Insurance product intelligent pushing mode based on deep neural network

Abstract

The invention relates to the technical field of insurance pushing, in particular to an insurance product intelligent pushing mode based on a deep neural network, which comprises a user portrait construction module, a causal reasoning module, a personalized pushing module and a feedback optimization module, and is characterized in that a user portrait is constructed through extracted features and a modeling result, and the user portrait is pushed to the user. The causal reasoning module is based on mass user data, determines a variable to be deduced and a conditional probability through the model establishment module, establishes a model for Bayesian reasoning, and calculates a posterior probability based on the Bayesian theorem through the calculation analysis module; and through the posterior probability, the client with the higher purchase intention for the insurance product can be analyzed and judged. According to the invention, on the basis of existing massive customer data, portrait customer characteristics and customer touch data are analyzed through customer purchase, and characteristics, demands and behavior preferences of customers are described through deep analysis, so that targeted sales strategies and high-quality services are provided for marketing personnel of a company.

Description

Technical Field

[0001] The present invention relates to the field of insurance push technology, and in particular to an insurance product intelligent push mode based on a deep neural network. Background Art

[0002] With the rapid development of Internet technology, the insurance industry is gradually transforming towards digitalization, among which intelligent push of insurance products has become an important means to improve user experience and enhance market competitiveness. At present, the mainstream intelligent push systems for insurance products on the market mainly rely on the following technical implementation solutions: Push based on user portraits: By collecting basic information of users (such as age, gender, occupation), historical browsing behavior, purchase records, etc., build user portraits, and then recommend products based on the preferences of similar user groups; Collaborative filtering algorithm: Analyze the similarity between users and push based on the logic of "if user A likes product X, then user B who is similar to user A may also like product X"; Content-based recommendation: Analyze the content characteristics of insurance products (such as coverage, premiums, claims conditions, etc.), and recommend similar or complementary products based on the content characteristics of products that users have previously browsed or purchased.

[0003] However, the current push methods often lack in-depth personalized understanding. User portraits and collaborative filtering are often based on surface features, making it difficult to capture users’ deep insurance needs and their changes, resulting in insufficient dynamic adaptability. User needs and preferences are dynamically changing, and existing systems are often based on static or historical data. It is difficult to adjust push strategies in a timely manner to adapt to these changes, making it impossible to provide the company’s marketers with targeted sales strategies and high-quality services. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent push mode for insurance products based on deep neural networks.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: an intelligent push mode of insurance products based on deep neural network, including a user portrait construction module, a causal reasoning module, a personalized push module and a feedback optimization module:

[0006] The user portrait construction module collects the basic data, historical behavior data, and customer contact data of a large number of existing customers through multiple channels through the data collection module, extracts features from the collected data, and then constructs a portrait model based on the extracted data features through the data modeling module, and constructs a user portrait based on the extracted features and the modeling results;

[0007] The causal reasoning module is based on the collected multi-channel user data. The model building module determines the variables and conditional probabilities to be inferred and builds a model for Bayesian reasoning. The calculation and analysis module then calculates the posterior probability based on the Bayesian theorem. The posterior probability can be used to analyze and determine which customers have a higher willingness to purchase insurance products.

[0008] The personalized push module converts the data into feature representations available to the deep neural network through the data conversion module based on the extracted user features and the results of the causal reasoning module, and then builds a deep neural network model suitable for personalized recommendation through the model design and training module, and uses the extracted user features and causal reasoning results to train the deep neural network model to optimize the push strategy, and uses the push generation module to dynamically adjust customers with high purchase intention characteristics that can be pushed in combination with the optimized push strategy, and use multiple push methods to push to the adjusted customers;

[0009] The feedback optimization module collects real-time feedback and behavior data of users after receiving push notifications through multiple channels through the feedback collection module, and then uses the machine learning algorithm to regularly update the user portrait based on the collected feedback data through the update optimization module, and uses the online learning mechanism to update and optimize the causal reasoning model in real time.

