Customer behavior prediction and automatic response system based on artificial intelligence

Abstract

The invention discloses a customer behavior prediction and automatic response system based on artificial intelligence, relates to the technical field of customer behavior analysis, and solves the technical problem that the change condition of interest of a customer in interest product types is difficult to analyze in a targeted manner according to the interest product types of the customer. By analyzing the browsing record, the purchase record and the shopping cart data of the customer, a more accurate user portrait can be conveniently constructed, the interest field of the customer can be accurately identified, and personalized product recommendation is provided for the customer. By accurately identifying the interest field of the customer, the customer behavior data is distributed to the interest user data storage node of the corresponding product type for storage, and the subsequent data query and analysis efficiency is remarkably improved. Through the LSTM deep learning technology, the long-term dependency relationship in the time sequence can be effectively captured, so that the model can identify the purchase intention of the customer more accurately.

Description

Technical Field

[0001] The present invention belongs to the field of customer behavior analysis, and specifically relates to a customer behavior prediction and automatic response system based on artificial intelligence. Background Art

[0002] With the acceleration of digital transformation, enterprises are facing the rapid growth of a large amount of customer data. Customer behavior data includes user registration information, browsing history, purchase history, shopping cart data and customer feedback. These data provide enterprises with opportunities to gain in-depth understanding of customer needs and preferences. However, traditional data analysis methods are often unable to process these massive data in a timely manner, resulting in slow response of enterprises in formulating marketing strategies and optimizing customer experience. At the same time, the development of artificial intelligence (AI) technology has provided enterprises with new solutions. Through technologies such as machine learning, deep learning and natural language processing, enterprises can extract valuable information from complex data and achieve accurate prediction of customer behavior. In addition, the automated response system can adjust the marketing strategy in real time according to the prediction results, thereby improving customer satisfaction and conversion rate.

[0003] Most of the existing customer behavior prediction and automated response systems conduct a unified analysis of the customer's user portrait, but it is difficult to conduct a targeted analysis of the changes in the customer's interest in the product categories of interest to the customer. At the same time, there is a lack of comprehensive analysis of the customer's purchase intention and purchase desire loss of the product of interest, which also makes it difficult to accurately analyze the changes in interest in the product categories of interest. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an artificial intelligence-based customer behavior prediction and automated response system to solve the technical problem that it is difficult to analyze the changes in customers' interests in product categories of interest in a targeted manner.

[0005] To solve the above problems, the first aspect of the present invention provides a customer behavior prediction and automatic response system based on artificial intelligence, comprising:

[0006] Data collection module: used to collect customer behavior data through multiple channels authorized by customers, including user registration information, account activities, browsing history, purchase history, shopping cart data and customer feedback;

[0007] Storage node establishment module: divide all the commodities on the platform into product categories, and set up user data storage nodes of interest for the corresponding product categories in several regions with the most merchant registered addresses in each product category;

[0008] Data allocation module: Analyzes the product categories that customers are interested in based on browsing history, purchase history and shopping cart data, and allocates customer behavior data to the user data storage nodes of interest corresponding to the product categories for storage;

[0009] Feature analysis module: Based on the customer behavior data of the interest user data storage node, analyze the customer behavior characteristics, including: visit duration, frequency of visit to interest products, dwell time on interest products, and purchase conversion rate;

[0010] Behavior prediction module: Based on the analyzed customer behavior characteristics, the LSTM deep learning technology is used to capture time series characteristics, build a customer behavior prediction model, and predict the customer's purchase intention and optimal recommendation frequency;

[0011] Real-time analysis engine: Analyzes real-time customer behavior data through stream processing technology, analyzes the degree of loss of users' purchasing desire for products of interest, and dynamically adjusts the recommendation strategy.

[0012] Optionally, in an example of the above aspect, the data allocation module analyzes the product categories that the customer is interested in based on the browsing history, purchase history and shopping cart data, and allocates the customer behavior data to the interest user data storage node corresponding to the product category for storage, including the following steps:

[0013] Distributed storage module: used to store browsing records, purchase records and shopping cart data of the latest detection time period in the customer behavior data, and divide the latest detection time period into several detection time intervals;

[0014] Count the product categories that appear most frequently in the browsing records, purchase records, and shopping cart data during the user detection time period and the most recent detection time interval, and use the corresponding product categories as the product categories that the user is most interested in during the user detection time period and the product categories that the user is most interested in during the most recent detection time interval;

[0015] According to the product categories that the users are most interested in during the analyzed detection time period and the product categories that they are most interested in during the most recent detection time interval, customers are grouped into corresponding product category target groups;

[0016] The behavior data of customers in the corresponding product category target group is stored in the interest user data storage node of the corresponding product category.

