Product recommendation method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of finance and data processing, and discloses a product recommendation method and device, computer equipment and a storage medium, and the method comprises the steps: determining a channel and a user identifier in response to a vehicle insurance signal of a user, and obtaining vehicle-mounted non-vehicle feature data associated with the user identifier; determining first significant feature data according to the vehicle-mounted non-vehicle feature data, and inputting the first significant feature data into a purchase intention model corresponding to the channel to obtain a purchase intention probability; when the purchase intention probability is lower than a preset threshold value, recommending a first type of non-vehicle insurance products to the user; when the purchase intention probability is not lower than a preset threshold value, determining second significant feature data according to the vehicle-mounted non-vehicle feature data, and inputting second significant features into a purchase gear model to obtain purchase gear preference; and determining a second type of non-vehicle insurance products according to the purchase gear preference and the seat team driving factors, and recommending the second type of non-vehicle insurance products to the user. Recommendation strategies are implemented according to intentions and grades, so that the conversion rate of non-vehicle-mounted products is increased.

Description

Technical Field

[0001] The present invention relates to the field of finance and data processing technology, and in particular to a product recommendation method, device, computer equipment and storage medium. Background Art

[0002] When users purchase auto insurance, insurance companies generally recommend related non-auto insurance products. Cross-recommendation of non-auto insurance products by using the auto insurance user base can effectively expand non-auto insurance business and increase overall insurance business volume. By bundling non-auto insurance products with auto insurance, the stickiness between insurance companies and users can be improved.

[0003] However, there are many types of non-auto insurance products, including personal accident insurance, property insurance, health insurance, travel insurance, etc. Moreover, the prices, coverage, and terms of non-auto insurance products vary. Insurance companies often find it difficult to match users' real needs when recommending them to users, resulting in a relatively low conversion rate for non-auto products, and the potential of non-auto insurance products has not been fully tapped.

[0004] In addition, there are many data labels related to users, and the number of combinations of these data label dimensions will grow exponentially, forming a "dimensionality disaster". However, many combinations of data label dimensions are extremely underutilized in practical applications, making traditional cross-label-based analysis difficult to perform.

[0005] Therefore, there is an urgent need to propose a product recommendation method that can accurately locate user needs, accurately analyze user preferences, and then accurately recommend non-vehicle products, in order to improve the conversion rate of non-vehicle products. Summary of the invention

[0006] The present invention provides a product recommendation method, device, computer equipment and storage medium, which identify the user's purchase intention probability by building a purchase intention model, and identify the user's purchase gear preference by building a purchase gear model, so as to implement recommendation strategies based on intention and gear, in order to improve the conversion rate of non-vehicle products accompanying the vehicle.

[0007] In a first aspect, a product recommendation method is provided, comprising:

[0008] In response to a vehicle insurance subscription signal from a user, determining a channel of the vehicle insurance subscription signal and a user identifier, and acquiring vehicle-related non-vehicle feature data associated with the user identifier;

[0009] Determine first significant feature data according to the accompanying non-vehicle feature data, input the first significant feature data into a purchase intention model corresponding to the channel, and obtain a purchase intention probability;

[0010] When the purchase intention probability is lower than the preset threshold, the first category of non-auto insurance products is recommended to the user;

[0011] When the purchase intention probability is not lower than a preset threshold, second significant feature data is determined according to the accompanying non-vehicle feature data, and the second significant feature is input into the purchase gear model to obtain the purchase gear preference;

[0012] The second category of non-auto insurance products is determined based on the purchase gear preference and seat team driving factors, and the second category of non-auto insurance products is recommended to the user.

[0013] In a second aspect, a product recommendation device is provided, comprising:

[0014] A data acquisition unit, for responding to a user's auto insurance subscription signal, determining a channel of the auto insurance subscription signal and a user identifier, and acquiring vehicle-related non-vehicle feature data associated with the user identifier;

[0015] A purchase intention prediction unit, used to determine first significant feature data according to the accompanying non-vehicle feature data, input the first significant feature data into a purchase intention model corresponding to the channel, and obtain a purchase intention probability;

[0016] A first recommendation unit, configured to recommend a first type of non-auto insurance product to the user when the purchase intention probability is lower than a preset threshold;

[0017] A purchase gear prediction unit, for determining second significant feature data according to the accompanying vehicle non-vehicle feature data when the purchase intention probability is not lower than a preset threshold, inputting the second significant feature into a purchase gear model, and obtaining a purchase gear preference;

[0018] The second recommendation unit is used to determine the second category of non-auto insurance products according to the purchase gear preference and the seat team driving factors, and recommend the second category of non-auto insurance products to the user.

[0019] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned product recommendation method when executing the computer program.

[0020] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned product recommendation method are implemented.

[0021] In the scheme implemented by the above-mentioned product recommendation method, device, computer equipment and storage medium, in response to the user's auto insurance purchase signal, the channel of the auto insurance purchase signal and the user identification are determined, and the vehicle-related non-vehicle feature data associated with the user identification are obtained; first significant feature data are determined based on the vehicle-related non-vehicle feature data, and the first significant feature data are input into the purchase intention model corresponding to the channel to obtain the purchase intention probability; when the purchase intention probability is lower than the preset threshold, the first category of non-auto insurance products is recommended to the user; when the purchase intention probability is not lower than the preset threshold, second significant feature data are determined based on the vehicle-related non-vehicle feature data, and the second significant feature is input into the purchase gear model to obtain the purchase gear preference; the second category of non-auto insurance products is determined based on the purchase gear preference and the seat team driving factor, and the second category of non-auto insurance products is recommended to the user.