[0010] As a further solution of the present invention, the user portrait component module includes a data acquisition module and a data modeling module, and the data acquisition module includes a data collection module and a feature extraction module;

[0011] The data collection module is used to collect basic data, historical behavior data and customer contact data of existing customers;

[0012] The feature extraction module extracts user basic features, user behavior features and derived features from the collected data based on the data collection module.

[0013] As a further solution of the present invention, the data modeling module includes a machine learning module and a labeling module;

[0014] The machine learning module is used to select an appropriate machine learning algorithm, conduct in-depth analysis of user characteristics, and mine potential user behavior patterns;

[0015] The labeling module is used to classify and aggregate different characteristics of users, label users accordingly, and form clear user classification.

[0016] As a further solution of the present invention, the causal reasoning module includes a model building module and a calculation and analysis module, and the model building module includes an event definition module and a priori probability module;

[0017] The event definition module is used to set target events and define potential causes;

[0018] The prior probability module is used for a preliminary view of the cause in the causal reasoning module before new data is observed.

[0019] As a further solution of the present invention, the computing and analyzing module includes a computing module and an analyzing module;

[0020] The calculation module is used to calculate the conditional probability and the posterior probability. The posterior probability can be calculated by setting the prior probability and the calculated conditional probability in conjunction with the Bayesian theorem;

[0021] The analysis module analyzes and determines which customers have a higher willingness to purchase insurance products based on the posterior probability calculated by the calculation module.

[0022] As a further solution of the present invention, the personalized push module includes a data conversion module, a model design training module and a push generation module, and the data conversion module includes a discrete feature encoding module and a continuous feature processing module;

[0023] The discrete feature encoding module is used to process categorical features using one-hot encoding;

[0024] The continuous feature processing module is used to perform normalization processing on numerical features.

[0025] As a further solution of the present invention, the model design training module includes a design module and a training module;

[0026] The design module constructs a deep neural network model suitable for personalized recommendation based on a multi-layer perceptron, a deep learning recommendation model or a sequence model;

[0027] The training module is based on the design module and trains the deep neural network model through the extracted user features and causal reasoning results.

[0028] As a further solution of the present invention, the push generation module includes a generation module and a push module;

[0029] The generation module can dynamically generate customers with high willingness suitable for push based on the optimized push strategy;

[0030] The push module can use a variety of push methods to push customers with higher purchasing intention to marketing personnel.

[0031] As a further solution of the present invention, the feedback optimization module includes a feedback collection module and an update optimization module;

[0032] The feedback collection module is used to collect real-time feedback and behavior data from users after they receive insurance product push notifications through various channels.

[0033] As a further solution of the present invention, the update optimization module includes a portrait update module and a model update module;

[0034] The portrait updating module is used to regularly update the user portrait;

[0035] The model updating module is used to perform real-time updates when the model receives new data, so that the model can adapt to the dynamic changes of user behavior.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are:

[0037] In the present invention, through the user portrait construction module, causal reasoning module, and personalized push module, it is possible to profile customer characteristics based on the existing massive customer data through customer purchase analysis, combine customer contact data, conduct in-depth analysis, and depict the causal relationship between customer characteristics and needs, and cooperate with the personalized recommendation model to provide the company's marketing personnel with targeted sales strategies and high-quality services.

[0038] In the present invention, the real-time feedback and behavior data of users after receiving the push are collected through multiple channels through the feedback collection module, and then the user portrait is regularly updated by the update optimization module based on the collected feedback data using a machine learning algorithm, and the causal reasoning model is updated and optimized in real time using an online learning mechanism to ensure the accuracy and adaptability of the push system, thereby enabling the system to dynamically adjust the push strategy, keep up with changes in user needs, and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a system flow chart of the present invention;

[0040] Figure 2 It is a schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] Embodiment 1

[0043] See also Figure 1 and Figure 2 The present invention provides a technical solution: an intelligent push mode for insurance products based on a deep neural network, including a user portrait construction module, a causal reasoning module, a personalized push module and a feedback optimization module:

