Household service recommendation method, device and equipment based on home and property insurance and storage medium

Abstract

The invention relates to the technical field of artificial intelligence and the field of financial science and technology, and discloses a household service recommendation method based on home and property insurance, which performs data format conversion and standardization processing by collecting real estate information, regional information, customer information, home and property insurance information and service provider information of customers and processing missing values and abnormal values in data. And according to the cleaned data, extracting features influencing housekeeping service demands, and determining key features for housekeeping service recommendation. And establishing a preliminary housekeeping service recommendation model based on the key features, and performing evaluation and optimization by adopting a contrast experiment to generate a personalized housekeeping service recommendation model. And using the personalized housekeeping service recommendation model to provide housekeeping service recommendation meeting the requirements of the customer for the customer. According to the method, the precision of household service recommendation can be improved, the customer experience is enhanced, and meanwhile, the attraction of household and financial insurance products and the customer stickiness are promoted.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a method, device, equipment and storage medium for recommending domestic services based on home property insurance. Background Art

[0002] In recent years, with the continuous development of social economy and the improvement of residents' income level, household property insurance (home property insurance) has gradually become one of the important products in the insurance industry. Home property insurance provides families with property loss protection, covering accidents such as natural disasters, theft, fire, etc. However, with the diversification of customer needs, the competitiveness of traditional home property insurance products in the market has gradually declined. In order to meet this challenge, insurance companies have gradually turned their attention to improving service added value, especially introducing personalized housekeeping services as value-added services of insurance products to enhance customer stickiness and the attractiveness of insurance products.

[0003] In the existing home property insurance service model, insurance companies mostly provide standardized insurance services and cannot make personalized adjustments based on the specific circumstances of customers. Each family's property characteristics, housing conditions, living habits, and safety and climate conditions in the area are very different. Traditional insurance services cannot fully consider these personalized factors, resulting in low customer satisfaction and low renewal rates, and it is difficult for insurance companies to improve customer loyalty through additional services.

[0004] In order to improve customer experience and satisfaction, some insurance companies have begun to explore the integration of housekeeping services as value-added services into home property insurance products. Housekeeping services not only cover daily services such as daily cleaning, home appliance maintenance, and house inspections, but can also help customers reduce the potential risk of property loss due to negligence through regular inspections and maintenance. For example, regular appliance maintenance services can not only provide convenience to customers, but also reduce the risk of accidents such as fires caused by electrical failures, thereby reducing the possibility of insurance claims.

[0005] However, the current way of providing housekeeping services is still mainly based on the traditional model, which cannot achieve truly personalized recommendations. The service content is usually divided according to the customer's premium and insurance type, lacking in-depth analysis of the customer's real needs. Insurance companies need a more intelligent system that can combine the customer's real estate information, the environmental conditions of the area, the composition of family members and living habits and other multi-dimensional factors to tailor housekeeping service plans for customers. Summary of the invention

[0006] The main purpose of the present invention is to provide a method, device, equipment and storage medium for recommending housekeeping services based on home property insurance, aiming to solve the technical problem in the prior art that there is a lack of effective combination between housekeeping services and home property insurance, and it is impossible to provide customers with personalized housekeeping service recommendations based on home property insurance.

[0007] To achieve the above object, the present invention provides a method for recommending housekeeping services based on home property insurance, comprising:

[0008] Collect the client's property information, region information, customer information, home insurance information and service provider information;

[0009] Clean the collected data, process missing values ​​and outliers, and convert and standardize the data format;

[0010] From the processed data, extract the features that affect the demand for domestic services and determine the key features for domestic service recommendations;

[0011] Establishing a preliminary housekeeping service recommendation model based on the key features, evaluating and optimizing the preliminary housekeeping service recommendation model through a control experiment, and generating a personalized housekeeping service recommendation model;

[0012] The personalized housekeeping service recommendation model is used to output personalized housekeeping service recommendations.

[0013] In one embodiment, establishing a preliminary housekeeping service recommendation model based on the key features includes:

[0014] Dividing the key features into a training data set and a validation data set;

[0015] Constructing an initial housekeeping service recommendation model, randomly selecting a preset number of data points from the training data set as initial cluster centers;

[0016] Based on the distance between each data point in the training data set and the initial cluster center, the data point is assigned to the nearest initial cluster center, and the position of the initial cluster center is updated;

[0017] Through multiple iterations, adjusting the parameters of the initial housekeeping service recommendation model, gradually reducing the distance between each data point and the cluster center;

[0018] Use the validation set to evaluate the initial housekeeping service recommendation model and adjust the location of the cluster center;

[0019] When the position change of the cluster center is less than the preset convergence condition, the training is stopped to obtain the preliminary housekeeping service recommendation model.

[0020] In one embodiment, the parameters of the initial housekeeping service recommendation model are adjusted through multiple iterations to gradually reduce the distance between each data point and the cluster center, including:

[0021] Initialize the Q-table's recommended strategy Q value to zero or a random value;

[0022] In each iteration, the initial housekeeping service recommendation model is used to generate recommendation results, feedback data is collected, and a reward value is generated based on the feedback data;

[0023] Update the recommended strategy Q value in the Q-table based on the reward value generated in each iteration;

[0024] Updating the location of the cluster center and optimizing the parameters of the initial housekeeping service recommendation model based on the reward value of each iteration;

[0025] The Q value of the recommendation strategy is updated through multiple iterations, gradually reducing the gap between each data point and the cluster center.

[0026] In one embodiment, a control experiment is used to evaluate and optimize the preliminary housekeeping service recommendation model to generate a personalized housekeeping service recommendation model, including:

[0027] Using the preliminary housekeeping service recommendation model as the experimental group and the preset recommendation strategy as the control group, and tracking core indicators including recommendation accuracy, click-through rate and customer retention rate;

[0028] Collect the recommendation result data of the experimental group and the control group, including the accuracy of the recommendations, the clicks on the recommendation results, and the actual conversion rate after the service selection, and compare and analyze the specific performance of the recommendation result data of the control group and the experimental group on the core indicators;

[0029] Optimizing weights and hyperparameters in the preliminary housekeeping service recommendation model based on the analysis results of the recommendation result data;

[0030] The personalized housekeeping service recommendation model is generated through multiple rounds of control experiments and model optimization.

[0031] In one embodiment, in the processed data, features that affect the demand for housekeeping services are extracted, and key features for housekeeping service recommendations are determined, including:

[0032] Extract categorical features and raw numerical features from processed data, and create derived features based on business knowledge;

[0033] Encoding the extracted categorical features, converting them into numerical features through one-hot encoding or label encoding, and normalizing or standardizing the converted numerical features and the original numerical features;

[0034] Using a chi-square test or an ANOVA test method, selecting features that affect the demand for domestic services from the derived features, the converted numerical features, and the original numerical features;

[0035] The number of features is reduced by recursive feature elimination or random forest feature importance ranking to extract the key features for housekeeping service recommendation.

[0036] In one embodiment, the collected data is cleaned, missing values ​​and outliers are processed, and data format conversion and standardization are performed, including:

[0037] Clean the collected data and fill in missing data by interpolation or K-nearest neighbor method;

[0038] Detect and handle outliers using box plots, Z-scores, or interquartile range methods, and replace outliers with non-outliers or the median;

[0039] Perform consistency checks on data fields and convert data fields in different formats into a unified format;

[0040] Normalize or standardize numerical data so that the data are processed within the same scale range.

[0041] In one embodiment, using the personalized housekeeping service recommendation model to output personalized housekeeping service recommendations includes:

[0042] Obtain a recommended home insurance plan for the customer, and extract insurance type, premium, and housing information from the recommended home insurance plan;

[0043] The extracted insurance type, premium and housing information are used as input and imported into the personalized housekeeping service recommendation model;

[0044] Through the personalized housekeeping service recommendation model, combined with the customer's historical behavior data, personalized housekeeping service recommendations that match customer needs are generated;

[0045] According to the personalized housekeeping service recommendation, the housekeeping service recommendation contents are arranged and displayed in order of priority.