[0017] Optionally, in an example of the above aspect, the feature analysis module analyzes the customer behavior features according to the customer behavior data of the interest user data storage node, including the following steps:

[0018] Account activity, browsing history, purchase history, shopping cart data and customer feedback of customer behavior data stored in interest user data nodes;

[0019] Analyzed customer behavior characteristics, including:

[0020] Count the total duration of user visits during the most recent detection time interval;

[0021] According to the user's browsing history, purchase history and time data in the shopping cart data within the most recent detection time interval, the user's daily visit frequency and the time spent on the products of interest are counted;

[0022] Count the number of times a user browses a product and the corresponding number of times a user purchases it within the most recent detection time interval, and use the ratio of the number of times a user browses a product to the corresponding number of times a user purchases it as the user's purchase conversion rate;

[0023] The obtained data on visit duration, frequency of visits to products of interest, dwell time on products of interest and purchase conversion rate are used to form a customer behavior feature matrix.

[0024] Optionally, in an example of the above aspect, the behavior prediction module captures time series features through LSTM deep learning technology based on the analyzed customer behavior characteristics, builds a customer behavior prediction model, and predicts the customer's purchase intention and optimal recommendation frequency, including the following steps:

[0025] Obtain the latest customer behavior data within the detection time period and subdivide the detection time interval into several test time intervals;

[0026] According to the customer behavior data of the customers in the test time period, the visit duration, the visit frequency of the products of interest, the stay time of the products of interest and the purchase conversion rate in the test time period are counted to build a customer behavior feature matrix;

[0027] Count the customer's purchase records and customer feedback data within each test time interval, count the number of purchases by the customer, and whether the recommended frequency in the feedback is greater than the demand, less than the demand, or the recommended frequency is appropriate;

[0028] Determine the test time interval as the time step and convert the feature matrix into a format suitable for LSTM input;

[0029] By taking the customer behavior feature matrix of the previous time step as input, the number of purchases of the customer in the next time step, and the feedback recommendation frequency greater than demand, less than demand, or appropriate recommendation frequency as output labels, the LSTM network model is trained to build a customer behavior prediction model.

[0030] Based on the customer behavior characteristics analyzed in real time by the feature analysis module, the customer behavior prediction model is constructed to predict the customer's purchase frequency, whether the recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate, and analyze the customer's purchase intention and the optimal recommendation frequency.

[0031] Optionally, in an example of the above aspect, by constructing a customer behavior prediction model, predicting the number of purchases of a customer, and whether the recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate, analyzing the customer's purchase intention and the optimal recommendation frequency, including the following steps:

[0032] Statistics are constructed through the customer behavior prediction model to predict the number of customer purchases, and the customer's purchase intention analysis value is calculated using the following formula:

[0033]

[0034] Among them, Pn is the customer's purchase intention analysis value, p is the predicted number of purchases by the customer, and Pt is the frequency of visits to the product of interest;

[0035] If the recommended frequency is greater than the demand, increase the current recommended frequency by 5%. If it is less than the demand, decrease the current recommended frequency by 5%. If the recommended frequency is appropriate, maintain the current recommended frequency.

[0036] Optionally, in an example of the above aspect, the real-time analysis engine analyzes the real-time customer behavior data through stream processing technology, analyzes the degree of loss of users' desire to purchase interested products, and dynamically adjusts the recommendation strategy, including the following steps:

[0037] Process customer behavior data through stream processing technology, count the real-time access duration, access frequency of products of interest, dwell time on products of interest, and purchase conversion rate of user data in each test time interval during the detection period, and draw the corresponding data waveform graph;

[0038] In the waveform chart, the time periods when the visit duration drops by more than 30%, the visit frequency of the interested products drops by more than 50%, and the stay time of the interested products drops by more than 50% are marked as the time periods of possible loss of purchase intention, and the time periods when the purchase conversion rate drops by more than 50% are marked as the time periods of loss of purchase intention;

[0039] Take the data waveform of a detection time interval before the purchase intention loss period as input, add the purchase intention loss label as output, take the data waveform of a detection time interval before the purchase intention loss period as input, add the purchase intention loss label as output, and train the deep learning model;

[0040] Process customer behavior data through stream processing technology, count the real-time access duration, access frequency, stay time and purchase conversion rate of user data in each test time interval in the most recent detection time interval, and draw the corresponding data waveform;

[0041] Input the data waveform of the most recent detection time interval into the trained deep learning model, and output a label of possible purchase intention loss, a label of purchase intention loss, or no label;

[0042] According to the output tag type, analyze the degree of loss of users' purchasing desire for the products of interest and dynamically adjust the recommendation strategy.

