimization algorithm, with sample insurance feature data of several sample users as input and the user category of each sample user as output.

[0105] The insurance product matching unit is used to match the motorcycle insurance product corresponding to the user category from the insurance product library according to the user category of the target user, so as to serve as the recommended insurance product.

[0106] The recommendation unit is used to send the recommended insurance product to the client corresponding to the target user to complete the personalized recommendation of the motorcycle insurance product for the target user.

[0107] The working process, working details and technical effects of the device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0108] like Figure 3 As shown, the third aspect of this embodiment provides another personalized recommendation device for motorcycle insurance products. Taking the device as an electronic device as an example, it includes: a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the personalized recommendation method for motorcycle insurance products as described in the first aspect of the embodiment.

[0109] For example, the memory may include, but is not limited to, random access memory (RAM), read only memory (ROM), flash memory, first input first output (FIFO) and / or first in last out (FILO), etc. Specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). At the same time, the processor may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0110] In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. For example, the processor may not be limited to a microprocessor of the STM32F105 series, a reduced instruction set computer (RISC) microprocessor, an X86 architecture processor, or a processor with an integrated embedded neural network processing unit (NPU); the transceiver may be, but is not limited to, a wireless fidelity (WIFI) wireless transceiver, a Bluetooth wireless transceiver, a General Packet Radio Service (GPRS) wireless transceiver, a ZigBee protocol (a low-power LAN protocol based on the IEEE802.15.4 standard, ZigBee) wireless transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. In addition, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0111] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0112] The fourth aspect of this embodiment provides a storage medium that stores instructions for the method for personalized recommendation of motorcycle insurance products as described in the first aspect of the embodiment, that is, the storage medium stores instructions, and when the instructions are run on a computer, the method for personalized recommendation of motorcycle insurance products as described in the first aspect of the embodiment is executed.

[0113] The storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, a CD, a hard disk, a flash memory, a USB flash drive and / or a memory stick, and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0114] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0115] The fifth aspect of this embodiment provides a computer program product containing instructions, which, when executed on a computer, causes the computer to execute the method for personalized recommendation of motorcycle insurance products as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0116] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. 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 personalized recommendation method for motorcycle insurance products, characterized in that: include: Obtaining user information of a target user and vehicle information of a motorcycle corresponding to the target user, wherein the vehicle information includes the age of the vehicle, the vehicle model, and the historical accident record of the vehicle, and the user information includes driving behavior data and personnel data; generating insurance characteristic data of the target user based on the user information and the vehicle information; Inputting the insurance feature data into a classification model to obtain a user category of the target user, wherein the classification model is trained based on a parameter optimization algorithm and takes sample insurance feature data of several sample users as input and the user category of each sample user as output; According to the user category of the target user, a motorcycle insurance product corresponding to the user category is matched from an insurance product library as a recommended insurance product; The recommended insurance product is sent to the client corresponding to the target user to complete the personalized recommendation of motorcycle insurance products for the target user.

2. The method according to claim 1, characterized in that The classification model is obtained by training in the following manner: Obtaining sample user information of several sample users and sample vehicle information of motorcycles corresponding to each sample user; Determining, based on the sample user information and the sample vehicle information of each sample user, a plurality of initial insurance characteristic indicators of each sample user and characteristic data corresponding to each initial insurance characteristic indicator; Based on the characteristic data of each initial insurance characteristic indicator, each initial insurance characteristic indicator is screened for characteristic importance, so as to screen out a plurality of actual insurance characteristic indicators from a plurality of initial insurance characteristic indicators; Using characteristic data corresponding to a plurality of actual insurance characteristic indicators of each sample user, sample insurance characteristic data of each sample user is generated, and the sample insurance characteristic data of each sample user is used to form a training data set; Based on a parameter optimization algorithm, the sample insurance feature data of each sample user in the training data set is used as input, and the user category of each sample user is used as output to train the neural network model, so as to obtain the classification model after the training is completed.

3. The method according to claim 2, characterized in that Based on the characteristic data of each initial insurance characteristic indicator, each initial insurance characteristic indicator is screened for characteristic importance, so as to screen out a plurality of actual insurance characteristic indicators from a plurality of initial insurance characteristic indicators, including: Performing feature mapping processing on the feature data corresponding to each initial insurance feature indicator to obtain mapped feature data corresponding to each initial insurance feature indicator, wherein the dimension of the mapped feature data corresponding to any initial insurance feature indicator is higher than the feature data corresponding to any initial insurance feature indicator; Using all the mapped feature data, a feature space set is formed, and a first information entropy of the feature space set is calculated; Deleting the i-th mapping feature data from the feature space set to obtain an updated feature space set; Calculating a second information entropy of the updated feature space set; Determine whether the difference between the second information entropy and the first information entropy is greater than 0; If yes, then the initial insurance characteristic index corresponding to the i-th mapped characteristic data is added to the insurance characteristic index set; Add 1 to i and delete the i-th mapped feature data from the feature space set again until i is equal to n, to obtain the final insurance feature index set, wherein the initial value of i is 1 and n is the total number of mapped feature data; Using the final insurance characteristic indicator set, a plurality of actual insurance characteristic indicators are determined.

4. The method according to claim 3, characterized in that Calculate the first information entropy of the feature space set, including: For the i-th mapping feature data in the feature space set, calculating the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set, wherein the target feature data set contains all the mapping feature data in the feature space set except the i-th mapping feature data; Calculating the entropy value between the i-th mapping feature data and each target feature data according to the similarity between the i-th mapping feature data and each target feature data to obtain a plurality of entropy values ​​corresponding to the i-th mapping feature data; Add 1 to i, and recalculate the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set until i equals n, and obtain multiple entropy values ​​corresponding to each mapping feature data; A plurality of entropy values ​​corresponding to each mapped feature data are summed, and a ratio between the summation result and a target parameter is used as a first information entropy of the feature space set, wherein the target parameter is the square of the total number of mapped feature data.