[0046] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a home service recommendation device based on home property insurance, the home service recommendation device based on home property insurance comprising a memory, a processor and a home service recommendation program based on home property insurance stored in the memory and executable on the processor, the home service recommendation program based on home property insurance, when executed by the processor, implements the steps of the home service recommendation method based on home property insurance as described above.

[0047] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer storage medium, on which a home service recommendation program based on home property insurance is stored, and when the home service recommendation program based on home property insurance is executed by a processor, the steps of the home service recommendation method based on home property insurance as described above are implemented.

[0048] Beneficial effects: The present invention provides a method for recommending housekeeping services based on home property insurance. By collecting the customer's real estate information, regional information, customer information, home property insurance information and service provider information, the missing values ​​and outliers in the data are processed, and data format conversion and standardization are performed. According to the cleaned data, the features that affect the demand for housekeeping services are extracted, and the key features for housekeeping service recommendations are determined. A preliminary housekeeping service recommendation model is established based on the key features, and control experiments are used for evaluation and optimization to generate a personalized housekeeping service recommendation model. The personalized housekeeping service recommendation model is used to provide customers with housekeeping service recommendations that meet their needs. The present invention can improve the accuracy of housekeeping service recommendations, enhance customer experience, and at the same time promote the attractiveness and customer stickiness of home property insurance products. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0050] Figure 1 It is a flow chart of an embodiment of a method for recommending housekeeping services based on household property insurance according to the present invention;

[0051] Figure 2 It is a functional module diagram of a preferred embodiment of a housekeeping service recommendation device based on household property insurance of the present invention;

[0052] Figure 3 It is a structural schematic diagram of the equipment hardware operating environment involved in the embodiment scheme of the recommended equipment for housekeeping services based on household property insurance of the present invention. DETAILED DESCRIPTION

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

[0054] In recent years, with the continuous development of social economy and the improvement of residents' income level, household property insurance (home property insurance) has gradually become one of the important products in the insurance industry. Home property insurance provides families with property loss protection, covering accidents such as natural disasters, theft, fire, etc. However, with the diversification of customer needs, the competitiveness of traditional home property insurance products in the market has gradually declined. In order to meet this challenge, insurance companies have gradually turned their attention to improving service added value, especially introducing personalized housekeeping services as value-added services of insurance products to enhance customer stickiness and the attractiveness of insurance products.

[0055] In the existing home property insurance service model, insurance companies mostly provide standardized insurance services and cannot make personalized adjustments based on the specific circumstances of customers. Each family's property characteristics, housing conditions, living habits, and safety and climate conditions in the area are very different. Traditional insurance services cannot fully consider these personalized factors, resulting in low customer satisfaction and low renewal rates, and it is difficult for insurance companies to improve customer loyalty through additional services.

[0056] In order to improve customer experience and satisfaction, some insurance companies have begun to explore the integration of housekeeping services as value-added services into home property insurance products. Housekeeping services not only cover daily services such as daily cleaning, home appliance maintenance, and house inspections, but can also help customers reduce the potential risk of property loss due to negligence through regular inspections and maintenance. For example, regular appliance maintenance services can not only provide convenience to customers, but also reduce the risk of accidents such as fires caused by electrical failures, thereby reducing the possibility of insurance claims.

[0057] However, the current way of providing housekeeping services is still mainly based on the traditional model, which cannot achieve truly personalized recommendations. The service content is usually divided according to the customer's premium and insurance type, lacking in-depth analysis of the customer's real needs. Insurance companies need a more intelligent system that can combine the customer's real estate information, the environmental conditions of the area, the composition of family members and living habits and other multi-dimensional factors to tailor housekeeping service plans for customers.

[0058] Due to the limitations of the interface calling method, the scalability of the system is severely restricted. Every time an upstream business service is added, the system needs to develop and test the interface, which not only prolongs the time of system expansion, but also increases the difficulty of expansion. In addition, when the system needs to adjust the interface calling logic, a lot of code modification and testing work is required, which increases the complexity and risk of system adjustment.

[0059] In summary, the existing interface calling methods have many problems and challenges when processing payment information feedback, especially in terms of complexity, efficiency, performance, stability and scalability. There is an urgent need for a more efficient, flexible and stable solution to improve the overall performance of the system and user experience.

[0060] See also Figure 1 , Figure 1 This is a flowchart of an embodiment of a method for recommending housekeeping services based on home property insurance provided by the present invention. It should be noted that although a logical sequence is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0061] like Figure 1 As shown, the method for recommending housekeeping services based on household property insurance proposed by the present invention comprises the following steps:

[0062] S10, collecting the client's property information, region information, client information, home property insurance information, and service provider information;

[0063] In this embodiment, the real estate information includes the type of house, area, construction age, decoration style, etc. These data can be obtained by filling out a form by the customer, connecting to the real estate database interface, or from the government's public real estate registration system. It is also possible to use an API interface to connect to a third-party real estate data platform, or to parse the form information submitted by the customer and store it in the database. Real estate data can also be obtained from property companies, real estate companies, etc., and using OCR technology to extract relevant data from paper documents is also a feasible way.

[0064] Regional information includes the climate, geographical location, surrounding public facilities, etc. of the customer's location. Detailed regional information can be obtained through address resolution services (such as Google Maps API or Baidu Maps API). Geocoding technology can also be used to convert the customer's address information into longitude and latitude, combined with relevant regional information such as climate data, traffic conditions, crime rates, etc., to generate regional data required for service recommendations. The postal code or detailed address entered by the customer can also be used as the input source of regional information, or data can be dynamically obtained by connecting with third-party meteorological platforms and traffic services.

[0065] Customer information includes age, occupation, income, family structure, etc. This data can be obtained by filling in the form when the customer registers, or by connecting to a third-party authentication platform for the customer's personal information. It can also be obtained by filling in the form or connecting to the interface of the customer data platform, using encryption technology to ensure the secure transmission and storage of customer privacy data. Customer information can also be obtained through authentication by banks, social security and other institutions to ensure the authenticity and accuracy of the data.

[0066] Home property insurance information includes policy type, premium amount, insurance terms, etc. This information can be extracted from the insurance company's internal system or parsed through the insurance policy provided by the customer. It can also be connected to the insurance company's API interface, or the electronic insurance policy submitted by the customer can be used for data parsing; at the same time, natural language processing technology is used to extract key information from the policy text. It is also possible to obtain the customer's home property insurance information through a third-party insurance platform, or extract relevant information from the paper policy through OCR technology.

[0067] Service provider information includes housekeeping company qualifications, service items, customer reviews, service coverage areas, etc. Service provider data can be obtained through the housekeeping service platform, or provided by the insurance company's internal service management system. It is also possible to connect to the data of the housekeeping service platform through the API interface, or directly retrieve relevant data from the insurance company's supplier management system. It is also possible to dynamically screen qualified service providers based on the customer's region, or use recommendation algorithms to automatically recommend suitable service providers based on customer preferences and historical choices.

[0068] By comprehensively collecting and processing customer property information, regional information, customer information, home property insurance information, and service provider information, the system can accurately analyze the actual needs of customers and make intelligent housekeeping service recommendations based on multi-dimensional data. This not only improves the accuracy of recommendations, but also increases customer satisfaction and loyalty through personalized recommendations. It also helps optimize the additional services of home property insurance products and improve the competitiveness of insurance companies.

[0069] S20, clean the collected data, process missing values ​​and outliers, and perform data format conversion and standardization;

[0070] In this embodiment, data cleaning refers to deleting or correcting incomplete, erroneous or invalid data from the collected data to improve data quality. Common cleaning methods include deleting blank data, filling missing values, and correcting erroneous values. Programming languages ​​(such as the Pandas library in Python) can be used to automatically detect and clean invalid data in the data. Missing values ​​can be filled using mean filling, interpolation, or K-nearest neighbor (KNN) to ensure data integrity. Cleaning methods may be different for different types of data. For example, text data may need to be spell-checked and semantically corrected; numerical data requires reasonable value range judgment, and outliers can be replaced with medians or means.