[0043] Optionally, in an example of the above aspect, analyzing the degree of loss of the user's desire to purchase the item of interest according to the output tag type and dynamically adjusting the recommendation strategy includes the following steps:

[0044] According to the output tag type and the customer's purchase intention analysis results, obtain the customer's shopping cart data in each test time interval within the detection time interval, including: shopping cart update frequency and shopping cart item browsing time;

[0045] The following formula is used to analyze the degree of loss of users' purchasing desire for products of interest:

[0046]

[0047] Among them, Pwa is the analysis result of the user's desire to buy the products of interest, Tci is the browsing time of the products in the shopping cart in the i-th test time interval, and Pci is the shopping cart update frequency in the i-th test time interval;

[0048] Dynamically adjust the recommendation strategy based on the analysis results of the user's loss of purchasing desire for the products of interest.

[0049] Optionally, in an example of the above aspect, dynamically adjusting the recommendation strategy according to the analysis result of the loss of the user's desire to purchase the interested product includes the following steps:

[0050] Compare the data of the most recent test time interval with the data of the previous test time interval within the detection time interval;

[0051] If the analysis result of the user's desire to buy the product of interest increases by more than 5%, and the customer's purchase intention analysis value decreases by more than 10%, the recommendation frequency of the customer's product of interest will be reduced by 5%, and the recommendation frequency of related products will be increased by 3%;

[0052] If the analysis result of the user's purchase intention loss for the product of interest increases by more than 5%, and the customer's purchase intention analysis value does not decrease by more than 10%, the recommendation frequency of the customer's product of interest will increase by 5%, and at the same time, the recommendation frequency of related products will decrease by 5%;

[0053] If the user's purchase intention loss analysis result for the product of interest does not increase by more than 5%, and the customer's purchase intention analysis value decreases by more than 10%, maintain the recommendation frequency of the customer's product of interest, and increase the recommendation frequency of related products by 3%;

[0054] If the analysis result of the user's purchase intention loss for the interested products does not increase by more than 5%, and the customer's purchase intention analysis value does not decrease by more than 10%, the recommendation frequency of the customer's interested products is maintained, and at the same time, the recommendation frequency of related products is maintained.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] The present invention facilitates the construction of a more accurate user portrait by analyzing the customer's browsing history, purchase history and shopping cart data, which helps to accurately identify the customer's areas of interest, thereby providing personalized product recommendations for the customer. By accurately identifying the customer's areas of interest, the customer behavior data is allocated to the interest user data storage node of the corresponding product category for storage, which significantly improves the efficiency of subsequent data query and analysis. When specific product category analysis is required, the relevant user and behavior data can also be quickly located to reduce query time. Real-time monitoring of customer behavior changes, timely adjustment of recommendation strategies to adapt to changes in customer needs.

[0057] The present invention can effectively capture the long-term dependencies in the time series through LSTM deep learning technology, so that the model can more accurately identify the customer's purchase intention; based on the prediction of purchase intention, it is convenient to provide more personalized and relevant product recommendations for each customer; through stream processing technology, the system can analyze customer behavior data in real time and timely identify the degree of loss of user's desire to buy goods of interest. According to the real-time analysis results, the recommendation strategy is dynamically adjusted to adapt to the changing user needs and market environment. By monitoring the degree of loss of user's desire to buy goods of interest, it is convenient to take measures before the user loses interest. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 It is a schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0060] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1

[0062] See also Figure 1 The first embodiment of the present invention provides a customer behavior prediction and automatic response system based on artificial intelligence, including:

[0063] Data collection module: used to collect customer behavior data through multiple channels authorized by customers, including user registration information, account activities, browsing history, purchase history, shopping cart data and customer feedback;