5. The method according to claim 4, characterized in that Calculating the similarity between the i-th mapping feature data and each target feature data in the corresponding target feature data set includes: For the mth target feature data in the target feature data set, the similarity between the i-th mapping feature data and the m-th target feature data is calculated using the following formula (1); （1） In the above formula (1), represents the similarity between the i-th mapping feature data and the m-th target feature data, Represents the k-th dimension data in the i-th mapping feature data, represents the data of the kth dimension in the target feature data, represents the standard deviation of the target sequence, A control factor represented as a constant, wherein the target sequence contains the k-th dimension data of each mapped feature data in the feature space set, and is the total dimension of the i-th mapping feature data; Correspondingly, according to the similarity between the i-th mapping feature data and each target feature data, calculating the entropy value between the i-th mapping feature data and each target feature data includes: The entropy value between the i-th mapping feature data and the m-th target feature data is calculated using the following formula (2); （2） In the above formula (2), Represents the entropy value between the i-th mapping feature data and the m-th target feature data.

6. The method according to claim 2, characterized in that Based on the parameter optimization algorithm, the neural network model is trained with the sample insurance feature data of each sample user in the training data set as input and the user category of each sample user as output, including: Initializing the model parameters of the neural network model to obtain several groups of model parameters; Using several sets of model parameters, an initial population is generated, wherein the initial population includes several individuals, and the position vector of each individual corresponds to a set of model parameters; Initialize the number of iterations t, and obtain the individual population at the t-th iteration, wherein the initial value of t is 1, and when t is 1, the individual population at the t-th iteration is the initial population; Generate a neural network model corresponding to each individual in the individual population at the t-th iteration according to the position vector of each individual in the individual population at the t-th iteration; Using the training data set to train the neural network model corresponding to each individual at the t-th iteration, and based on the loss function of each neural network model, calculating the fitness of each individual at the t-th iteration, wherein the smaller the loss function value of the neural network model of any individual at the t-th iteration, the greater the fitness value of the any individual; Based on the fitness of each individual at the tth iteration, the global optimal position vector is determined; Determine whether the iteration stop condition is met; If not, a nonlinear iteration factor at the t-th iteration is generated, and a first search coefficient and a second search coefficient at the t-th iteration are determined according to the nonlinear iteration factor; Determining whether the absolute value of the first search coefficient is less than a first threshold; If yes, generate the first random number at the t-th iteration, and perform a local position search on each individual at the t-th iteration based on the first random number at the t-th iteration, the first search coefficient, and the second search coefficient to obtain each updated individual, so as to form a new individual population using each updated individual, wherein the value interval of the first random number is [0,1]; Increment t by 1, and update the individual population at the t-th iteration to the new individual population, and regenerate the neural network model corresponding to each individual at the t-th iteration based on the position vector of each individual in the individual population at the t-th iteration, until the iteration stopping condition is met, and use the neural network model of the individual corresponding to the global optimal position vector that meets the iteration stopping condition as the classification model.

7. The method according to claim 6, characterized in that If the absolute value of the first search coefficient is greater than or equal to a first threshold, the method further includes: For any individual at the t-th iteration, randomly select the position vector of an individual from all individuals at the t-th iteration as the random position vector; A global position update factor is generated, and based on the global position update factor and the random position vector, a global position search is performed on any individual to obtain the updated individual, and after all individuals at the t-th iteration are polled, each updated individual is obtained.

8. The method according to claim 6, characterized in that Generate the nonlinear iteration factor for the tth iteration, including: The nonlinear iteration factor at the tth iteration is generated using the following formula (3); （3） In the above formula (3), represents the nonlinear iteration factor at the tth iteration, is the maximum number of iterations, Indicates a positive number; Correspondingly, based on the first random number, the first search coefficient and the second search coefficient at the t-th iteration, a local position search is performed on each individual at the t-th iteration to obtain each updated individual, which includes: For any individual at the tth iteration, the following formula (4) is used to perform local position search; （4） In the above formula (4), represents the position vector of any individual at the tth iteration, represents the updated position vector of any individual, denote the first search coefficient and the second search coefficient respectively, represents the global optimal position vector, represents the first random number, represents the second random number, represents the spiral enclosure coefficient, Indicates the second threshold.

9. A personalized recommendation device for motorcycle insurance products, characterized in that: include: An acquisition unit, wherein the user acquires user information of a target user and vehicle information of a motorcycle corresponding to the target user, wherein the vehicle information includes the age of the vehicle, the vehicle model, and the historical accident record of the vehicle, and the user information includes driving behavior data and personnel data; A feature extraction unit, used to generate insurance feature data of a target user based on the user information and the vehicle information; A classification unit, used for inputting the insurance feature data into a classification model to obtain a user category of the target user, wherein the classification model is trained based on a parameter optimization algorithm and takes sample insurance feature data of a number of sample users as input and the user category of each sample user as output; An insurance product matching unit, configured to match a motorcycle insurance product corresponding to the user category from an insurance product library according to the user category of the target user, so as to serve as a recommended insurance product; The recommendation unit is used to send the recommended insurance product to the client corresponding to the target user to complete the personalized recommendation of the motorcycle insurance product for the target user.

10. A computer program product comprising instructions, characterized in that When the instructions are executed on a computer, the computer is caused to execute the method for personalized recommendation of motorcycle insurance products as claimed in any one of claims 1 to 8.

Images