[0071] When there are missing values ​​in the data, the system needs to select a reasonable missing value processing method based on different data types. For numerical data, common methods include mean filling, median filling, or prediction filling based on historical data. You can use machine learning models (such as KNN or linear regression) to predict missing values, or quickly complete the data through simple mean and median filling methods. Text data can use frequency analysis or context prediction methods to handle missing parts. When there are too many missing data, you may choose to directly delete these incomplete records or mark them as "unknown" categories to avoid affecting the overall analysis.

[0072] Outliers are extreme values ​​that deviate from the normal range in a data set, and are usually detected and processed by statistical methods. This includes detecting outliers based on Z-score, interquartile range (IQR) or box plots, and correcting or deleting them as appropriate. You can also use the Z-score method to calculate the standard deviation difference between each data point and the mean, and values ​​that exceed the set range are marked as abnormal. The IQR method detects values ​​with large deviations by calculating the interquartile range of the data distribution. For outliers, you can choose to delete them directly or replace them with reasonable values, such as the median or the average of the previous and next records. Abnormal text or category data can be corrected through word frequency analysis.

[0073] Data format conversion refers to converting data from different sources and formats into a unified format to ensure the consistency of subsequent processing. For example, the date format is unified into the "YYYY-MM-DD" format, or the currency unit is unified into the same currency. Inconsistent formats can be converted to a unified format by customizing conversion rules or using third-party data format conversion tools. Different types of data such as time, currency, and numerical values ​​are standardized using specific conversion methods. The system can check the type and format of the data field and convert the inconsistent parts uniformly to ensure that data from different data sources are comparable in the same dimension.

[0074] Standardization refers to converting data of different scales into the same dimension so that the model can handle it better. For example, compress all numerical data to between 0 and 1, or perform Z-score normalization on the data. Commonly used normalization methods include Min-Max Normalization and Z-score normalization. The former scales the data to a specific range, and the latter scales based on the mean and standard deviation of the data. Choose an appropriate normalization method based on different model requirements. For example, for neural network models, normalization is often used, while for linear models, Z-score normalization is usually used.

[0075] By cleaning the data, processing missing values, outliers, converting data formats and standardizing them, the system ensures the integrity, consistency and normalization of the data. It not only improves the accuracy of data analysis and model training, but also reduces erroneous results and recommendation deviations caused by data quality issues, thereby significantly improving the accuracy of housekeeping service recommendations and customer experience.

[0076] S30, extracting features that affect the demand for domestic services from the processed data, and determining key features for domestic service recommendation;

[0077] In this embodiment, features related to the customer's housekeeping service needs are extracted from the cleaned data, such as house area, family composition, geographical location, customer income level, etc. These features are important factors for predicting customer housekeeping service needs. Data mining methods can be used to analyze categorical features (such as house type, geographical location) and numerical features (such as house area, income level) in the data, and features that are highly relevant to customer housekeeping needs can be screened out through business knowledge and statistical analysis. Derived features (such as the ratio of house area to the number of family members) can also be created through feature engineering to enhance the understanding of customer needs. Feature selection based on business rules can be used to quickly determine which features have a high impact on housekeeping service recommendations.

[0078] Among the extracted features, key features that have a significant impact on housekeeping service recommendations are further screened out. These key features are the most important input variables in the model recommendation process. Statistical analysis methods such as chi-square test, ANOVA test, or recursive feature elimination are used to analyze the correlation between each feature and the demand for housekeeping services, and to screen out features that are significantly correlated with service demand. At the same time, machine learning methods (such as feature importance ranking of random forests) are used to further refine features to ensure the optimality of model inputs. The optimal features are selected by integrating multiple feature selection techniques (such as L1 regularization and random forests). For complex feature selection processes, automated machine learning tools can also be used to automatically optimize feature selection.

[0079] In one specific implementation, additional financial health characteristics of customers are extracted from the processed data, including credit score, debt-to-asset ratio, consumption pattern, etc. These financial characteristics directly affect customers' demand for housekeeping services. For example, financially healthy customers may give priority to high-end housekeeping services. Through cluster analysis, the financial behavior patterns of customer groups are identified, and financial health status, asset categories, etc. are associated with housekeeping service needs, and the core financial characteristics that affect service recommendations are selected.

[0080] By extracting features and screening key features from the cleaned data, the system can identify the core factors that affect customers' demand for housekeeping services. This improves the accuracy of the recommendation model, reduces data redundancy, and reduces the computational complexity of the model. Through efficient feature selection, the system can more accurately provide housekeeping service recommendations that meet customer needs, improving customer experience and satisfaction.

[0081] S40, establishing a preliminary housekeeping service recommendation model based on the key features, evaluating and optimizing the preliminary housekeeping service recommendation model using a control experiment, and generating a personalized housekeeping service recommendation model;

[0082] In this embodiment, a preliminary housekeeping service recommendation model is established based on the extracted key features. The model can be a machine learning model, such as a commonly used classification model based on a decision tree, random forest, or support vector machine (SVM). These models predict the type of housekeeping service that a customer may need through key features (such as the customer's property information, region, customer information, insurance information, etc.). The labeled training data can be modeled by a machine learning algorithm, and different algorithms can be used to select models to build a preliminary recommendation model. The results of feature engineering (such as the customer's family structure, insurance type, etc.) will be used as input to output the corresponding recommended services. Deep learning (such as a multi-layer neural network) can also be used to process more complex nonlinear features, or a rule-based recommendation model can be used to generate service recommendations based on specific business rules and decision trees derived from data analysis.

[0083] A controlled experiment was used to evaluate the effectiveness of the preliminary housekeeping service recommendation model. The recommendation strategy of the control group was based on traditional recommendation methods, usually simple fixed rules or based on the customer's historical preferences. This recommendation method is logically simple and usually provides recommendations to customers based on their previous choices, conventional service items, or market-standard service packages. The recommendation model of the control group does not use complex machine learning algorithms or personalized models, but relies on predefined rule sets. For example, for a certain type of customer, a few types of housekeeping services may be recommended without considering the customer's specific personal characteristics, such as age, property area, insurance information, etc. .

[0084] The experimental group uses a preliminary housekeeping service recommendation model based on key features, and makes recommendations based on the multi-dimensional features of the customer (such as age, number of family members, property information, home insurance information, etc.). This model generates housekeeping service recommendations that best meet the customer's needs based on the multi-dimensional features collected from the customer. The recommendation model of the experimental group uses data mining, machine learning and other technical means to dynamically analyze customer characteristics and generate personalized service recommendations. The recommendation results will be combined with the customer's home insurance characteristics, service preferences and historical choices to optimize service content and timing.

[0085] The control group is based on traditional recommendation algorithms, which may be rule-driven recommendations. For example, based on common customer choices, the system may recommend standard housekeeping services, such as basic cleaning, deep cleaning, etc. This recommendation strategy will not be personalized based on the customer's detailed information. For the control group, a simple recommendation method based on customer historical behavior can be used, such as an association rule algorithm or a recommendation model based on customer segmentation. The data comes from the types of services previously selected by customers and the mainstream service items in the market.

[0086] The experimental group is based on a personalized recommendation model that generates recommendations that meet the personalized needs of customers by analyzing their detailed characteristics (such as age, property type, geographic location, home insurance plan, etc.). The recommendations for different customers will change dynamically based on their specific characteristics. The experimental group model uses machine learning or deep learning models to input the customer's multi-dimensional characteristics, including home insurance information (such as insurance type, premium, coverage), family structure and property information, to generate personalized recommendations. The performance of the experimental group can be tracked in real time through the A / B testing platform, and the model parameters can be optimized based on customer feedback.