[0022] In the present invention, a purchase intention model is constructed to identify the probability of user purchase intention, and a purchase gear model is constructed to identify the user's purchase gear preference, so as to implement a recommendation strategy based on intention and gear, in order to improve the conversion rate of non-car products accompanying the vehicle. The method proposed in the present invention is a comprehensive recommendation strategy that combines user intention, user gear, and product orientation, which can help improve the conversion rate of non-car products accompanying the vehicle to a greater extent. The present invention abandons the traditional cross-dimensional recommendation method and proposes a more comprehensive recommendation strategy, which provides richer recommendation content for users and improves user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0024] Figure 1 is a schematic diagram of an application environment of a product recommendation method in one embodiment of the present invention;

[0025] Figure 2 is a flowchart of a product recommendation method in one embodiment of the present invention;

[0026] Figure 3 1 is a flow chart of a product recommendation method in an embodiment of the present invention;

[0027] Figure 4 is a structural schematic diagram of a product recommendation device in one embodiment of the present invention;

[0028] Figure 5 It is a schematic diagram of the structure of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The idea of ​​the present invention is that when a user purchases auto insurance, the insurance company will generally recommend related non-auto insurance products. If, after purchasing auto insurance, the user purchases corresponding non-auto insurance products based on the recommended related non-auto insurance products, the insurance company has achieved the conversion of non-auto products. In order to improve the conversion rate of non-auto products, insurance companies hope to recommend related non-auto insurance products to users in a more targeted manner. The present invention identifies the user's purchase intention probability by constructing a purchase intention model, and identifies the user's purchase gear preference by constructing a purchase gear model, thereby implementing recommendation strategies based on intention and gear, in order to improve the conversion rate of non-auto products.

[0031] The product recommendation method provided by the embodiment of the present invention can be applied to Figure 1 In the application environment shown, the server is connected to the client for communication.

[0032] The client is used to instruct the user. The client may include, but is not limited to, personal computers, laptops, smart phones, portable wearable devices, and other devices that can realize input and display. The client can purchase auto insurance based on interaction with the server, and send an auto insurance purchase signal to the server after the auto insurance purchase is successful. The auto insurance purchase signal can indicate to the server that the user's auto insurance purchase is successful, and can start recommending non-auto insurance products to the user.

[0033] The server side responds to the user's auto insurance subscription signal, determines the channel of the auto insurance subscription signal and the user identifier, and obtains the vehicle-related non-vehicle feature data associated with the user identifier; determines first significant feature data based on the vehicle-related non-vehicle feature data, inputs the first significant feature data into a purchase intention model corresponding to the channel, and obtains a purchase intention probability; when the purchase intention probability is lower than a preset threshold, recommends a first category of non-auto insurance products to the user; when the purchase intention probability is not lower than the preset threshold, determines second significant feature data based on the vehicle-related non-vehicle feature data, inputs the second significant feature into a purchase gear model, and obtains a purchase gear preference; determines a second category of non-auto insurance products based on the purchase gear preference and the agent team driving factor, and recommends the second category of non-auto insurance products to the user.

[0034] After the server sends the first or second non-auto insurance products to the client, the user can choose whether to purchase the non-auto insurance products on the client. Since the first or second non-auto insurance products are non-auto insurance products that are more suitable for users determined through analysis, users are more likely to choose non-auto insurance products for insurance, and insurance companies hope to obtain a higher conversion rate of non-auto products based on this.

[0035] It is particularly emphasized that the present invention is intended to recommend more suitable non-auto insurance products to users, and the choice of whether to purchase non-auto insurance products still lies with the users themselves.

[0036] The present invention is described in detail below through specific embodiments.

[0037] See also Figure 2 As shown, Figure 2 A flowchart of a product recommendation method provided by an embodiment of the present invention includes the following steps:

[0038] Step S210: In response to the user's car insurance purchase signal, determine the channel of the car insurance purchase signal and the user identification, and obtain the non-car feature data associated with the user identification.

[0039] The insurance company interacts with the user's client through the server to enable the user to complete the car insurance. After the client sends the car insurance signal to the server, the server determines its channel and user ID based on the car insurance signal.

[0040] Channels are used to indicate the path that users take to purchase auto insurance. Users from different channels may have different user behavior characteristics, and subsequent targeted recommendations need to take into account the differences in the groups of users from different channels.

[0041] The user ID is used to indicate the identity of the user who purchases the auto insurance. Through the user ID, the server can retrieve the non-car feature data of the user from the database as the basic data for subsequent analysis of targeted non-car insurance products.

[0042] Step S220: determining first significant feature data according to the accompanying non-vehicle feature data, inputting the first significant feature data into a purchase intention model corresponding to the channel, and obtaining a purchase intention probability.

[0043] The server analyzes the user's purchase intention probability and uses the purchase intention model corresponding to the channel. The purchase intention model is a pre-trained classification model corresponding to the channel. The input of the pre-trained purchase intention model is the first significant feature data, and the output of the pre-trained purchase intention model is the probability of the user purchasing non-car products.

[0044] The amount of vehicle-related non-vehicle feature data is large and the dimensions are rich. After the server obtains the vehicle-related non-vehicle feature data, it is necessary to extract the first significant feature data of the vehicle-related non-vehicle feature data in order to use the purchase intention model to predict the purchase intention probability.

[0045] Step S230: When the purchase intention probability is lower than a preset threshold, recommend the first category of non-auto insurance products to the user.

[0046] If the server determines that the user's purchase intention is low based on the prediction result of the purchase intention model, the first category of non-auto insurance products can be recommended to the user. The first category of non-auto insurance products may include, but are not limited to: products with a high historical conversion probability, low premium products, etc.

[0047] Since users have a low purchasing intention, recommending the first category of non-auto insurance products to users may stimulate users' desire to learn about non-auto insurance products, thereby making it possible for users to purchase non-auto insurance products.

[0048] Step S240, when the purchase intention probability is not lower than the preset threshold, second significant feature data is determined according to the accompanying non-vehicle feature data, and the second significant feature data is input into the purchase gear model to obtain the purchase gear preference.

