A customer behavior prediction and recommendation system based on AI big model

Abstract

The present invention relates to the field of data processing technology, and in particular to a customer behavior prediction and recommendation system based on an AI large model, comprising: a data collection module for collecting customer behavior data in an e-commerce group purchase platform and storing it in a database; a data processing module, comprising a feature extraction unit for extracting features of optimization data output by preprocessing the customer behavior data; a model training module, comprising a model training unit connected to the feature extraction unit for training a deep learning model; a prediction and recommendation module, comprising a commodity recommendation unit for recommending commodities based on the prediction results output by predicting the customer's behavior; a storage module for storing the customer behavior data and the optimization data respectively; and a control module for determining the storage capacity of a database connection pool based on the proportion of lost data in the customer behavior data. The present invention improves the accuracy of customer behavior recommendation.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a customer behavior prediction and recommendation system based on an AI big model. Background Art

[0002] In the prior art, with the popularization of the Internet, the advantages of online shopping have become more prominent and it has become an increasingly important form of shopping. At present, people's pace of life is getting faster and faster, and traditional online shopping methods have caused obstacles for people to quickly obtain information. If we want to overcome the shortcomings of low efficiency, poor timeliness and difficulty in operation of information utilization, we need to study how to quickly and effectively obtain and predict users' interest preferences and adopt efficient and convenient push technology to present content resources that users are interested in to users in a timely manner, thereby improving user experience.

[0003] Chinese Patent Publication No.: CN113421146A discloses a method and system for predicting and recommending user online shopping behavior, which includes at least the following steps: S1. Collecting user information: Through data mining, collect the user's social behavior and shopping behavior to obtain and store the user's interest preference characteristics; S2. Analyze the user's interest preference characteristics, extract feature vectors from the collected user information, obtain user attribute characteristics and pre-process the user attribute characteristics, and filter out the user characteristics required for online shopping behavior prediction and recommendation; S3. Collect product feature labels, classify and store the products according to their different feature labels; S4. Take the products purchased by users according to their interest preference characteristics as the parent event , taking the commodity feature label as the sub-event, analyzing the relationship between the parent event and the sub-event; S5. According to the relationship between the parent event and the sub-event, construct a causal structure diagram, obtain the rules based on the causal relationship, and complete the mining of the causal relationship; S6. Combined with the output of steps S2 to S5, analyze the user's social behavior and shopping behavior, select features according to the causal relationship, and construct a user online shopping behavior prediction and recommendation model; S7. According to the user online shopping behavior prediction and recommendation model, predict the user's online shopping behavior and generate recommendation information; S8. The recommendation information is returned to the user end, and the user end provides data to the shopping homepage according to the recommendation information for presenting the user's shopping recommendation content. It can be seen that the user online shopping behavior prediction and recommendation method and system have the problems of cumbersome steps, complex operation and long time consumption. Summary of the invention

[0004] To this end, the present invention provides a customer behavior prediction and recommendation system based on an AI large model to overcome the problem in the prior art of reduced accuracy in customer behavior recommendations due to network delays, overfitting of model training, and inaccurate algorithm recommendations.

[0005] To achieve the above-mentioned purpose, the present invention provides a customer behavior prediction and recommendation system based on AI big model, including: a data acquisition module, used to collect customer behavior data in the e-commerce group buying platform and store it in a database; a data processing module, which is connected to the data acquisition module and includes a feature extraction unit for extracting features of optimized data output by preprocessing the customer behavior data; a model training module, which is connected to the data processing module and includes a model training unit connected to the feature extraction unit for training a deep learning model; a prediction and recommendation module, which is connected to the model training module and includes a product recommendation unit for recommending products according to the prediction results output by predicting the customer's behavior; a storage module, which is used to analyze the customer behavior data; and a prediction and recommendation module for recommending products according to the prediction results output by predicting the customer's behavior. The control module is respectively connected to the data acquisition module, the data processing module, the model training module and the prediction and recommendation module, and is used to store the customer behavior data, the optimization data, the characteristics of the optimization data, the deep learning model, the prediction results and the product recommendation results respectively; the control module is respectively connected to the data acquisition module, the data processing module, the model training module, the prediction and recommendation module and the storage module, and is used to determine the storage capacity of the database connection pool according to the proportion of lost data of the customer behavior data, or to determine the learning rate of the deep learning model training according to the average delay time of updating different product recommendations with the same data volume, and to determine the similarity threshold between products according to the accuracy of the product recommendations.