[0087] The control experiment evaluates the effectiveness of the personalized recommendation model by comparing the performance of the experimental group with that of the control group. Specific indicators include customer satisfaction, recommendation click-through rate, conversion rate, and service repurchase rate. The recommendation results of the experimental group and the control group will be compared and analyzed through multiple indicators (such as user interaction and actual selected services). The difference in the effects of the two groups can be obtained by monitoring user behavior data, collecting feedback, analyzing satisfaction surveys, etc.

[0088] Based on the evaluation results, the personalized recommendation model of the experimental group can be further optimized. The optimization process includes adjusting the weight of the recommendation model, selecting different features, or adding new data sources to improve the accuracy of the recommendation. Reinforcement learning or hyperparameter tuning techniques can be used to adjust the parameters of the recommendation model in real time based on the feedback data of the experimental group to improve the model's predictive ability and recommendation quality. The continuous conduct of control experiments allows the model to be continuously verified and optimized.

[0089] Through controlled experiments, we can objectively evaluate the advantages of personalized recommendation models over traditional recommendation strategies. The experimental group can significantly improve the accuracy of recommendations and customer satisfaction by combining machine learning algorithms with detailed user information. Compared with the control group, the experimental group can provide housekeeping service recommendations that better meet customer needs, improve conversion rates and service utilization rates, and ultimately increase customer loyalty and the added value of insurance products.

[0090] S50: Outputting personalized housekeeping service recommendations using the personalized housekeeping service recommendation model.

[0091] In this embodiment, the input of the personalized housekeeping service recommendation model includes the customer's property information, regional information, family structure, home property insurance information, etc. These feature data are cleaned and standardized as the input of the model. By extracting the user's feature data and using feature engineering technology, the raw data is converted into an input vector suitable for model processing. The recommendation results are calculated using a machine learning model (such as a decision tree, random forest, or neural network). For different types of customers (such as different family sizes and different insurance amounts), the input features will be adjusted, and the system will adjust the weights of the features in real time according to the specific needs of the customer. For example, for high-net-worth customers, the system will focus on features such as property area and insurance coverage.

[0092] The personalized recommendation model generates personalized housekeeping service recommendations based on the input customer characteristics by learning the customer's historical behavior and current needs. The model can determine the priority of the recommended services through classification or sorting algorithms. Use classification models (such as logistic regression, SVM) or sorting models to generate recommendation results. The model calculates the recommendation score for each housekeeping service based on the input characteristics, and outputs a list of housekeeping services that best meet the customer's needs after sorting. The collaborative filtering algorithm can be used to generate personalized recommendations based on the historical selection behavior of similar customers. The model can also continuously optimize the recommendation plan through reinforcement learning methods, making the recommendation results more dynamic and personalized.

[0093] The generated recommendation plan is presented to customers through the service platform. The recommendation content includes a detailed description of the service, service provider information, price, time, etc. Customers can directly select the recommended service to make a reservation or adjust the recommendation plan. The recommendation results output by the model are presented to customers in a list form through the recommendation system interface. The recommended content is sorted by priority, and customers can select or further adjust it according to their personal needs. The recommended content can be presented to customers through mobile APP, web pages or telephone customer service channels, supporting customers to obtain and select housekeeping services in a variety of ways.

[0094] By using a personalized recommendation model to generate personalized housekeeping service recommendations, the system can provide accurate service recommendations based on the needs and characteristics of different customers. This not only improves customer satisfaction with housekeeping services, but also increases the selection rate and conversion rate of services. The use of a personalized recommendation system reduces the difficulty of customer selection, makes the recommendation plan more in line with the actual needs of customers, and improves customer experience and the added value of insurance products.

[0095] The present invention relates to a method for recommending housekeeping services based on household property insurance. By collecting the client's property information, regional information, client information, household property insurance information and service provider information, the missing values ​​and outliers in the data are processed, and data format conversion and standardization are performed. According to the cleaned data, the features that affect the demand for housekeeping services are extracted, and the key features for housekeeping service recommendations are determined. A preliminary housekeeping service recommendation model is established based on the key features, and a control experiment is used for evaluation and optimization to generate a personalized housekeeping service recommendation model. The personalized housekeeping service recommendation model is used to provide clients with housekeeping service recommendations that meet their needs. The present invention can improve the accuracy of housekeeping service recommendations, enhance customer experience, and at the same time promote the attractiveness and customer stickiness of household property insurance products.

[0096] In one embodiment, in the above S40, establishing a preliminary housekeeping service recommendation model based on the key features includes:

[0097] S401, dividing the key features into a training data set and a verification data set;

[0098] S402, constructing an initial housekeeping service recommendation model, randomly selecting a preset number of data points from the training data set as initial cluster centers;

[0099] S403, based on the distance between each data point in the training data set and the initial cluster center, assign the data point to the nearest initial cluster center and update the position of the initial cluster center;

[0100] S404, adjusting the parameters of the initial housekeeping service recommendation model through multiple iterations to gradually reduce the distance between each data point and the cluster center;

[0101] S405, using the validation set to evaluate the initial housekeeping service recommendation model and adjust the location of the cluster center;

[0102] S406, when the position change of the cluster center is less than the preset convergence condition, stop training and obtain the preliminary housekeeping service recommendation model.

[0103] In this embodiment, a training data set and a validation data set are divided from the extracted key feature data set. The training data is used for model training, and the validation data is used for model evaluation. The common division ratio is 80% training data and 20% validation data, but it can be flexibly adjusted according to the amount of data. Use a programming language (such as the train_test_split function in Python) to randomly divide the data set into a training set and a validation set to ensure that the training set is used for model learning and the validation set is used for model evaluation. The cross-validation method (k-fold cross-validation) can be used to divide the data set into k subsets, and different subsets are used in turn as validation sets to ensure that the model performs consistently under different data.

[0104] By randomly selecting K data points from the training data set as the initial cluster centers for the subsequent clustering process. The initial model can be constructed based on a clustering algorithm (such as K-means). Using a clustering algorithm such as K-means, by selecting random K data points as the initial cluster centers, and using these centers for subsequent data point allocation and updating. You can also use the K-means++ algorithm to optimize the selection of the initial cluster centers, reduce the convergence time of the model, and improve the clustering effect.

[0105] Calculate the distance between each data point and each cluster center, and assign the data point to the closest cluster center. Update the location of the cluster center, and the new cluster center is the average value of the data points in the group to which it belongs. Use Euclidean distance or other distance metrics to calculate the distance between the data point and the cluster center. Each time the cluster center is updated, calculate the mean of all data points in the class as the new cluster center. You can also use other distance metrics (such as Manhattan distance) to calculate the distance between the data point and the cluster center, or use weighted average to update the cluster center.

[0106] Through multiple iterations, the system gradually adjusts the parameters in the clustering model so that each data point gradually approaches the cluster center to which it belongs, until the cluster center changes very little in each iteration. During the iteration process, the distance between the data point and the cluster center is repeatedly calculated, and the cluster center is updated until the convergence condition is reached. The maximum number of iterations and the convergence threshold can be set. Accelerated convergence techniques can also be used, such as using momentum updates of cluster centers, or combining clustering algorithms that accelerate convergence (such as MiniBatchK-means).

[0107] The performance of the model is evaluated through the data of the validation set, including the clustering effect and the accuracy of the recommendation service. If the model performs poorly, the model parameters or the location of the cluster center can be adjusted to further optimize the model. Evaluation indicators (such as silhouette coefficient, SSE, etc.) are used to evaluate the performance of the model, and the results of the validation set are used to adjust the location of the cluster center and optimize the accuracy of the model. The validation set can also be used for fine-grained model adjustments, or the model parameters can be further optimized through automated parameter adjustment tools (such as grid search).