[0064] Storage node establishment module: divide all the commodities on the platform into product categories, and set up user data storage nodes of interest for the corresponding product categories in several regions with the most merchant registered addresses in each product category;

[0065] Data allocation module: Analyzes the product categories that customers are interested in based on browsing history, purchase history and shopping cart data, and allocates customer behavior data to the user data storage nodes of interest corresponding to the product categories for storage;

[0066] Feature analysis module: Based on the customer behavior data of the interest user data storage node, analyze the customer behavior characteristics, including: visit duration, frequency of visit to interest products, dwell time on interest products, and purchase conversion rate;

[0067] Behavior prediction module: Based on the analyzed customer behavior characteristics, the LSTM deep learning technology is used to capture time series characteristics, build a customer behavior prediction model, and predict the customer's purchase intention and optimal recommendation frequency;

[0068] Real-time analysis engine: Analyzes real-time customer behavior data through stream processing technology, analyzes the degree of loss of users' purchasing desire for products of interest, and dynamically adjusts the recommendation strategy.

[0069] Specifically, in this embodiment, the data collection module collects user registration information, account activities, browsing history, purchase history, shopping cart data and customer feedback data through customer-authorized channels, collects basic information when the user registers, and ensures that the user agrees to the privacy policy.

[0070] Account activities are implemented through back-end logging or database triggers to record user account login, logout and other activities.

[0071] Browsing records are recorded by using page tracking technology to record the pages users visit, their stay time, and their click behavior.

[0072] Purchase records are integrated through the order management system, and when the user completes the purchase, the transaction details are recorded, including product ID, purchase time, amount and other information.

[0073] The shopping cart data is captured on the front end through JavaScript events and stored in the back end database by tracking the items that the user has added to the shopping cart and their status.

[0074] Customer Feedback Allow users to submit feedback on your product or service by providing a feedback form or rating system.

[0075] Through API or regular data export, extract historical behavior data authorized by customers from existing CRM systems, including account activities, purchase records and customer feedback.

[0076] All collected data should be stored in a secure database, ensuring compliance with relevant laws and regulations, using data encryption technology to protect sensitive information, and implementing access controls to prevent unauthorized access.

[0077] Clean the collected data to remove duplicates and invalid data. This step can be achieved through automated scripts.

[0078] Integrate data from different channels into a unified platform, for example, using ETL tools to aggregate data from multiple sources into a centralized database.

[0079] The storage node establishment module determines the classification standards of products based on business needs and market research. For example, classification by category includes: electronic products, electronic accessories, men's clothing, women's clothing, household items, etc.

[0080] Create a database table to store all product category information, including: product category ID, product name, product description, and rating category.

[0081] Extract the merchant's registered location from the merchant registration information, count the merchant addresses for each product category, and calculate the number of registered merchants in each region. This can be achieved through SQL queries or data analysis tools.

[0082] By establishing a storage node module, all products on the platform are reasonably divided into product categories, and interest user data storage nodes in the corresponding regions are set up in each product category, which can better analyze the behavioral data of interested customers and understand customer needs.

[0083] Set up a data allocation module to analyze the types of products that customers are interested in based on browsing history, purchase history and shopping cart data. By dividing all the goods on the platform into reasonable product categories and setting up interest user data storage nodes in the corresponding regions in each product category, customer behavior data is allocated to the interest user data storage nodes of the corresponding product categories for storage, which can better analyze the behavior data of interested customers and understand customer needs.

[0084] By analyzing customers' browsing history, purchase history, and shopping cart data, it is easy to build a more accurate user portrait, which helps to accurately identify customers' areas of interest, thereby providing customers with personalized product recommendations. By accurately identifying customers' areas of interest, customer behavior data is allocated to the interest user data storage nodes of the corresponding product categories for storage, which significantly improves the efficiency of subsequent data query and analysis. When specific product category analysis is required, relevant user and behavior data can also be quickly located to reduce query time. Monitor changes in customer behavior in real time and adjust recommendation strategies in a timely manner to adapt to changes in customer needs.

[0085] The behavior prediction module captures time series features through LSTM deep learning technology based on the analyzed customer behavior characteristics, builds a customer behavior prediction model, and predicts the customer's purchase intention and optimal recommendation frequency. The real-time analysis engine analyzes real-time customer behavior data through stream processing technology, analyzes the degree of loss of user desire to buy products of interest, and dynamically adjusts the recommendation strategy.