[0049] If the server determines that the user has a high purchase intention based on the prediction result of the purchase intention model, the server analyzes the user's purchase gear preference and uses the purchase gear model for analysis. The purchase gear model is a pre-trained classification model. The input of the pre-trained purchase gear model is the second significant feature data, and the output of the pre-trained purchase gear model is the gear of the user to purchase non-car products.

[0050] The amount of vehicle-related and non-vehicle feature data is large and the dimensions are rich. After the server obtains the vehicle-related and non-vehicle feature data, it is necessary to extract the second significant feature data from the vehicle-related and non-vehicle feature data to predict the purchase gear preference using the purchase gear model.

[0051] Step S250, determining a second type of non-auto insurance product according to the purchase gear preference and the seat team driving factor, and recommending the second type of non-auto insurance product to the user.

[0052] If the server determines the user's purchase gear preference based on the prediction results of the purchase gear model, the second type of non-auto insurance products can be recommended to the user. The second type of non-auto insurance products can be determined based on the prediction results of the purchase gear model and the seat team driving factors.

[0053] The prediction result of the purchase gear model indicates the user's purchase gear preference. For example, the preference can be a premium range, and the purchase gear model can predict the premium range of the user's non-auto insurance products.

[0054] The factors driving the agent team may include, but are not limited to, non-auto insurance products promoted by the product department and / or non-auto insurance products with policy preferences.

[0055] Since users have a high purchasing intention, recommending the second type of non-auto insurance products to users can more efficiently recommend the main and preferred non-auto insurance products. Users' purchase of appropriate non-auto insurance products can also increase the profits of insurance companies.

[0056] pass Figure 2 The method shown in the invention responds to the user's auto insurance subscription signal, determines the channel of the auto insurance subscription signal and the user identification, obtains the vehicle-related non-vehicle feature data associated with the user identification; determines the first significant feature data based on the vehicle-related non-vehicle feature data, inputs the first significant feature data into the purchase intention model corresponding to the channel, and obtains the purchase intention probability; when the purchase intention probability is lower than a preset threshold, recommends a first category of non-auto insurance products to the user; when the purchase intention probability is not lower than the preset threshold, determines the second significant feature data based on the vehicle-related non-vehicle feature data, inputs the second significant feature into the purchase gear model, and obtains the purchase gear preference; determines the second category of non-auto insurance products based on the purchase gear preference and the agent team driving factor, and recommends the second category of non-auto insurance products to the user.

[0057] In the present invention, a purchase intention model is constructed to identify the probability of user purchase intention, and a purchase gear model is constructed to identify the user's purchase gear preference, so as to implement a recommendation strategy based on intention and gear, in order to improve the conversion rate of non-car products accompanying the vehicle. The method proposed in the present invention is a comprehensive recommendation strategy that combines user intention, user gear, and product orientation, which can help improve the conversion rate of non-car products accompanying the vehicle to a greater extent. The present invention abandons the traditional cross-dimensional recommendation method and proposes a more comprehensive recommendation strategy, which provides richer recommendation content for users and improves user satisfaction.

[0058] In some optional embodiments, S210 responds to the user's auto insurance subscription signal, determines the channel and user identification of the auto insurance subscription signal, and obtains vehicle-related non-vehicle feature data associated with the user identification, including: receiving the auto insurance subscription signal, identifying the auto insurance subscription signal to determine the channel and user identification; querying in the database according to the user identification to obtain user basic feature data, user auto insurance feature data, and user non-auto insurance feature data associated with the user identification.

[0059] After receiving the car insurance subscription signal, the server identifies the car insurance subscription signal to determine its channel and user identification.

[0060] The server can determine the channel based on the path identification of the received car insurance application signal, and determine the user identification based on the information carried by the car insurance application signal.

[0061] For example: the server determines the channel as "Mini Program" and the user ID as "ID123456" based on the car insurance purchase signal.

[0062] The server searches the database according to the user ID, thereby retrieving the user's basic characteristic data, the user's auto insurance characteristic data, and the user's non-auto insurance characteristic data associated with the user ID.

[0063] The database can store a large number of user's vehicle and non-vehicle feature data. When the server determines the user ID, it can match the user ID with a large number of users stored in the database and retrieve the relevant user basic feature data, user vehicle insurance feature data and user non-vehicle insurance feature data.

[0064] For example: user basic characteristic data may include but is not limited to: age, gender, occupation, income level, family situation, marital status, price sensitivity, etc. User auto insurance characteristic data may include but is not limited to: insurance data (insurance channel, insurance institution, etc.), vehicle data (vehicle age, new car purchase price, etc.), insurance data (signing time, renewal period, auto insurance premium, insurance combination, third-party limit, number of accidents, etc.). Non-auto insurance characteristic data may include but is not limited to: last year's non-auto premium, last year's non-auto insurance amount, non-auto retention period, number of non-auto claims in the past two years, number of non-auto insurance years in the past five years, etc.

[0065] In some optional embodiments, the purchase intention model is pre-trained by the following method: obtaining historical vehicle-related and non-vehicle feature data and historical results of whether non-vehicles are purchased with the vehicle that are associated with historical users of multiple channels; defining the first target variable as whether non-vehicles are purchased with the vehicle, preprocessing the historical vehicle-related and non-vehicle feature data, performing numerical conversion on the preprocessed historical vehicle-related and non-vehicle feature data, and screening the first historical significant feature data in the historical vehicle-related and non-vehicle feature data that has a significant impact on the first target variable; constructing a first initial LightGBM model, and setting the parameters of the first initial LightGBM model, wherein the first initial LightGBM model adopts a cross entropy loss function; using the first historical significant feature data of each channel and the corresponding historical results of whether non-vehicles are purchased with the vehicle as training samples, using grid search to tune the parameters of the first initial LightGBM model, and training the first initial LightGBM model separately to obtain multiple purchase intention models corresponding to each channel.