[0044] The user portrait construction module collects the basic data, historical behavior data, and customer contact data of a large number of existing customers through multiple channels through the data collection module, extracts features from the collected data, and then constructs a portrait model based on the extracted data features through the data modeling module, and constructs a user portrait based on the extracted features and modeling results;

[0045] The causal reasoning module is based on the collected multi-channel user data. The model building module determines the variables and conditional probabilities to be inferred and builds a model for Bayesian reasoning. The calculation and analysis module then calculates the posterior probability based on the Bayesian theorem. The posterior probability can be used to analyze and determine which customers have a higher willingness to purchase insurance products.

[0046] The personalized push module is based on the extracted user features and the results of the causal reasoning module. The data conversion module converts the data into feature representations that can be used by the deep neural network. The model design and training module then builds a deep neural network model suitable for personalized recommendations. The extracted user features and causal reasoning results are used to train the deep neural network model to optimize the push strategy. The push generation module is used to dynamically adjust customers with high purchase intention characteristics that can be pushed in combination with the optimized push strategy, and push to the adjusted customers using multiple push methods.

[0047] The feedback optimization module collects real-time feedback and behavior data of users after receiving push notifications through multiple channels through the feedback collection module. Then, based on the collected feedback data, the update optimization module uses machine learning algorithms to regularly update user portraits and uses online learning mechanisms to update and optimize causal reasoning models in real time.

[0048] The user portrait component module includes a data acquisition module and a data modeling module. The data acquisition module includes a data collection module and a feature extraction module. The data collection module is used to collect basic data, historical behavior data and customer contact data of existing customers, where customer types include customers who have purchased and customers who are in contact. The basic data includes the customer's age, gender, annual income and occupation, etc. The historical behavior data includes different types of insurance products purchased by the customer and the frequency of purchase. The customer contact data is the face-to-face interview or invitation to the customer. The feature extraction module is based on the data collection module, and extracts user basic features, user behavior features and derived features from the collected data. The basic features can directly use the user's basic information, such as age, gender, geographic location, etc. The behavioral features include the user's activity level in different time periods, such as the frequency of users consulting a certain insurance product, the frequency of purchasing behavior and the user's preference for different categories of insurance products. The derived features are new features generated based on existing features, such as the ratio of the user's purchase amount to the contact consultation time. Before feature extraction, the collected data is cleaned by deduplication, filling missing values ​​and format standardization.

[0049] The data modeling module includes a machine learning module and a labeling module. The machine learning module is used to select appropriate machine learning algorithms, including decision trees, random forests, neural networks, etc., to conduct in-depth analysis of user characteristics and explore potential user behavior patterns. The labeling module is used to classify and aggregate different user characteristics, label users accordingly, and form a clear user classification.

[0050] The causal reasoning module includes a model building module and a computational analysis module. The model building module includes an event definition module and a prior probability module. The event definition module is used to set the target event A and define the potential feature B. For example, the target event A can be defined as the customer having a high willingness to purchase a certain type of insurance product, while the potential feature B includes user interests, promotional activities, social influence, etc. The prior probability module P(B) is used to obtain a preliminary view of the cause in the causal reasoning module before new data is observed, and the prior can be determined through historical data or industry research.

[0051] The calculation and analysis module includes a calculation module and an analysis module. The calculation module is used to calculate the conditional probability P(A|B) and the posterior probability P(B|A). The posterior probability can be calculated by setting the prior probability and the calculated conditional probability with Bayes' theorem. The calculation formula of the posterior probability is:

[0052]

[0053] Where P(A) can be calculated by the total probability formula:

[0054]

[0055] Based on the posterior probability calculated by the calculation module, the analysis module analyzes and determines which customers have the characteristics that have a higher willingness to purchase insurance products.

[0056] The personalized push module includes a data conversion module, a model design and training module, and a push generation module. The data conversion module includes a discrete feature encoding module and a continuous feature processing module. The discrete feature encoding module is used to process categorical features using one-hot encoding, and the continuous feature processing module is used to normalize numerical features.