[0086] LSTM deep learning technology can effectively capture long-term dependencies in time series, allowing the model to more accurately identify customers' purchasing intentions. Based on the prediction of purchasing intentions, it is convenient to provide more personalized and relevant product recommendations to each customer. Through stream processing technology, the system can analyze customer behavior data in real time and promptly identify the degree of loss of users' purchasing desire for products of interest.

[0087] According to the real-time analysis results, the recommendation strategy is dynamically adjusted to adapt to the ever-changing user needs and market environment. By monitoring the loss of users' desire to buy products of interest, it is convenient to take measures before users lose interest.

[0088] Combining LSTM models with real-time analysis engines makes it easier to gain deep insights into consumer behaviors, preferences, and trends, providing data support for strategic decision-making. Based on customer behavior analysis, more accurate and efficient marketing activities can be implemented.

[0089] In an optional embodiment, the data distribution module analyzes the product categories that the customer is interested in based on the browsing history, purchase history and shopping cart data, and distributes the customer behavior data to the interest user data storage node corresponding to the product category for storage, including the following steps:

[0090] Distributed storage module: used to store browsing records, purchase records and shopping cart data of the latest detection time period in the customer behavior data, and divide the latest detection time period into several detection time intervals;

[0091] Count the product categories that appear most frequently in the browsing records, purchase records, and shopping cart data during the user detection time period and the most recent detection time interval, and use the corresponding product categories as the product categories that the user is most interested in during the user detection time period and the product categories that the user is most interested in during the most recent detection time interval;

[0092] According to the product categories that the users are most interested in during the analyzed detection time period and the product categories that they are most interested in during the most recent detection time interval, customers are grouped into corresponding product category target groups;

[0093] The behavior data of customers in the corresponding product category target group is stored in the interest user data storage node of the corresponding product category.

[0094] Furthermore, the feature analysis module analyzes the customer behavior features according to the customer behavior data of the interest user data storage node, including the following steps:

[0095] Account activity, browsing history, purchase history, shopping cart data and customer feedback of customer behavior data stored in interest user data nodes;

[0096] Analyzed customer behavior characteristics, including:

[0097] Count the total duration of user visits during the most recent detection time interval;

[0098] According to the user's browsing history, purchase history and time data in the shopping cart data within the most recent detection time interval, the user's daily visit frequency and the time spent on the products of interest are counted;

[0099] Count the number of times a user browses a product and the corresponding number of times a user purchases it within the most recent detection time interval, and use the ratio of the number of times a user browses a product to the corresponding number of times a user purchases it as the user's purchase conversion rate;

[0100] The obtained data on visit duration, frequency of visits to products of interest, dwell time on products of interest and purchase conversion rate are used to form a customer behavior feature matrix.

[0101] In an optional embodiment, the behavior prediction module captures time series features through LSTM deep learning technology based on the analyzed customer behavior characteristics, builds a customer behavior prediction model, and predicts the customer's purchase intention and optimal recommendation frequency, including the following steps:

[0102] Obtain the latest customer behavior data within the detection time period and subdivide the detection time interval into several test time intervals;

[0103] According to the customer behavior data of the customers in the test time period, the visit duration, the visit frequency of the products of interest, the stay time of the products of interest and the purchase conversion rate in the test time period are counted to build a customer behavior feature matrix;

[0104] Count the customer's purchase records and customer feedback data within each test time interval, count the customer's purchase frequency, and whether the feedback recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate;

[0105] Determine the test time interval as the time step and convert the feature matrix into a format suitable for LSTM input;

[0106] By taking the customer behavior feature matrix of the previous time step as input, the number of purchases of the customer in the next time step, and the feedback recommendation frequency greater than demand, less than demand, or appropriate recommendation frequency as output labels, the LSTM network model is trained to build a customer behavior prediction model.

[0107] Based on the customer behavior characteristics analyzed in real time by the feature analysis module, the customer behavior prediction model is constructed to predict the customer's purchase frequency, whether the recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate, and analyze the customer's purchase intention and the optimal recommendation frequency.