[0006] Furthermore, the data processing module also includes a preprocessing unit for preprocessing the customer behavior data to output the optimized data.

[0007] Furthermore, the model training module also includes a model updating unit connected to the model training unit for updating the deep learning model.

[0008] Furthermore, the prediction and recommendation module also includes a behavior prediction unit connected to the model training unit for predicting the customer's behavior to output a prediction result.

[0009] Furthermore, the control module is used to determine the recommendation accuracy of the customer behavior according to the proportion of lost data of the customer behavior data. If the proportion of lost data of the customer behavior data is greater than a preset first proportion, it is determined that the recommendation accuracy of the customer behavior does not meet the requirements.

[0010] Furthermore, the control module is used to preliminarily determine that the prediction accuracy of the customer behavior data does not meet the requirements when the proportion of lost data in the customer behavior data is greater than the preset first proportion and less than or equal to a preset second proportion, and to determine whether the prediction accuracy of the customer behavior data meets the requirements based on the average delay time for updating different product recommendations with the same data volume.

[0011] Furthermore, the control module is used to reduce the storage capacity of the database connection pool when the proportion of lost data of the customer behavior data is greater than the second proportion;

[0012] The reduction extent of the storage capacity of the database connection pool is determined by the difference between the proportion of lost data of the customer behavior data and a preset second proportion.

[0013] Further, the control module is used to determine the prediction accuracy of the customer behavior data according to the average delay time of updating the recommendation of different products with the same data volume, and if the average delay time of updating the recommendation of different products with the same data volume is greater than the preset first delay time, it is determined that the prediction accuracy of the customer behavior data does not meet the requirements;

[0014] The control module is used to preliminarily determine that the recommendation accuracy of the product does not meet the requirements when the average delay time for updating the recommendations of different products with the same data volume is greater than a preset second delay time, and determine whether the recommendation accuracy of the product meets the requirements based on the accuracy of the product recommendations.

[0015] Furthermore, the control module is used to reduce the learning rate of deep learning model training when the average delay time of updating different commodity recommendations with the same data volume is greater than the preset first delay time and less than or equal to the preset second delay time;

[0016] Among them, the reduction range of the learning rate of the deep learning model training is determined by the difference between the average delay time of updating different product recommendations with the same data volume and the preset first delay time.

[0017] Furthermore, the control module is used to determine the recommendation accuracy of the product according to the accuracy of the product recommendation. If the accuracy of the product recommendation is less than a preset accuracy, it is determined that the recommendation accuracy of the product does not meet the requirements, and the similarity threshold between the products is reduced;

[0018] The reduction range of the similarity threshold between the commodities is determined by the difference between the preset accuracy and the accuracy of the commodity recommendation.

[0019] Compared with the prior art, the beneficial effect of the present invention lies in that the system of the present invention determines the storage capacity of the database connection pool according to the proportion of lost data of customer behavior data by setting a data acquisition module, a data processing module, a model training module, a prediction and recommendation module, a storage module and a control module. Since the overall network traffic will increase significantly during the peak period of online shopping, a large number of users will conduct transactions and browse products at the same time, resulting in network congestion, which will cause data loss during transmission. By reducing the storage capacity of the database connection pool, the burden on the database server can be reduced, so that it can more effectively process important transaction data related operations, thereby reducing the risk of data loss. The learning rate of deep learning model training is determined according to the average delay time of recommendation updates of different products with the same data volume. Due to the different popularity of different products, some products are purchased many times and the amount of data generated is also large, while the number of purchases is large. The amount of data for a small number of products is relatively small, resulting in uneven data distribution, which makes the model overly focused on the customer behavior patterns of popular products and overfitting. By reducing the learning rate of deep learning model training, the model can have more time to balance the learning of different product data during training, which helps the model to learn popular product data while taking into account unpopular product data, so as to better learn the overall data pattern. The similarity threshold between products is determined according to the accuracy of product recommendations. Due to the lack of sufficient analysis of the characteristics of customer behavior data, the setting of recommendation algorithm parameters deviates from the actual situation, resulting in the algorithm finding too few similar users or similar products, resulting in a decrease in the accuracy of product recommendations. By reducing the similarity threshold between products, the range of similar products can be expanded, thereby increasing the range of products that can be selected by the recommendation system, improving the diversity of recommended products, and improving the accuracy of customer behavior recommendations.