[0108] When the position change of the cluster center in the model is less than the set convergence threshold, the iteration stops, the model training ends, and a stable housekeeping service recommendation model is output. Set the convergence condition, usually when the moving distance of the cluster center is less than a certain threshold, or when the maximum number of iterations is reached. After the training is completed, the final model is output for subsequent recommendations. You can also set an adaptive convergence threshold to dynamically adjust the convergence condition according to the performance of the model, or use an early stopping mechanism to avoid overfitting.

[0109] This embodiment can efficiently establish a housekeeping service recommendation model based on customer characteristics through clustering and iterative optimization, ensuring that the model can dynamically adjust the recommended content according to the needs of different customers. Through verification and optimization, the model improves the accuracy of recommendations and the matching degree of services, thereby improving customer satisfaction and conversion rate.

[0110] In one embodiment, the above S404 includes:

[0111] S4041, initializing the Q-table's recommended strategy Q value to zero or a random value;

[0112] S4042, in each iteration, using the initial housekeeping service recommendation model to generate recommendation results, collecting feedback data, and generating a reward value based on the feedback data;

[0113] S4043, updating the recommended strategy Q value in the Q-table based on the reward value generated in each iteration;

[0114] S4044, updating the location of the cluster center, and optimizing the parameters of the initial housekeeping service recommendation model based on the reward value of each iteration;

[0115] S4045, updates the Q value of the recommendation strategy through multiple iterations, gradually reducing the gap between each data point and the cluster center.

[0116] In this embodiment, in the initial stage of reinforcement learning, the Q-table's recommended strategy Q value is usually initialized to zero or a random value, indicating that the system has no empirical data to guide the expected reward of the state-action pair. The Q-table is initialized to all zeros or a random matrix through a programming language (such as the numpy library in Python). Each Q value represents the expected reward after selecting a recommended strategy in a specific state. The empirical initial value can be used, that is, the initial Q value is set based on the past recommendation history, thereby accelerating the learning process.

[0117] In each iteration, the system generates recommendation results based on the current recommendation strategy and collects customer feedback data on the recommendation results. The feedback data may include customer clicks, selections, or usage. Based on these feedback data, the system generates a reward value for Q-value updates in reinforcement learning. The recommendation results are generated through existing recommendation models (such as recommendation algorithms based on clustering or collaborative filtering). Customer feedback data can be collected by real-time monitoring of user behavior. The generation of reward values ​​can be calculated based on customer behavior data (such as click-through rate, purchase conversion rate). The reward value can change dynamically based on different customer behaviors. For example, clicking on a recommended service will result in a positive reward, while giving up the recommendation will result in a negative reward.

[0118] Each time according to the generated reward value, update the Q value in the Q-table. The update formula is:

[0119] Q(s,a)←Q(s,a)+α[r+γmax(Q(s′,a′))-Q(s,a)].

[0120] Among them, the leftmost Q(s,a) is the updated Q(s,a) value, which represents the new Q value updated after executing action a in state s, that is, the expected reward value after this learning. This value is the new expected reward that the system can obtain by executing action a in the current state s. The Q value represents the reward that the system expects to obtain from the environment after executing a certain action in a certain state. As learning progresses, the Q value will be gradually updated to reflect more accurate reward expectations.

[0121] The Q(s,a) in the middle is the Q value before the update, that is, the current Q value of performing action a in state s. This value is used to calculate the gap between the current state and the ideal reward. The formula gradually adjusts this Q value, making the model's predictions more and more accurate.

[0122] The rightmost Q(s,a) is the current Q value of executing action a in the current state s (i.e., the Q value before the update). It is the expected reward value of the system for executing action a in state s before the current iteration. This value is used to calculate the TD error (TemporalDifference error), which is the difference between the current expected reward and the actual reward. The system updates the Q value by correcting this difference.

[0123] max(Q(s′,a′)) is the maximum Q value of all possible actions a′ in the next state s′, indicating the maximum expected reward that the system can obtain in the next state s′. This value is used to help the system evaluate the maximum future expected reward that can be obtained after entering the next state s′ from the current state s, and guide the action selection in the current state.

[0124] α represents the learning rate, which determines how close the Q value is to the new value each time it is updated, and its value range is between 0 and 1. If the α value is large, the Q value will update to the new value faster, but it may cause instability; if the α value is small, the Q value update speed will slow down, the model convergence will be more stable but more iterations will be required.

[0125] r is the immediate reward value obtained from the environment after executing action a. The reward value r reflects the feedback obtained by taking action a in the current state s. If r is a positive value, it means that this action brings positive feedback; if it is a negative value, it means that this action is not effective.

[0126] γ refers to the discount factor, which is between 0 and 1 and measures the importance of future rewards. If γ is close to 1, it means that the system will pay more attention to future long-term rewards; if γ is close to 0, the system will pay more attention to immediate short-term rewards. By adjusting γ, you can control whether the model pursues long-term benefits or short-term gains.

[0127] max(Q(s′,a′)) is the maximum value of the Q values ​​of all possible actions a′ in the next state s′. It represents the maximum expected reward that can be obtained in the next state s′. The Q-learning algorithm selects the best action in the current state by maximizing the expected reward in the future.

[0128] The part α[r+γmax(Q(s′,a′))-Q(s,a)] is called TD error (Temporal Differenceerror), which represents the difference between the expected value and the current value. This part is used to measure the gap between the current Q value and the maximum reward obtained from the next state, and adjust the Q value through the learning rate to gradually approach the ideal value.

[0129] After each iteration, the Q-table is updated based on the customer feedback data, and the expected reward value of each recommendation strategy is gradually adjusted using the update formula of the reinforcement learning algorithm. Different update methods can be used, such as the Dyna-Q algorithm, to perform more efficient Q-value updates by simulating future customer behavior and feedback.

[0130] After each iteration, based on the results of the Q value update, the position of the cluster center is readjusted to optimize the parameters of the recommendation model. The accuracy of the recommendation model is improved by continuously reducing the distance between the data point and the cluster center. Clustering algorithms (such as K-means) are used to update the position of the cluster center, and the cluster center is adjusted based on customer feedback to make the recommendation results more in line with customer needs. Combined with reinforcement learning to optimize the recommendation strategy, the model will dynamically adjust the recommended content. Other clustering algorithms (such as DBSCAN) can be used in combination with different distance measurement methods (such as Manhattan distance or cosine similarity) to update the cluster center.

[0131] Through multiple iterations and continuous updates of the Q-table, the distance between the data point and the cluster center is gradually reduced. With each iteration, the parameters of the recommendation model will gradually converge and eventually be optimized to the optimal state. By combining reinforcement learning and clustering algorithms, the model parameters are continuously optimized iteratively, making the recommendation results more in line with the personalized needs of customers. An adaptive learning rate can be used to dynamically adjust the Q value update rate according to the number of iterations and convergence speed to accelerate convergence.

[0132] This embodiment combines reinforcement learning and cluster analysis, and adjusts the model parameters iteratively for multiple times, which can gradually optimize the housekeeping service recommendation results and ensure that the recommendation results are highly matched with customer needs. It can effectively improve the accuracy of recommendations and customer satisfaction, while continuously optimizing model parameters to achieve an organic combination of dynamic adjustment and personalized recommendations.

[0133] In one embodiment, in the above S40, a control experiment is used to evaluate and optimize the preliminary housekeeping service recommendation model to generate a personalized housekeeping service recommendation model, including:

[0134] S407, using the preliminary housekeeping service recommendation model as an experimental group and the preset recommendation strategy as a control group, and tracking core indicators including recommendation accuracy, click-through rate, and customer retention rate;

[0135] S408, collecting the recommendation result data of the experimental group and the control group, including the accuracy of the recommendation, the click status of the recommendation result, and the actual conversion rate after the service selection, and comparing and analyzing the specific performance of the recommendation result data of the control group and the experimental group on the core indicators;

[0136] S409, optimizing weights and hyperparameters in the preliminary housekeeping service recommendation model based on the analysis results of the recommendation result data;

[0137] S410, generating the personalized housekeeping service recommendation model through multiple rounds of control experiments and model optimization.