[0108] In an optional embodiment, by constructing a customer behavior prediction model, predicting the number of customer purchases, and whether the recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate, analyzing the customer's purchase intention and the optimal recommendation frequency, including the following steps:

[0109] Statistics are constructed through the customer behavior prediction model to predict the number of customer purchases, and the customer's purchase intention analysis value is calculated using the following formula:

[0110]

[0111] Among them, Pn is the customer's purchase intention analysis value, p is the predicted number of purchases by the customer, and Pt is the frequency of visits to the product of interest;

[0112] If the recommended frequency is greater than the demand, increase the current recommended frequency by 5%. If it is less than the demand, decrease the current recommended frequency by 5%. If the recommended frequency is appropriate, maintain the current recommended frequency.

[0113] Example 2

[0114] Based on Example 1, the real-time analysis engine analyzes the real-time customer behavior data through stream processing technology, analyzes the degree of loss of users' desire to buy the products of interest, and dynamically adjusts the recommendation strategy, including the following steps:

[0115] Process customer behavior data through stream processing technology, count the real-time access duration, access frequency of products of interest, dwell time on products of interest, and purchase conversion rate of user data in each test time interval during the detection period, and draw the corresponding data waveform graph;

[0116] In the waveform chart, the time periods when the visit duration drops by more than 30%, the visit frequency of the interested products drops by more than 50%, and the stay time of the interested products drops by more than 50% are marked as the time periods of possible loss of purchase intention, and the time periods when the purchase conversion rate drops by more than 50% are marked as the time periods of loss of purchase intention;

[0117] Take the data waveform of a detection time interval before the purchase intention loss period as input, add the purchase intention loss label as output, take the data waveform of a detection time interval before the purchase intention loss period as input, add the purchase intention loss label as output, and train the deep learning model;

[0118] Process customer behavior data through stream processing technology, count the real-time access duration, access frequency, stay time and purchase conversion rate of user data in each test time interval in the most recent detection time interval, and draw the corresponding data waveform;

[0119] Input the data waveform of the most recent detection time interval into the trained deep learning model, and output a label of possible purchase intention loss, a label of purchase intention loss, or no label;

[0120] According to the output tag type, analyze the degree of loss of users' purchasing desire for the products of interest and dynamically adjust the recommendation strategy.

[0121] Furthermore, according to the output tag type, the user's purchase desire loss for the product of interest is analyzed, and the recommendation strategy is dynamically adjusted, including the following steps:

[0122] According to the output tag type and the customer's purchase intention analysis results, obtain the customer's shopping cart data in each test time interval within the detection time interval, including: shopping cart update frequency and shopping cart item browsing time;

[0123] The following formula is used to analyze the degree of loss of users' purchasing desire for products of interest:

[0124]

[0125] Among them, Pwa is the analysis result of the user's desire to buy the products of interest, Tci is the browsing time of the products in the shopping cart in the i-th test time interval, and Pci is the shopping cart update frequency in the i-th test time interval;

[0126] Dynamically adjust the recommendation strategy based on the analysis results of the user's loss of purchasing desire for the products of interest.

[0127] In an optional embodiment, dynamically adjusting the recommendation strategy according to the analysis result of the loss of the user's desire to purchase the product of interest includes the following steps:

[0128] Compare the data of the most recent test time interval with the data of the previous test time interval within the detection time interval;

[0129] If the analysis result of the user's desire to buy the product of interest increases by more than 5%, and the customer's purchase intention analysis value decreases by more than 10%, the recommendation frequency of the customer's product of interest will be reduced by 5%, and the recommendation frequency of related products will be increased by 3%;

[0130] If the analysis result of the user's purchase intention loss for the product of interest increases by more than 5%, and the customer's purchase intention analysis value does not decrease by more than 10%, the recommendation frequency of the customer's product of interest will increase by 5%, and at the same time, the recommendation frequency of related products will decrease by 5%;

[0131] If the user's purchase intention loss analysis result for the product of interest does not increase by more than 5%, and the customer's purchase intention analysis value decreases by more than 10%, maintain the recommendation frequency of the customer's product of interest, and increase the recommendation frequency of related products by 3%;

[0132] If the analysis result of the user's purchase intention loss for the interested products does not increase by more than 5%, and the customer's purchase intention analysis value does not decrease by more than 10%, the recommendation frequency of the customer's interested products is maintained, and at the same time, the recommendation frequency of related products is maintained.