[0066] The present invention trains corresponding purchase intention models for different channels. That is, the server pre-trains multiple purchase intention models corresponding to different channels. Therefore, during data processing, the historical vehicle-related and non-vehicle feature data associated with historical users of different channels and the historical results of whether the non-vehicle was purchased with the vehicle are obtained respectively. During model training, multiple purchase intention models are pre-trained using training samples corresponding to different channels.

[0067] Pre-training of the purchase intention model can include the following steps:

[0068] First, data processing.

[0069] The data processing process can further include: data collection, data preprocessing, numerical conversion and feature engineering.

[0070] Data collection: Obtain historical vehicle-related and non-vehicle feature data associated with historical users from multiple channels and historical results of whether non-vehicle purchases were made along with the vehicle.

[0071] Data preprocessing: Define the first target variable as whether non-vehicle items are purchased along with the vehicle, and preprocess the historical feature data of non-vehicle items along with the vehicle.

[0072] Specifically, the historical vehicle-related and non-vehicle feature data are preprocessed, including: performing missing value processing and outlier processing on the historical vehicle-related and non-vehicle feature data.

[0073] Missing values ​​can be handled by filling in the mean, filling in the median, or deleting missing values.

[0074] Outlier processing can be done by truncation, Winsorize, or by deleting outliers.

[0075] Preprocess historical vehicle and non-vehicle feature data to ensure data quality.

[0076] Numerical conversion: Perform numerical conversion on the preprocessed historical vehicle and non-vehicle feature data.

[0077] Specifically, the preprocessed historical vehicle-related and non-vehicle feature data are numerically converted, including: using one-hot encoding to convert the categorical data in the preprocessed historical vehicle-related and non-vehicle feature data into numerical data; using label encoding to convert the sequential data in the preprocessed historical vehicle-related and non-vehicle feature data into numerical data.

[0078] For the preprocessed historical vehicle-related and non-vehicle feature data whose values ​​are categories and there is no order relationship between categories, the unique hot encoding method is used to convert them into numerical data. For the preprocessed historical vehicle-related and non-vehicle feature data whose values ​​are categories and there is an order relationship between categories, the label encoding method is used to convert them into numerical data.

[0079] Feature engineering: Filter the first historical significant feature data that has a significant impact on the first target variable from the historical vehicle-related and non-vehicle feature data.

[0080] Correlation analysis and statistical methods, such as variance analysis and chi-square test, can be used to screen out the first historical significant feature data that has a significant impact on "whether to purchase non-car items along with the car". Significant impact can be a manually set classification standard. Significant impact can mean that a preset number of historical non-car feature data with a high degree of impact on "whether to purchase non-car items along with the car" are used as the first historical significant feature data. Significant impact can also mean that historical non-car feature data with a degree of impact on "whether to purchase non-car items along with the car" that is higher than a preset limit is used as the first historical significant feature data.

[0081] Second, model training.

[0082] The model training process can further include: initializing the decision tree and training the decision tree using training samples.

[0083] Initialize the decision tree: build the first initial LightGBM model and set the parameters of the first initial LightGBM model, where the first initial LightGBM model adopts the cross entropy loss function.

[0084] Use training samples to train the decision tree: take the first historical significant feature data of each channel and the corresponding historical results of whether the non-car was purchased with the car as training samples, use grid search to tune the parameters of the first initial LightGBM model, train the first initial LightGBM model separately, and obtain multiple purchase intention models corresponding to each channel.

[0085] The first historical significant feature data of the first channel and the corresponding historical results of whether the non-car was purchased along with the car are input into the first initial LightGBM model as training samples, and the parameters of the first initial LightGBM model are tuned using grid search to obtain the purchase intention model corresponding to the first channel.

[0086] The first historical significant feature data of the second channel and the corresponding historical results of whether the non-car was purchased along with the car were input into the first initial LightGBM model as training samples, and the parameters of the first initial LightGBM model were tuned using grid search to obtain the purchase intention model corresponding to the second channel.

[0087] And so on, we can obtain multiple purchase intention models corresponding to each channel.

[0088] In some optional embodiments, step S220 determines the first significant feature data based on the vehicle-related non-vehicle feature data, inputs the first significant feature data into the purchase intention model corresponding to the channel, and obtains the purchase intention probability, including: extracting the first significant feature data from the vehicle-related non-vehicle feature data, inputting the first significant feature data into the purchase intention model corresponding to the channel, and determining the purchase intention probability according to the prediction result of the purchase intention model corresponding to the channel.

[0089] After the server obtains the vehicle-related non-vehicle feature data associated with the user identifier, it extracts the first significant feature data according to the input requirements of the purchase intention model.

[0090] The first significant feature data is input into the purchase intention model corresponding to the channel, and the purchase intention model corresponding to the channel calculates the predicted value of the output variable according to the weight of the decision tree. That is, the purchase intention model corresponding to the channel outputs the probability that the user purchases a non-car along with the car.

[0091] In some optional implementations, S230 recommends a first category of non-auto insurance products to the user when the purchase intention probability is lower than a preset threshold, including: when the purchase intention probability is lower than a preset threshold, recommends a non-auto insurance product with a historically high conversion probability to the user.

[0092] For example, the preset threshold can be 0.7. If the prediction result of the purchase intention model is that the probability of the user purchasing a non-car along with the car is 0.5, based on 0.5<0.7, it is determined that the user's purchase intention is low. At this time, non-car insurance products with a high historical conversion probability are recommended to the user.