[0057] The model design and training module includes a design module and a training module. The design module constructs a deep neural network model suitable for personalized recommendation based on a multi-layer perceptron, a deep learning recommendation model or a sequence model. The multi-layer perceptron can be used to directly input the feature vectors of users and projects, and perform feature learning through multiple fully connected layers. The deep learning recommendation model can combine user features and project features, and parse interactive information through an embedding layer. The sequence model includes a long short-term memory network and a gated recurrent unit, which are used to capture the time dependencies in user behavior sequences. The training module is based on the design module and trains the deep neural network model through the extracted user features and causal reasoning results.

[0058] The push generation module includes a generation module and a push module. The generation module scores or predicts customers based on the optimized push strategy and the data of existing customers in the database to obtain the customer's preference values ​​for different insurance products. Then, all customers are sorted according to the preference values ​​for a certain insurance product, and the top-ranked customers are recommended to marketers, so as to dynamically generate pushed customers and the insurance products preferred by the corresponding customers. The push module can push customers to marketers in a variety of ways, including notification push, SMS push, and social media push, etc., so as to improve the accuracy and personalization of recommendations and achieve the effect of personalized and precise push.

[0059] Embodiment 2

[0060] See also Figure 1 and Figure 2 The present invention provides a technical solution: an intelligent push mode for insurance products based on a deep neural network, including a user portrait construction module, a causal reasoning module, a personalized push module and a feedback optimization module:

[0061] The user portrait construction module collects the basic data, historical behavior data and customer contact data of a large number of existing customers through multiple channels through the data collection module, extracts features from the collected data, and then constructs a portrait model based on the extracted data features through the data modeling module, and constructs a user portrait based on the extracted features and modeling results;

[0062] The causal reasoning module is based on the collected multi-channel user data. The model building module determines the variables and conditional probabilities to be inferred and builds a model for Bayesian reasoning. The calculation and analysis module then calculates the posterior probability based on the Bayesian theorem. The posterior probability can be used to analyze and determine which customers have a higher willingness to purchase insurance products.

[0063] The personalized push module is based on the extracted user features and the results of the causal reasoning module. The data conversion module converts the data into feature representations that can be used by the deep neural network. The model design and training module then builds a deep neural network model suitable for personalized recommendations. The extracted user features and causal reasoning results are used to train the deep neural network model to optimize the push strategy. The push generation module is used to dynamically adjust customers with high purchase intention characteristics that can be pushed in combination with the optimized push strategy, and push to the adjusted customers using multiple push methods.

[0064] The feedback optimization module collects real-time feedback and behavior data of users after receiving push notifications through multiple channels through the feedback collection module. Then, based on the collected feedback data, the update optimization module uses machine learning algorithms to regularly update user portraits and uses online learning mechanisms to update and optimize causal reasoning models in real time.

[0065] The update optimization module includes a portrait update module and a model update module. The portrait update module is used to regularly update user portraits. The portrait update module is based on machine learning algorithms, such as cluster analysis and recommendation systems. The model update module is used to perform real-time updates when the model receives new data, so that the model can adapt to the dynamic changes in user behavior. The model update module is based on an online learning mechanism, which is a machine learning method, thereby ensuring the accuracy and adaptability of the push system, enabling the system to dynamically adjust the push strategy, keep up with changes in user needs, and improve the user experience.