[0138] In this embodiment, in order to evaluate the effect of the preliminary housekeeping service recommendation model, a controlled experiment method is adopted. The preliminary recommendation model is applied to the customers in the experimental group, and the traditional preset recommendation strategy (such as recommendation based on historical preferences or rules) is applied to the customers in the control group. The core indicators include the accuracy of the recommendation (whether the housekeeping service recommended by the model meets the customer's needs), the click-through rate (the proportion of customers clicking on the recommended service), and the customer retention rate (the probability that customers continue to use the housekeeping service after using it). The A / B testing platform is used for grouping and monitoring. The experimental group uses the preliminary recommendation model to generate recommended content, and the control group uses the existing traditional recommendation strategy. The system tracks the customer's behavior data (such as clicks and records of selecting housekeeping services) and records the core indicator data. Multiple groups of experimental groups and control groups can be used for stratified experiments, and the experimental effects can be further subdivided according to different customer characteristics (such as age, income, property type, etc.).

[0139] By tracking customer behavior in real time, collect the recommendation result data of the experimental group and the control group. The recommendation accuracy is evaluated by whether the customer chooses the recommended service, the click-through rate is measured by whether the customer clicks on the recommended service link, and the conversion rate refers to whether the customer finally uses the recommended housekeeping service. Use data collection tools to track customer behavior in real time, store the collected recommendation data (such as clicks, conversion rates) in the data warehouse, and use statistical analysis tools (such as the pandas library in Python) to compare and analyze the recommendation effects of the experimental group and the control group. Use customer satisfaction surveys as additional evaluation indicators, and supplement the evaluation of conversion rate and retention rate through questionnaire feedback.

[0140] Based on the experimental results, optimize the feature weights and hyperparameters in the preliminary recommendation model. For example, if a certain type of feature (such as the customer's real estate information) has a greater impact on the accuracy of the recommendation, the system will increase the weight of this feature. Hyperparameters (such as the number of cluster centers, learning rate, etc.) can be adjusted through tuning methods such as grid search or random search to improve the overall performance of the model. Use automated parameter adjustment tools (such as grid search or Bayesian optimization) to adjust the weights and hyperparameters of the model. By evaluating the feedback on the recommendation effect, select the optimal parameter combination to improve the model's recommendation accuracy and customer conversion rate. Reinforcement learning algorithms can be introduced to dynamically adjust the model's weights and parameters based on real-time customer feedback to further improve the model's personalized recommendation effect.

[0141] Experimentation and optimization is a cyclical process. Through multiple rounds of experiments, the recommendation strategy and parameters of the model are continuously optimized. After each round of experiments, the system will adjust the model according to the latest experimental data until the model performance reaches the expected effect, and finally generate a personalized housekeeping service recommendation model. A / B testing is combined with automatic parameter adjustment to continuously optimize model parameters, gradually improve recommendation accuracy, click-through rate and customer retention rate, and generate a personalized housekeeping service recommendation model. The experimental cycle can be flexibly adjusted according to business needs. Short-term experiments are used to quickly optimize recommendation strategies, and long-term experiments are used to analyze customers' long-term behavior patterns.

[0142] This embodiment uses a control experiment to evaluate the preliminary recommendation model. The system can verify the effectiveness of the model in actual business and objectively evaluate the accuracy of personalized recommendations and customer satisfaction by comparing the performance of the experimental group and the control group. Multiple rounds of experiments and optimization ensure the dynamic adjustment and improvement of the model. The personalized housekeeping service recommendation model finally generated can more accurately meet customer needs and improve the conversion rate and customer retention rate of recommended services.

[0143] In one embodiment, the above S30 includes:

[0144] S301, extracting categorical features and original numerical features from the processed data, and creating derived features based on business knowledge;

[0145] S302, encoding the extracted categorical features, converting them into numerical features through one-hot encoding or label encoding, and normalizing or standardizing the converted numerical features and the original numerical features;

[0146] S303, using a chi-square test or an ANOVA test method, selecting a feature that affects the demand for domestic services from the derived features, the converted numerical features, and the original numerical features;

[0147] S304, reducing the number of features by recursive feature elimination or random forest feature importance ranking, and extracting the key features for housekeeping service recommendation.

[0148] In this embodiment, categorical features (such as the region where the customer is located, the type of house, the type of insurance product, etc.) and original numerical features (such as property area, insurance amount, number of family members, etc.) are extracted from the processed data. Based on business knowledge, derived features are created, for example, by combining existing features to generate new features (such as "house area / number of family members" as living density features). Feature engineering tools (such as pandas and sklearn libraries in Python) can be used to extract and create new derived features. These derived features are defined according to business scenarios and can be generated by performing arithmetic operations or logical combinations on the original features. Complex interactive features can also be further created through customer historical behavior data, such as the cross-features of customer service selection frequency and the number of family members.

[0149] Categorical features (such as region, house type, etc.) need to be converted to numerical features. Common encoding methods include One-Hot Encoding and Label Encoding. One-Hot Encoding represents each category as a binary vector, while label encoding maps categories to integers. Numerical features and encoded categorical features need to be normalized or standardized to ensure that the features are input into the model on the same scale. You can use OneHotEncoder or LabelEncoder in sklearn to encode categorical features. For numerical features, use Min-Max Scaling or Z-score normalization to compress the data into the range of 0-1 or standard normal distribution. For high-cardinality categorical features (features with a very large number of categories), you can choose frequency encoding or target encoding to process them to avoid dimensionality explosion caused by one-hot encoding.

[0150] The chi-square test is used to determine the correlation between categorical features and the demand for domestic services, and the ANOVA test is used to evaluate the impact of numerical features on the demand for domestic services. Through these statistical test methods, the system can filter out features that are highly correlated with the demand for domestic services. Use the chi2 function in sklearn for the chi-square test and f_classif for the ANOVA test. Compare the statistical value of each feature with the critical value to filter out features that significantly affect the demand for domestic services. You can set the significance level (p-value) threshold according to business needs, and only features that pass the test will be retained for subsequent models.

[0151] Recursive feature elimination (RFE) iteratively trains the model to remove less influential features one by one, and finally retain the most influential features. Random forest feature importance ranking is based on the decision tree model, which selects the most influential features by evaluating the contribution of each feature to the prediction accuracy. Use the RFE function in sklearn for recursive feature elimination, or use the feature importance score in the random forest model (RandomForestClassifier) ​​for sorting, and finally filter out the key features for recommendation. You can combine multiple feature selection methods (such as L1 regularization, PCA) to further reduce the feature dimension and improve the training efficiency of the model.

[0152] In this embodiment, by extracting and selecting key features that affect the demand for housekeeping services from the processed data, the system can improve the accuracy and efficiency of the recommendation model. The process of feature extraction and selection ensures the high quality and relevance of the model input and reduces the noise caused by redundant data. Through recursive feature elimination or random forest feature importance ranking, the system can efficiently screen out features that have a significant impact on recommendation decisions, optimize the accuracy of housekeeping service recommendations, and thus improve customer satisfaction and user experience.

[0153] In one embodiment, the above S20 includes:

[0154] S201, cleaning the collected data and filling in missing data by interpolation or K-nearest neighbor method;

[0155] S202, detect and handle outliers using box plots, Z-score or interquartile range methods, and replace outliers with non-outlier values ​​or medians;

[0156] S203, performing consistency check on the data fields, and converting the data fields in different formats into a unified format;

[0157] S204, normalizing or standardizing the numerical data so that the data are processed within the same scale range.