[0093] In some optional embodiments, the gear purchase model is pre-trained by the following method: obtaining historical vehicle-related non-vehicle feature data and historical vehicle-related non-vehicle gear purchase results associated with historical users; defining a second target variable as the vehicle-related non-vehicle gear purchase, preprocessing the historical vehicle-related non-vehicle feature data, performing numerical conversion on the preprocessed historical vehicle-related non-vehicle feature data, and screening the second historical significant feature data in the historical vehicle-related non-vehicle feature data that has a significant impact on the second target variable; constructing a second initial LightGBM model, setting the parameters of the second initial LightGBM model, wherein the second initial LightGBM model adopts a defined loss function: Loss = CrossEntropyLoss-αR, (1); wherein CrossEntropyLoss represents the cross entropy loss function, α represents the adjustment factor, and R represents the preset expected reward; using the second historical significant feature data and the corresponding historical vehicle-related non-vehicle gear purchase results as training samples, tuning the parameters and adjustment factors of the second initial LightGBM model, training the second initial LightGBM model, and obtaining the gear purchase model.

[0094] Pre-training of the purchase gear model can include the following steps:

[0095] First, data processing.

[0096] The data processing process can further include: data collection, data preprocessing, numerical conversion and feature engineering.

[0097] Data collection: Obtain historical vehicle-related and non-vehicle feature data associated with historical users and historical vehicle-related and non-vehicle gear purchase results.

[0098] Data preprocessing: Define the second target variable as the non-vehicle purchase gear with the vehicle, and preprocess the historical non-vehicle feature data with the vehicle. Taking the premium range as an example, the non-vehicle purchase gear with the vehicle can include: (0-300] yuan, (300-600] yuan, (600-1000] yuan, and more than 1000 yuan.

[0099] Specifically, the historical vehicle-related and non-vehicle feature data are preprocessed, including: performing missing value processing and outlier processing on the historical vehicle-related and non-vehicle feature data.

[0100] Missing values ​​can be handled by filling in the mean, filling in the median, or deleting missing values.

[0101] Outlier processing can be done by truncation, Winsorize, or by deleting outliers.

[0102] Preprocess historical vehicle and non-vehicle feature data to ensure data quality.

[0103] Numerical conversion: Perform numerical conversion on the preprocessed vehicle and non-vehicle feature data.

[0104] Specifically, the preprocessed historical vehicle-related and non-vehicle feature data are numerically converted, including: using one-hot encoding to convert the categorical data in the preprocessed historical vehicle-related and non-vehicle feature data into numerical data; using label encoding to convert the sequential data in the preprocessed historical vehicle-related and non-vehicle feature data into numerical data.

[0105] For the preprocessed historical vehicle-related and non-vehicle feature data whose values ​​are categories and there is no order relationship between categories, the unique hot encoding method is used to convert them into numerical data. For the preprocessed historical vehicle-related and non-vehicle feature data whose values ​​are categories and there is an order relationship between categories, the label encoding method is used to convert them into numerical data.

[0106] Feature engineering: Screen the second historical significant feature data that has a significant impact on the second target variable from the historical vehicle-related and non-vehicle feature data.

[0107] Correlation analysis and statistical methods, such as variance analysis and chi-square test, can be used to screen out the second historical significant feature data that has a significant impact on "non-vehicle gears purchased with the vehicle". Significant impact can be a manually set classification standard. Significant impact can mean a preset number of historical non-vehicle feature data with a high degree of impact on "non-vehicle gears purchased with the vehicle" as the second historical significant feature data. Significant impact can also mean that historical non-vehicle feature data with a degree of impact on "non-vehicle gears purchased with the vehicle" that is higher than a preset limit is used as the second historical significant feature data.

[0108] Second, model training.

[0109] The model training process can further include: initializing the decision tree and training the decision tree using training samples.

[0110] Initialize the decision tree: construct a second initial LightGBM model and set the parameters of the second initial LightGBM model, wherein the second initial LightGBM model adopts the defined loss function: Loss = CrossEntropyLoss - αR, (1), wherein CrossEntropyLoss represents the cross entropy loss function, α represents the adjustment factor, and R represents the preset expected reward.

[0111] Since the purchase gear model predicts the user's preference for different gears, this is a multi-classification problem. Therefore, rewards are introduced on the basis of the cross entropy loss function to improve the prediction accuracy of the purchase gear model. In traditional machine learning models, the goal of the model is usually to optimize the prediction accuracy. Introducing rewards in the loss function can embed business goals directly into the model training process, so that the model can consider business goals at the same time when predicting, which is more in line with actual business needs.

[0112] The loss function is defined as: Loss = CrossEntropyLoss - αR, (1), where CrossEntropyLoss represents the cross entropy loss function, α represents the adjustment factor (used to balance the impact of the original loss and reward), and R represents the preset expected reward.

[0113] Use training samples to train the decision tree: use the second historical significant feature data and the corresponding historical non-vehicle gear purchase results as training samples, tune the parameters and adjustment factors of the second initial LightGBM model, train the second initial LightGBM model, and obtain the purchase gear model.

[0114] The LightFBM model is applied to multi-classification problems, trained using a defined loss function, and parameters and adjustment factors are tuned based on the training results. That is, the original prediction accuracy and expected rewards are optimized at the same time, so that the pre-trained purchase gear model can consider the possibility of users accepting recommendations and subsequent transaction conditions when predicting gear preferences.

[0115] In some optional embodiments, S240 when the purchase intention probability is not lower than a preset threshold, second significant feature data is determined based on the vehicle-related non-vehicle feature data, the second significant feature is input into a purchase gear model, and a purchase gear preference is obtained, including: extracting the second significant feature data from the vehicle-related non-vehicle feature data, inputting the second significant feature data into the purchase gear model, and determining the purchase gear preference according to the prediction result of the purchase gear model.

[0116] For example, the preset threshold can be 0.7. If the prediction result of the purchase intention model is that the probability of the user purchasing a non-car along with the car is 0.8, based on 0.8>0.7, it is determined that the user's purchase intention is high. At this time, the server obtains the non-car along with the car feature data associated with the user identifier, and extracts the second significant feature data according to the input requirements of the purchase gear model.

[0117] The second significant feature data is input into the purchase gear model, and the purchase gear model calculates the predicted value of the output variable according to the weight of the decision tree. That is, the purchase gear model outputs the gear of the non-car purchased by the user along with the car.