[0133] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. Customer behavior prediction and automated response system based on artificial intelligence, characterized by: include: Data collection module: used to collect customer behavior data through multiple channels authorized by customers, including user registration information, account activities, browsing history, purchase history, shopping cart data and customer feedback; Storage node establishment module: divide all the commodities on the platform into product categories, and set up user data storage nodes of interest for the corresponding product categories in several regions with the most merchant registered addresses in each product category; Data allocation module: Analyzes the product categories that customers are interested in based on browsing history, purchase history and shopping cart data, and allocates customer behavior data to the user data storage nodes of interest corresponding to the product categories for storage; Feature analysis module: Based on the customer behavior data of the interest user data storage node, analyze the customer behavior characteristics, including: visit duration, frequency of visit to interest products, dwell time on interest products, and purchase conversion rate; Behavior prediction module: Based on the analyzed customer behavior characteristics, the LSTM deep learning technology is used to capture time series characteristics, build a customer behavior prediction model, and predict the customer's purchase intention and optimal recommendation frequency; Real-time analysis engine: Analyzes real-time customer behavior data through stream processing technology, analyzes the degree of loss of users' purchasing desire for products of interest, and dynamically adjusts the recommendation strategy.

2. The artificial intelligence-based customer behavior prediction and automated response system according to claim 1, characterized in that: The data distribution module analyzes the product categories that customers are interested in based on browsing records, purchase records and shopping cart data, and distributes customer behavior data to the interested user data storage nodes of the corresponding product categories for storage, including the following steps: Distributed storage module: used to store browsing records, purchase records and shopping cart data of the latest detection time period in the customer behavior data, and divide the latest detection time period into several detection time intervals; Count the product categories that appear most frequently in the browsing records, purchase records, and shopping cart data during the user detection time period and the most recent detection time interval, and use the corresponding product categories as the product categories that the user is most interested in during the user detection time period and the product categories that the user is most interested in during the most recent detection time interval; According to the product categories that the users are most interested in during the analyzed detection time period and the product categories that they are most interested in during the most recent detection time interval, customers are grouped into corresponding product category target groups; The behavior data of customers in the corresponding product category target group is stored in the interest user data storage node of the corresponding product category.

3. The artificial intelligence-based customer behavior prediction and automated response system according to claim 1, characterized in that: The feature analysis module analyzes the customer behavior features according to the customer behavior data of the interest user data storage node, including the following steps: Account activity, browsing history, purchase history, shopping cart data and customer feedback of customer behavior data stored in interest user data nodes; Analyzed customer behavior characteristics, including: Count the total duration of user visits during the most recent detection time interval; According to the user's browsing history, purchase history and time data in the shopping cart data within the most recent detection time interval, the user's daily visit frequency and the time spent on the products of interest are counted; Count the number of times a user browses a product and the corresponding number of times a user purchases it within the most recent detection time interval, and use the ratio of the number of times a user browses a product to the corresponding number of times a user purchases it as the user's purchase conversion rate; The obtained data on visit duration, frequency of visits to products of interest, dwell time on products of interest and purchase conversion rate are used to form a customer behavior feature matrix.

4. The artificial intelligence-based customer behavior prediction and automated response system according to claim 1, characterized in that: The behavior prediction module captures time series features through LSTM deep learning technology based on the analyzed customer behavior characteristics, builds a customer behavior prediction model, and predicts the customer's purchase intention and optimal recommendation frequency, including the following steps: Obtain the latest customer behavior data within the detection time period and subdivide the detection time interval into several test time intervals; According to the customer behavior data of the customers in the test time period, the visit duration, the visit frequency of the products of interest, the stay time of the products of interest and the purchase conversion rate in the test time period are counted to build a customer behavior feature matrix; Count the customer's purchase records and customer feedback data within each test time interval, count the customer's purchase frequency, and whether the feedback recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate; Determine the test time interval as the time step and convert the feature matrix into a format suitable for LSTM input; By taking the customer behavior feature matrix of the previous time step as input, the number of purchases of the customer in the next time step, and the feedback recommendation frequency greater than demand, less than demand, or appropriate recommendation frequency as output labels, the LSTM network model is trained to build a customer behavior prediction model. Based on the customer behavior characteristics analyzed in real time by the feature analysis module, the customer behavior prediction model is constructed to predict the customer's purchase frequency, whether the recommendation frequency is greater than the demand, less than the demand, or the recommendation frequency is appropriate, and analyze the customer's purchase intention and the optimal recommendation frequency.