[0158] In this embodiment, missing data will have a negative impact on the accuracy and stability of the model, so it needs to be filled. Common methods include interpolation and K-nearest neighbor method (KNN). Interpolation fills missing data by calculating the values ​​between existing data points, while KNN method fills in by finding neighboring data points similar to the missing data. Interpolation can be used to fill in numerical data, using linear interpolation or polynomial interpolation; KNN method can be used for categorical or numerical data, using similarity (such as Euclidean distance) to determine the closest K neighbors, and fill in missing values ​​by their average or mode. Different filling strategies can be selected according to different data characteristics. For example, for time series data, linear interpolation is more common; for multidimensional data sets, KNN method can fill in missing values ​​more accurately.

[0159] Outliers are extreme values ​​in the data, which may be caused by data input errors or other reasons. Common outlier detection methods include box plots, Z-scores, and interquartile range methods. Detected outliers can be replaced by medians or non-outliers to ensure the normality of the data. When using the Z-score method, the data is standardized, and Z-score values ​​exceeding a certain threshold (such as ±3) can be considered outliers. The interquartile range method calculates the interquartile range (IQR) of the data, and values ​​that exceed the upper and lower limits can be considered outliers. Box plots can intuitively display data distribution and outliers. For outliers in categorical data, they can be replaced according to the categories with higher frequency or common categories. For numerical outliers, use the median or values ​​within the upper and lower quartiles to replace them.

[0160] When data comes from different sources, there are often inconsistencies in field names, types, or formats. Data consistency checks detect and unify the data field formats from different sources to ensure that the data can be correctly parsed and used in subsequent processing. Use programming languages ​​(such as Python) to check the format type of the field, and use regular expressions or built-in formatting tools to unify the format of date, currency, and text data. For numeric fields, convert the string format to a numeric field to ensure consistency. The formats of dates, currencies, etc. may be different in different countries and regions. You need to select the appropriate format standard based on the specific business scenario, such as unifying different date formats into the ISO format (YYYY-MM-DD).

[0161] In order to make the data have a consistent scale when training the model, numerical data usually needs to be normalized or standardized. Normalization compresses the data to a specific range (such as between 0 and 1), while standardization converts the data to a normal distribution with a mean of 0 and a standard deviation of 1. The min-max normalization method scales the numerical data to a specified range, and the Z-score standardization standardizes the data based on the mean and standard deviation of the data. These techniques can be implemented using the MinMaxScaler and StandardScaler functions in the sklearn library. Normalization is more suitable for data sets with a relatively fixed range, while standardization is suitable for scenarios with a normal distribution. The specific choice depends on the actual distribution characteristics of the data and the model requirements.

[0162] This embodiment cleans the data, processes missing values ​​and outliers, and performs format conversion and standardization. The system can ensure that the data input to the model is of high quality, unified format, and consistent dimensions. This not only improves the accuracy of model training, but also reduces the negative impact of data noise and outliers on the model. The cleaned data can effectively reduce the errors in the model, improve the accuracy of housekeeping service recommendations and customer satisfaction, and provide a reliable data foundation for subsequent feature selection and model optimization.

[0163] In one embodiment, the above S50 includes:

[0164] S501, obtaining a recommended home property insurance plan for the customer, and extracting insurance type, premium, and house information from the recommended home property insurance plan;

[0165] S502, taking the extracted insurance type, premium and house information as input and importing them into the personalized housekeeping service recommendation model;

[0166] S503, generating personalized housekeeping service recommendations that match customer needs by combining the personalized housekeeping service recommendation model with the customer's historical behavior data;

[0167] S504: Arrange and display the housekeeping service recommendation contents in order of priority according to the personalized housekeeping service recommendation.

[0168] In this embodiment, the system first extracts key information from the customer's recommended home property insurance plan, which includes insurance type (such as house insurance, property insurance), premium amount, and detailed information related to the house (such as house area, house type, etc.). By parsing the customer's insurance plan, the API interface can be used to extract data from the insurance system, or to extract these key information from the electronic policy document. Natural language processing technology can be used to extract key fields from text-formatted insurance policies. According to the policy formats of different insurance companies, regular expressions, NLP models and other methods can be used to extract information to ensure that information such as insurance type and premium can be consistently obtained in different document formats.

[0169] The insurance type, premium, and house information extracted from the insurance recommendation plan are used as input features of the personalized housekeeping service recommendation model. This information helps predict customers' demand for housekeeping services. For example, customers with larger houses may need deep cleaning services, while customers with high premiums may prefer high-end housekeeping services. Through preprocessing and encoding, the input insurance type (categorical data) is converted into numerical features (such as one-hot encoding), and the premium and house information are normalized or standardized to ensure that different types of features can be effectively used by the recommendation model. The input features can adjust the weights according to the specific needs of the customers to ensure that the more important features in the model (such as insurance type and house area) have a greater impact on the recommendation results.

[0170] The personalized housekeeping service recommendation model combines insurance information with the customer's historical behavior data (such as previously selected housekeeping services, service frequency, etc.) to generate personalized recommendations that meet customer needs. Recommendations may include cleaning services, repair services, maintenance services, etc. Use machine learning or deep learning models (such as collaborative filtering, random forests, deep neural networks) to integrate customer historical behavior with insurance information, and generate a recommended list of housekeeping services through model learning. Historical behavior data can be further subdivided into customer selection preferences, price sensitivity, etc. Use a hybrid recommendation system to combine content-based recommendations with collaborative filtering to generate more accurate recommendation results. Historical behavior data can also be used to dynamically adjust the recommended service types.

[0171] Based on the recommendation results generated by the model, the system arranges the recommendations in order of priority, giving priority to highly relevant housekeeping services that meet customer needs. Priority can be based on the urgency of the service, the customer's historical selection preferences, or the popularity of the service. The recommendation results are prioritized using a sorting algorithm, which can be based on indicators such as relevance score and historical click-through rate. The final list of housekeeping services displayed to the customer will be ranked from high to low priority. The order of recommended content can be dynamically adjusted based on different customer types, such as giving priority to high-end services for high-net-worth customers and giving priority to cost-effective services for thrifty customers.

[0172] By using a personalized housekeeping service recommendation model, this embodiment enables the system to generate highly personalized housekeeping service recommendations based on the customer's insurance plan and historical behavior data. The recommendation model not only improves the accuracy of the recommendation, but also optimizes the display order of the recommended content through priority sorting, so that customers can quickly find the service type that meets their needs. This personalized recommendation mechanism improves the customer experience, increases the selection rate and satisfaction of housekeeping services, and optimizes the value of insurance supplementary services.

[0173] The present invention also provides a housekeeping service recommendation device based on home property insurance, referring to Figure 2 , Figure 2 This is a functional module diagram of a preferred embodiment of a home service recommendation device based on home property insurance of the present invention. The home service recommendation device based on home property insurance includes:

[0174] Data collection module, used to collect customers' real estate information, regional information, customer information, home property insurance information and service provider information;

[0175] Data cleaning module, used to clean the collected data, process missing values ​​and outliers, and perform data format conversion and standardization;

[0176] A feature extraction module is used to extract features that affect the demand for domestic services from the processed data and determine key features for domestic service recommendations;

[0177] A model training and optimization module, used to establish a preliminary housekeeping service recommendation model based on the key features, evaluate and optimize the preliminary housekeeping service recommendation model using a control experiment, and generate a personalized housekeeping service recommendation model;

[0178] The housekeeping service recommendation output module is used to output personalized housekeeping service recommendations using the personalized housekeeping service recommendation model.

[0179] The specific implementation of the device for recommending housekeeping services based on home property insurance of the present invention is basically the same as the various embodiments of the method for recommending housekeeping services based on home property insurance described above, and will not be described in detail here.