[0118] In some optional embodiments, S250 determines a second category of non-auto insurance products based on purchase gear preferences and seat team driving factors, and recommends the second category of non-auto insurance products to users, including: determining gear preference non-auto insurance products based on purchase gear preferences; determining main push non-auto insurance products based on seat team driving factors; determining the second category of non-auto insurance products based on gear preference non-auto insurance products and main push non-auto insurance products; and recommending the second category of non-auto insurance products to users.

[0119] For example, based on the prediction result of the purchase gear model, it is determined that the gear of the non-car purchased by the user along with the car is (300-600] yuan, then based on the purchase gear preference, the gear preference non-car insurance products with a premium range of (300-600] yuan are selected, namely, Product B, Product E, and Product F.

[0120] For example: Based on the driving factors of the agent team, the non-auto insurance products and policy-oriented non-auto insurance products promoted by the product department are determined to be Product A (premium range (0-300] yuan), Product B (premium range (300-600] yuan), Product C (premium range (600-1000] yuan), and Product D (premium range more than 1000 yuan).

[0121] The server further selects product B with a premium range of RMB 300 to 600 and recommends it to the user as the second type of non-auto insurance product.

[0122] The product recommendation method proposed in the present invention has a comprehensive and flexible comprehensive recommendation strategy, showing excellent scalability and application potential. It can be reused for prediction of rights, activities and other scenarios, enriching the recommendation content and meeting the diversified needs of users.

[0123] See also Figure 3 As shown, Figure 3 Another flowchart of the product recommendation method provided by the embodiment of the present invention includes the following steps:

[0124] Step S301, obtaining historical vehicle-related and non-vehicle feature data associated with historical users from multiple channels and historical results of whether non-vehicles were purchased along with the vehicle. Go to step S302.

[0125] Step S302, define the first target variable as whether to purchase non-vehicles along with the vehicle, perform missing value processing and outlier processing on the historical feature data along with the vehicle and the non-vehicles; use one-hot encoding to convert the categorical data in the preprocessed historical feature data along with the vehicle and the non-vehicles; use label encoding to convert the ordinal data in the preprocessed historical feature data along with the vehicle and the non-vehicles to the numerical data, and select the first historical significant feature data in the historical feature data along with the vehicle and the non-vehicles that has a significant impact on the first target variable. Go to step S303.

[0126] Step S303, construct the first initial LightGBM model, set the parameters of the first initial LightGBM model, wherein the first initial LightGBM model adopts the cross entropy loss function; use the first historical significant feature data of each channel and the corresponding historical results of whether the non-car is purchased with the car as training samples, use grid search to tune the parameters of the first initial LightGBM model, train the first initial LightGBM model separately, and obtain multiple purchase intention models corresponding to each channel. Go to step S309.

[0127] Step S304, obtaining the historical vehicle-related non-vehicle feature data and the historical vehicle-related non-vehicle gear position purchase results associated with the historical user. Go to step S305.

[0128] Step S305, define the second target variable as the non-vehicle gear position purchased with the vehicle, perform missing value processing and outlier processing on the historical non-vehicle feature data with the vehicle; use one-hot encoding to convert the categorical data in the preprocessed historical non-vehicle feature data with the vehicle into numerical data; use label encoding to convert the ordinal data in the preprocessed historical non-vehicle feature data with the vehicle into numerical data, and screen the second historical significant feature data in the historical non-vehicle feature data with the vehicle that has a significant impact on the second target variable. Go to step S306.

[0129] Step S306, construct the second initial LightGBM model, set the parameters of the second initial LightGBM model, wherein the second initial LightGBM model adopts the defined loss function: Loss = CrossEntropyLoss-αR, wherein CrossEntropyLoss represents the cross entropy loss function, α represents the adjustment factor, and R represents the preset expected reward; the second historical significant feature data and the corresponding historical vehicle-purchased non-vehicle gear position results are used as training samples, the parameters and adjustment factors of the second initial LightGBM model are tuned, the second initial LightGBM model is trained, and the purchase gear position model is obtained. Go to step S312.

[0130] Step S307, receiving the car insurance purchase signal, identifying the car insurance purchase signal to determine the channel and user identification. Go to step S308.

[0131] Step S308, query the database according to the user identification to obtain the user basic characteristic data, the user auto insurance characteristic data and the user non-auto insurance characteristic data associated with the user identification. Go to step S309.

[0132] Step S309, extracting first significant feature data from the user's basic feature data, the user's auto insurance feature data, and the user's non-auto insurance feature data, inputting the first significant feature data into a purchase intention model corresponding to the channel, and determining the purchase intention probability according to the prediction result of the purchase intention model corresponding to the channel. Go to step S310.

[0133] Step S310, determine whether the purchase intention probability is lower than a preset threshold. If yes, go to step S311; if not, go to step S312.

[0134] Step S311, recommending non-auto insurance products with a historically high conversion probability to the user.

[0135] Step S312, extracting second significant feature data from the user's basic feature data, the user's auto insurance feature data, and the user's non-auto insurance feature data, inputting the second significant feature data into the purchase intention model, and determining the purchase gear preference according to the prediction result of the purchase intention model. Go to step S313.

[0136] Step S313, determine the gear preference non-auto insurance product according to the purchase gear preference, determine the main push non-auto insurance product according to the seat team driving factor, and determine the second type of non-auto insurance product according to the gear preference non-auto insurance product and the main push non-auto insurance product. Go to step S314.

[0137] Step S314, recommending the second type of non-auto insurance products to the user.