5. The artificial intelligence-based customer behavior prediction and automated response system according to claim 4, characterized in that: The customer behavior prediction model is constructed to predict the number of customer purchases, whether the recommendation frequency is greater than demand, less than demand, or the recommendation frequency is appropriate, and analyze the customer's purchase intention and the optimal recommendation frequency, including the following steps: Statistics are constructed through the customer behavior prediction model to predict the number of customer purchases, and the customer's purchase intention analysis value is calculated using the following formula: Among them, Pn is the customer's purchase intention analysis value, p is the predicted number of purchases by the customer, and Pt is the frequency of visits to the product of interest; If the recommended frequency is greater than the demand, increase the current recommended frequency by 5%. If it is less than the demand, decrease the current recommended frequency by 5%. If the recommended frequency is appropriate, maintain the current recommended frequency.

6. The artificial intelligence-based customer behavior prediction and automated response system according to claim 1, characterized in that: The real-time analysis engine analyzes real-time customer behavior data through stream processing technology, analyzes the degree of loss of users' desire to purchase products of interest, and dynamically adjusts the recommendation strategy, including the following steps: Process customer behavior data through stream processing technology, count the real-time access duration, access frequency of products of interest, dwell time on products of interest, and purchase conversion rate of user data in each test time interval during the detection period, and draw the corresponding data waveform graph; In the waveform chart, the time periods when the visit duration drops by more than 30%, the visit frequency of the interested products drops by more than 50%, and the stay time of the interested products drops by more than 50% are marked as the time periods of possible loss of purchase intention, and the time periods when the purchase conversion rate drops by more than 50% are marked as the time periods of loss of purchase intention; Take the data waveform of a detection time interval before the purchase intention loss period as input, add the purchase intention loss label as output, take the data waveform of a detection time interval before the purchase intention loss period as input, add the purchase intention loss label as output, and train the deep learning model; Process customer behavior data through stream processing technology, count the real-time access duration, access frequency, stay time and purchase conversion rate of user data in each test time interval in the most recent detection time interval, and draw the corresponding data waveform; Input the data waveform of the most recent detection time interval into the trained deep learning model, and output a label of possible purchase intention loss, a label of purchase intention loss, or no label; According to the output tag type, analyze the degree of loss of users' purchasing desire for the products of interest and dynamically adjust the recommendation strategy.

7. The artificial intelligence-based customer behavior prediction and automated response system according to claim 6, characterized in that: According to the output tag type, analyze the user's purchase desire loss for the product of interest and dynamically adjust the recommendation strategy, including the following steps: According to the output tag type and the customer's purchase intention analysis results, obtain the customer's shopping cart data in each test time interval within the detection time interval, including: shopping cart update frequency and shopping cart item browsing time; The following formula is used to analyze the degree of loss of users' purchasing desire for products of interest: Among them, Pwa is the analysis result of the user's desire to buy the products of interest, Tci is the browsing time of the products in the shopping cart in the i-th test time interval, and Pci is the shopping cart update frequency in the i-th test time interval; Dynamically adjust the recommendation strategy based on the analysis results of the user's loss of purchasing desire for the products of interest.

8. The artificial intelligence-based customer behavior prediction and automated response system according to claim 7, characterized in that: According to the analysis results of the loss of users' purchasing desire for the products of interest, the recommendation strategy is dynamically adjusted, including the following steps: Compare the data of the most recent test time interval with the data of the previous test time interval within the detection time interval; If the analysis result of the user's desire to buy the product of interest increases by more than 5%, and the customer's purchase intention analysis value decreases by more than 10%, the recommendation frequency of the customer's product of interest will be reduced by 5%, and the recommendation frequency of related products will be increased by 3%; If the analysis result of the user's purchase intention loss for the product of interest increases by more than 5%, and the customer's purchase intention analysis value does not decrease by more than 10%, the recommendation frequency of the customer's product of interest will increase by 5%, and at the same time, the recommendation frequency of related products will decrease by 5%; If the user's purchase intention loss analysis result for the product of interest does not increase by more than 5%, and the customer's purchase intention analysis value decreases by more than 10%, maintain the recommendation frequency of the customer's product of interest, and increase the recommendation frequency of related products by 3%; If the analysis result of the user's purchase intention loss for the interested products does not increase by more than 5%, and the customer's purchase intention analysis value does not decrease by more than 10%, the recommendation frequency of the customer's interested products is maintained, and at the same time, the recommendation frequency of related products is maintained.

Images