[0180] The present invention also provides a housekeeping service recommendation device based on household property insurance, such as Figure 3 As shown, the housekeeping service recommendation device based on household property insurance may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0181] Those skilled in the art will understand that Figure 3 The hardware structure of the home service recommendation device based on home property insurance shown in the figure does not constitute a limitation of the home service recommendation device based on home property insurance, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0182] like Figure 3 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a home service recommendation program based on home property insurance. The operating system is a program for managing and controlling the home service recommendation equipment and software resources based on home property insurance, and supports the operation of the network communication module, the user interface module, the home service recommendation program based on home property insurance, and other programs or software; the network communication module is used to manage and control the network interface 1004; the user interface module is used to manage and control the user interface 1003.

[0183] exist Figure 3 In the hardware structure of the household property insurance-based housekeeping service recommendation device shown, the network interface 1004 is mainly used to connect to the background server and communicate data with the background server; the user interface 1003 is mainly used to connect to the client and communicate data with the client; the processor 1001 can call the household property insurance-based housekeeping service recommendation program stored in the memory 1005, and perform the same operation as the household property insurance-based housekeeping service recommendation method.

[0184] The specific implementation of the home service recommendation device based on home property insurance of the present invention is basically the same as the various embodiments of the home service recommendation method based on home property insurance described above, and will not be repeated here.

[0185] In addition, an embodiment of the present invention also proposes a computer storage medium, on which a home service recommendation program based on home property insurance is stored. When the home service recommendation program based on home property insurance is executed by a processor, the steps of the home service recommendation method based on home property insurance as described above are implemented.

[0186] The specific implementation of the computer storage medium of the present invention is basically the same as the above-mentioned embodiments of the housekeeping service recommendation method based on home property insurance, and will not be repeated here.

[0187] It should be noted that if software tools or components other than those of our company appear in the embodiments of the present application, they are only used for illustrative purposes and do not represent actual use.

Claims

1. A method for recommending housekeeping services based on household property insurance, characterized in that: The following steps are involved: Collect the client's property information, region information, customer information, home insurance information and service provider information; Clean the collected data, process missing values ​​and outliers, and convert and standardize the data format; From the processed data, extract the features that affect the demand for domestic services and determine the key features for domestic service recommendations; Establishing a preliminary housekeeping service recommendation model based on the key features, evaluating and optimizing the preliminary housekeeping service recommendation model through a control experiment, and generating a personalized housekeeping service recommendation model; The personalized housekeeping service recommendation model is used to output personalized housekeeping service recommendations.

2. The method for recommending housekeeping services based on home property insurance as claimed in claim 1, characterized in that: A preliminary housekeeping service recommendation model is established based on the key features, including: Dividing the key features into a training data set and a validation data set; Constructing an initial housekeeping service recommendation model, randomly selecting a preset number of data points from the training data set as initial cluster centers; Based on the distance between each data point in the training data set and the initial cluster center, the data point is assigned to the nearest initial cluster center, and the position of the initial cluster center is updated; Through multiple iterations, adjusting the parameters of the initial housekeeping service recommendation model, gradually reducing the distance between each data point and the cluster center; Use the validation set to evaluate the initial housekeeping service recommendation model and adjust the location of the cluster center; When the position change of the cluster center is less than the preset convergence condition, the training is stopped to obtain the preliminary housekeeping service recommendation model.

3. The method for recommending housekeeping services based on home property insurance as claimed in claim 2, characterized in that: Through multiple iterations, the parameters of the initial housekeeping service recommendation model are adjusted to gradually reduce the distance between each data point and the cluster center, including: Initialize the Q-table's recommended strategy Q value to zero or a random value; In each iteration, the initial housekeeping service recommendation model is used to generate recommendation results, feedback data is collected, and a reward value is generated based on the feedback data; Update the recommended strategy Q value in the Q-table based on the reward value generated in each iteration; Updating the location of the cluster center and optimizing the parameters of the initial housekeeping service recommendation model based on the reward value of each iteration; The Q value of the recommendation strategy is updated through multiple iterations, gradually reducing the gap between each data point and the cluster center.

4. The method for recommending housekeeping services based on home property insurance as claimed in claim 1, characterized in that: A control experiment is used to evaluate and optimize the preliminary housekeeping service recommendation model to generate a personalized housekeeping service recommendation model, including: Using the preliminary housekeeping service recommendation model as the experimental group and the preset recommendation strategy as the control group, and tracking core indicators including recommendation accuracy, click-through rate and customer retention rate; Collect the recommendation result data of the experimental group and the control group, including the accuracy of the recommendations, the clicks on the recommendation results, and the actual conversion rate after the service selection, and compare and analyze the specific performance of the recommendation result data of the control group and the experimental group on the core indicators; Optimizing weights and hyperparameters in the preliminary housekeeping service recommendation model based on the analysis results of the recommendation result data; The personalized housekeeping service recommendation model is generated through multiple rounds of control experiments and model optimization.

5. The method for recommending housekeeping services based on home property insurance as claimed in claim 1, characterized in that: From the processed data, the features that affect the demand for domestic services are extracted, and the key features for domestic service recommendations are determined, including: Extract categorical features and raw numerical features from processed data, and create derived features based on business knowledge; Encoding the extracted categorical features, converting them into numerical features through one-hot encoding or label encoding, and normalizing or standardizing the converted numerical features and the original numerical features; Using a chi-square test or an ANOVA test method, selecting features that affect the demand for domestic services from the derived features, the converted numerical features, and the original numerical features; The number of features is reduced by recursive feature elimination or random forest feature importance ranking to extract the key features for housekeeping service recommendation.

6. The method for recommending housekeeping services based on home property insurance as claimed in claim 1, characterized in that: Clean the collected data, process missing values ​​and outliers, and convert and standardize the data format, including: Clean the collected data and fill in missing data by interpolation or K-nearest neighbor method; Detect and handle outliers using box plots, Z-scores, or interquartile range methods, and replace outliers with non-outliers or the median; Perform consistency checks on data fields and convert data fields in different formats into a unified format; Normalize or standardize numerical data so that the data are processed within the same scale range.

7. The method for recommending housekeeping services based on home property insurance as claimed in claim 1, characterized in that: Outputting personalized housekeeping service recommendations using the personalized housekeeping service recommendation model includes: Obtain a recommended home insurance plan for the customer, and extract insurance type, premium, and housing information from the recommended home insurance plan; The extracted insurance type, premium and housing information are used as input and imported into the personalized housekeeping service recommendation model; Through the personalized housekeeping service recommendation model, combined with the customer's historical behavior data, personalized housekeeping service recommendations that match customer needs are generated; According to the personalized housekeeping service recommendation, the housekeeping service recommendation contents are arranged and displayed in order of priority.

8. A housekeeping service recommendation device based on household property insurance, characterized in that: The housekeeping service recommendation device based on household property insurance comprises: Data collection module, used to collect customers' real estate information, regional information, customer information, home property insurance information and service provider information; Data cleaning module, used to clean the collected data, process missing values ​​and outliers, and perform data format conversion and standardization; A feature extraction module is used to extract features that affect the demand for domestic services from the processed data and determine key features for domestic service recommendations; A model training and optimization module, used to establish a preliminary housekeeping service recommendation model based on the key features, evaluate and optimize the preliminary housekeeping service recommendation model using a control experiment, and generate a personalized housekeeping service recommendation model; The housekeeping service recommendation output module is used to output personalized housekeeping service recommendations using the personalized housekeeping service recommendation model.

9. A housekeeping service recommendation device based on home property insurance, characterized in that: The household property insurance-based housekeeping service recommendation device includes a memory, a processor, and a household property insurance-based housekeeping service recommendation program stored in the memory and executable on the processor. When the household property insurance-based housekeeping service recommendation program is executed by the processor, the steps of the household property insurance-based housekeeping service recommendation method as described in any one of claims 1 to 7 are implemented.

10. A computer storage medium, characterized in that: The storage medium stores a home service recommendation program based on home property insurance, and when the home service recommendation program based on home property insurance is executed by the processor, the steps of the home service recommendation method based on home property insurance as described in any one of claims 1-7 are implemented.

Images