[0138] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0139] In one embodiment, a product recommendation device is provided, which corresponds one-to-one to the product recommendation method in the above embodiment. Figure 4 As shown, the product recommendation device 400 includes:

[0140] The data acquisition unit 401 is used to respond to the user's car insurance purchase signal, determine the channel of the car insurance purchase signal and the user identification, and obtain the non-car feature data associated with the user identification;

[0141] A purchase intention prediction unit 402 is used to determine first significant feature data according to the accompanying non-vehicle feature data, input the first significant feature data into a purchase intention model corresponding to the channel, and obtain a purchase intention probability;

[0142] The first recommendation unit 403 is used to recommend a first type of non-auto insurance product to the user when the purchase intention probability is lower than a preset threshold;

[0143] The purchase gear prediction unit 404 is used to determine the second significant feature data according to the accompanying vehicle non-vehicle feature data when the purchase intention probability is not lower than a preset threshold, input the second significant feature into the purchase gear model, and obtain the purchase gear preference;

[0144] The second recommendation unit 405 is used to determine the second type of non-auto insurance products according to the purchase gear preference and the seat team driving factor, and recommend the second type of non-auto insurance products to the user.

[0145] In some optional implementations, the data acquisition unit 401 is specifically used to: receive a motor vehicle insurance policy signal, identify the motor vehicle insurance policy signal to determine the channel and user identification; query in the database according to the user identification to obtain user basic feature data, user motor vehicle insurance feature data, and user non-motor vehicle insurance feature data associated with the user identification.

[0146] In some optional embodiments, the purchase intention prediction unit 402 includes: a purchase intention model training module, which is specifically used to: obtain historical vehicle-related and non-vehicle feature data associated with historical users of multiple channels and historical results of whether non-vehicles are purchased with the vehicle; define the first target variable as whether non-vehicles are purchased with the vehicle, preprocess the historical vehicle-related and non-vehicle feature data, perform numerical conversion on the preprocessed historical vehicle-related and non-vehicle feature data, and screen the first historical significant feature data in the historical vehicle-related and non-vehicle feature data that has a significant impact on the first target variable; construct a first initial LightGBM model, set the parameters of the first initial LightGBM model, wherein the first initial LightGBM model adopts a cross entropy loss function; use the first historical significant feature data of each channel and the corresponding historical results of whether non-vehicles are purchased with the vehicle as training samples, use grid search to tune the parameters of the first initial LightGBM model, train the first initial LightGBM model separately, and obtain multiple purchase intention models corresponding to each channel.

[0147] In some optional embodiments, the purchase intention prediction unit 402 is specifically used to: extract first significant feature data from the vehicle-related non-vehicle feature data, input the first significant feature data into a purchase intention model corresponding to the channel, and determine the purchase intention probability according to the prediction result of the purchase intention model corresponding to the channel.

[0148] In some optional implementations, the first recommendation unit 403 is specifically configured to: when the purchase intention probability is lower than a preset threshold, recommend a non-auto insurance product with a historically high conversion probability to the user.

[0149] In some optional embodiments, the purchase gear prediction unit 404 includes: a purchase gear model training module, which is specifically used to: obtain historical vehicle-related non-vehicle feature data associated with historical users and historical vehicle-related non-vehicle gear purchase results; define the second target variable as the vehicle-related non-vehicle gear purchase, preprocess the historical vehicle-related non-vehicle feature data, perform numerical conversion on the preprocessed historical vehicle-related non-vehicle feature data, and screen the second historical significant feature data in the historical vehicle-related non-vehicle feature data that has a significant impact on the second target variable; construct a second initial LightGBM model, and set the parameters of the second initial LightGBM model, wherein the second initial LightGBM model adopts a defined loss function: Loss = CrossEntropyLoss-αR, wherein CrossEntropyLoss represents the cross entropy loss function, α represents the adjustment factor, and R represents the preset expected reward; use the second historical significant feature data and the corresponding historical vehicle-related non-vehicle gear purchase results as training samples, tune the parameters and adjustment factors of the second initial LightGBM model, train the second initial LightGBM model, and obtain the purchase gear model.

[0150] In some optional implementations, the purchase gear prediction unit 404 is specifically used to extract second significant feature data from the vehicle-related non-vehicle feature data, input the second significant feature data into the purchase gear model, and determine the purchase gear preference according to the prediction result of the purchase gear model.

[0151] In some optional implementations, the purchase intention model training module or the purchase gear model training module is specifically used to: perform missing value processing and outlier processing on historical vehicle-related and non-vehicle feature data; use one-hot encoding to convert the categorical data in the preprocessed historical vehicle-related and non-vehicle feature data into numerical data; use label encoding to convert the ordinal data in the preprocessed historical vehicle-related and non-vehicle feature data into numerical data.

[0152] In some optional implementations, the second recommendation unit 405 is specifically used to: determine the gear preference non-auto insurance products according to the purchase gear preference; determine the main push non-auto insurance products according to the seat team driving factors; determine the second category of non-auto insurance products according to the gear preference non-auto insurance products and the main push non-auto insurance products; and recommend the second category of non-auto insurance products to users.

[0153] For the specific definition of the product recommendation device, please refer to the definition of the product recommendation method above, which will not be repeated here. Each module in the above-mentioned product recommendation device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0154] In one embodiment, a computer device is provided, the internal structure diagram of which can be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a product recommendation method.

[0155] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0156] In response to a vehicle insurance subscription signal from a user, determining a channel of the vehicle insurance subscription signal and a user identifier, and acquiring vehicle-related non-vehicle feature data associated with the user identifier;

[0157] Determine first significant feature data according to the accompanying non-vehicle feature data, input the first significant feature data into a purchase intention model corresponding to the channel, and obtain a purchase intention probability;

[0158] When the purchase intention probability is lower than the preset threshold, the first category of non-auto insurance products is recommended to the user;

[0159] When the purchase intention probability is not lower than a preset threshold, second significant feature data is determined according to the accompanying non-vehicle feature data, and the second significant feature is input into the purchase gear model to obtain the purchase gear preference;

[0160] The second category of non-auto insurance products is determined based on the purchase gear preference and seat team driving factors, and the second category of non-auto insurance products is recommended to the user.