[0066] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent push mode for insurance products based on deep neural networks, characterized in that: Including user portrait building module, causal reasoning module, personalized push module and feedback optimization module: The user portrait construction module collects the basic data, historical behavior data, and customer contact data of a large number of existing customers through multiple channels through the data collection module, extracts features from the collected data, and then constructs a portrait model based on the extracted data features through the data modeling module, and constructs a user portrait based on the extracted features and the modeling results; The causal reasoning module is based on the collected multi-channel user data. The model building module determines the variables and conditional probabilities to be inferred and builds a model for Bayesian reasoning. The calculation and analysis module then calculates the posterior probability based on the Bayesian theorem. The posterior probability can be used to analyze and determine which customers have a higher willingness to purchase insurance products. The personalized push module converts the data into feature representations available to the deep neural network through the data conversion module based on the extracted user features and the results of the causal reasoning module, and then builds a deep neural network model suitable for personalized recommendation through the model design and training module, and uses the extracted user features and causal reasoning results to train the deep neural network model to optimize the push strategy, and uses the push generation module to dynamically adjust customers with high purchase intention characteristics that can be pushed in combination with the optimized push strategy, and use multiple push methods to push to the adjusted customers; The feedback optimization module collects real-time feedback and behavior data of users after receiving push notifications through multiple channels through the feedback collection module, and then uses the machine learning algorithm to regularly update the user portrait based on the collected feedback data through the update optimization module, and uses the online learning mechanism to update and optimize the causal reasoning model in real time.

2. According to claim 1, a smart push mode for insurance products based on deep neural network is characterized in that: The user portrait component module includes a data acquisition module and a data modeling module, and the data acquisition module includes a data collection module and a feature extraction module; The data collection module is used to collect basic data, historical behavior data and customer contact data of existing customers; The feature extraction module extracts user basic features, user behavior features and derived features from the collected data based on the data collection module.

3. According to claim 2, a smart push mode for insurance products based on deep neural network is characterized in that: The data modeling module includes a machine learning module and a labeling module; The machine learning module is used to select an appropriate machine learning algorithm, conduct in-depth analysis of user characteristics, and mine potential user behavior patterns; The labeling module is used to classify and aggregate different characteristics of users, label users accordingly, and form clear user classification.

4. The insurance product intelligent push mode based on deep neural network according to claim 1 is characterized in that: The causal reasoning module includes a model building module and a calculation and analysis module, and the model building module includes an event definition module and a priori probability module; The event definition module is used to set target events and define potential causes; The prior probability module is used for a preliminary view of the cause in the causal reasoning module before new data is observed.

5. The insurance product intelligent push mode based on deep neural network according to claim 4 is characterized in that: The calculation and analysis module includes a calculation module and an analysis module; The calculation module is used to calculate the conditional probability and the posterior probability. The posterior probability can be calculated by setting the prior probability and the calculated conditional probability in conjunction with the Bayesian theorem; The analysis module analyzes and determines which customers have a higher willingness to purchase insurance products based on the posterior probability calculated by the calculation module.

6. The insurance product intelligent push mode based on deep neural network according to claim 1 is characterized in that: The personalized push module includes a data conversion module, a model design and training module, and a push generation module. The data conversion module includes a discrete feature encoding module and a continuous feature processing module. The discrete feature encoding module is used to process categorical features using one-hot encoding; The continuous feature processing module is used to perform normalization processing on numerical features.

7. The insurance product intelligent push mode based on deep neural network according to claim 6 is characterized in that: The model design training module includes a design module and a training module; The design module constructs a deep neural network model suitable for personalized recommendation based on a multi-layer perceptron, a deep learning recommendation model or a sequence model; The training module is based on the design module and trains the deep neural network model through the extracted user features and causal reasoning results.

8. The insurance product intelligent push mode based on deep neural network according to claim 6 is characterized in that: The push generation module includes a generation module and a push module; The generation module can dynamically generate customers with high willingness suitable for push based on the optimized push strategy; The push module can use a variety of push methods to push customers with higher purchasing intention to marketing personnel.

9. The insurance product intelligent push mode based on deep neural network according to claim 1 is characterized in that: The feedback optimization module includes a feedback collection module and an update optimization module; The feedback collection module is used to collect real-time feedback and behavior data from users after they receive insurance product push notifications through various channels.

10. The insurance product intelligent push mode based on deep neural network according to claim 9, characterized in that: The update optimization module includes a portrait update module and a model update module; The portrait updating module is used to regularly update the user portrait; The model updating module is used to perform real-time updates when the model receives new data, so that the model can adapt to the dynamic changes of user behavior.

Images