[0161] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant description in the aforementioned method embodiment. In order to avoid repetition, they will not be described one by one here.

[0162] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0163] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0164] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A product recommendation method, characterized in that: include: In response to a vehicle insurance subscription signal from a user, determining a channel and a user identifier of the vehicle insurance subscription signal, and acquiring vehicle-related non-vehicle feature data associated with the user identifier; Determining first significant feature data according to the accompanying non-vehicle feature data, inputting the first significant feature data into a purchase intention model corresponding to the channel to obtain a purchase intention probability; When the purchase intention probability is lower than a preset threshold, recommending a first category of non-auto insurance products to the user; When the purchase intention probability is not lower than the preset threshold, determining second significant feature data according to the accompanying non-vehicle feature data, inputting the second significant feature into a purchase gear model, and obtaining a purchase gear preference; A second category of non-auto insurance products is determined based on the purchase gear preference and the agent team driving factors, and the second category of non-auto insurance products is recommended to the user.

2. The product recommendation method according to claim 1, characterized in that: The step of responding to the vehicle insurance purchase signal of the user, determining the channel and user identification of the vehicle insurance purchase signal, and acquiring vehicle-related non-vehicle feature data associated with the user identification includes: receiving the vehicle insurance application signal, identifying the vehicle insurance application signal to determine the channel and the user identifier; The database is queried according to the user identification to obtain user basic feature data, user auto insurance feature data and user non-auto insurance feature data associated with the user identification.

3. The product recommendation method according to claim 1, characterized in that: The purchase intention model is pre-trained by the following method: Obtaining historical vehicle-related and non-vehicle feature data associated with multiple historical users of the channels and historical results of whether non-vehicle purchases were made along with the vehicle; A first target variable is defined as whether a non-vehicle is purchased along with the vehicle, the historical non-vehicle characteristic data along with the vehicle is preprocessed, the historical non-vehicle characteristic data along with the vehicle is numerically converted, and first historical significant characteristic data having a significant impact on the first target variable is screened from the historical non-vehicle characteristic data along with the vehicle; Construct a first initial LightGBM model and set parameters of the first initial LightGBM model, wherein the first initial LightGBM model adopts a cross entropy loss function; Taking the first historical significant feature data of each of the channels and the corresponding historical results of whether non-vehicle items were purchased along with the vehicle as training samples, grid search is used to tune the parameters of the first initial LightGBM model, and the first initial LightGBM model is trained separately to obtain multiple purchase intention models corresponding to each of the channels.

4. The product recommendation method according to claim 1, characterized in that: When the purchase intention probability is lower than a preset threshold, recommending a first type of non-auto insurance product to the user includes: When the purchase intention probability is lower than the preset threshold, a non-auto insurance product with a historically high conversion probability is recommended to the user.

5. The product recommendation method according to claim 1, characterized in that: The purchase gear model is pre-trained by the following method: Obtain historical vehicle-related and non-vehicle feature data associated with historical users and historical vehicle-related and non-vehicle gear purchase results; The second target variable is defined as the non-vehicle gear purchased along with the vehicle, the historical non-vehicle characteristic data along with the vehicle is preprocessed, the historical non-vehicle characteristic data along with the vehicle is numerically converted, and the second historical significant characteristic data having a significant impact on the second target variable is screened from the historical non-vehicle characteristic data along with the vehicle; Construct a second initial LightGBM model and set the parameters of the second initial LightGBM model, wherein the second initial LightGBM model adopts the defined loss function: Loss=CrossEntropyLoss-αR, (1); Among them, CrossEntropyLoss represents the cross entropy loss function, α represents the adjustment factor, and R represents the preset expected reward; Using the second historical significant feature data and the corresponding historical non-vehicle gear purchase results as training samples, the parameters of the second initial LightGBM model and the adjustment factor are tuned, the second initial LightGBM model is trained, and the purchase gear model is obtained.

6. The product recommendation method according to claim 3 or 5, characterized in that: The preprocessing of the historical vehicle-related non-vehicle feature data and the numerical conversion of the preprocessed historical vehicle-related non-vehicle feature data include: Performing missing value processing and outlier processing on the historical vehicle-related non-vehicle feature data; Using one-hot encoding to convert the categorical data in the preprocessed historical vehicle-related non-vehicle feature data into numerical data; Label encoding is used to convert the preprocessed sequential data in the historical non-vehicle feature data into the numerical data.

7. The product recommendation method according to claim 1, characterized in that: The determining of a second type of non-auto insurance product according to the purchase gear preference and the seat team driving factor, and recommending the second type of non-auto insurance product to the user comprises: Determining a gear preference non-auto insurance product according to the purchase gear preference; Determine the non-auto insurance products that are preferred based on the driving factors of the agent team; Determining the second category of non-auto insurance products according to the gear preference non-auto insurance products and the main push tendency non-auto insurance products; The second category of non-auto insurance products is recommended to the user.

8. A product recommendation device, characterized in that: include: A data acquisition unit, configured to respond to a user's auto insurance subscription signal, determine a channel and a user identifier of the auto insurance subscription signal, and acquire vehicle-related non-vehicle feature data associated with the user identifier; a purchase intention prediction unit, configured to determine first significant feature data according to the accompanying non-vehicle feature data, input the first significant feature data into a purchase intention model corresponding to the channel, and obtain a purchase intention probability; A first recommendation unit, configured to recommend a first type of non-auto insurance product to the user when the purchase intention probability is lower than a preset threshold; a purchase gear prediction unit, configured to determine second significant feature data according to the accompanying non-vehicle feature data when the purchase intention probability is not lower than the preset threshold, input the second significant feature into a purchase gear model, and obtain a purchase gear preference; The second recommendation unit is used to determine a second type of non-auto insurance product according to the purchase gear preference and the seat team driving factor, and recommend the second type of non-auto insurance product to the user.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the product recommendation method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the product recommendation method according to any one of claims 1 to 7 are implemented.

Images