[0020] Furthermore, the system of the present invention determines the storage capacity of the database connection pool by setting a preset first ratio and a preset second ratio. Since the overall network traffic will increase significantly during the peak period of online shopping, a large number of users will conduct transactions and browse products at the same time, causing network congestion, thereby causing data loss during transmission. By reducing the storage capacity of the database connection pool, the burden on the database server can be reduced, enabling it to more effectively process important transaction data-related operations, thereby reducing the risk of data loss and further improving the accuracy of customer behavior recommendations.

[0021] Furthermore, the system of the present invention determines the learning rate of deep learning model training by setting a preset first delay time and a preset second delay time. Due to the different popularity of different commodities, some commodities are purchased many times and generate a large amount of data, while the amount of data for commodities purchased less times is relatively small, resulting in uneven data distribution, so that the model pays too much attention to the customer behavior patterns of popular commodities and overfitting occurs. By reducing the learning rate of deep learning model training, the model can have more time to balance the learning of different commodity data during training, which helps the model to take into account the data of unpopular commodities while learning the data of popular commodities, so as to better learn the overall data pattern and further improve the accuracy of customer behavior recommendation.

[0022] Furthermore, the system of the present invention determines the similarity threshold between products by setting a preset accuracy rate. Since the characteristics of customer behavior data are not fully analyzed, the setting of recommendation algorithm parameters deviates from the actual situation, resulting in the algorithm finding too few similar users or similar products, thereby causing the accuracy of product recommendations to decrease. By reducing the similarity threshold between products, the range of similar products can be expanded, thereby increasing the range of products that can be selected by the recommendation system, improving the diversity of recommended products, and further improving the accuracy of customer behavior recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is an overall structural block diagram of a customer behavior prediction and recommendation system based on an AI big model according to an embodiment of the present invention;

[0024] Figure 2 This is a logic flow chart of a customer behavior prediction and recommendation system based on an AI big model according to an embodiment of the present invention;

[0025] Figure 3 This is a specific structural block diagram of a data processing module of a customer behavior prediction and recommendation system based on an AI big model according to an embodiment of the present invention;

[0026] Figure 4 This is a specific structural block diagram of the model training module of the customer behavior prediction and recommendation system based on the AI ​​big model in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; 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.

[0028] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0029] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0030] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0031] See also Figure 1 , Figure 2 , Figure 3 as well as Figure 4 As shown, they are respectively the overall structural block diagram, the logic flow chart, the specific structural block diagram of the data processing module and the specific structural block diagram of the model training module of the customer behavior prediction and recommendation system based on the AI ​​big model of the embodiment of the present invention. The present invention provides a customer behavior prediction and recommendation system based on the AI ​​big model, comprising:

[0032] The data collection module is used to collect customer behavior data on the e-commerce group buying platform and store it in the database;

[0033] A data processing module, which is connected to the data acquisition module and includes a feature extraction unit for extracting features of optimized data outputted by preprocessing the customer behavior data;

[0034] A model training module, which is connected to the data processing module, and includes a model training unit connected to the feature extraction unit for training a deep learning model;

[0035] A prediction and recommendation module, which is connected to the model training module and includes a product recommendation unit for recommending products based on the prediction results output by predicting the customer's behavior;

[0036] A storage module, which is respectively connected to the data acquisition module, the data processing module, the model training module and the prediction and recommendation module, and is used to store the customer behavior data, the optimization data, the features of the optimization data, the deep learning model, the prediction results and the product recommendation results respectively;

[0037] A control module, which is respectively connected to the data acquisition module, the data processing module, the model training module, the prediction and recommendation module and the storage module, is used to determine the storage capacity of the database connection pool according to the proportion of lost data in the customer behavior data, or to determine the learning rate of the deep learning model training according to the average delay time of updating different product recommendations with the same data volume, and to determine the similarity threshold between products according to the accuracy of product recommendations.

[0038] Specifically, customer behavior data includes the length of time customers spend browsing product pages, the names of products customers purchase, and the keywords customers use when searching for products.

[0039] Specifically, preprocessing includes cleaning customer behavior data, processing missing values ​​and outliers, and data conversion operations.

[0040] Specifically, the optimized data includes the purchase time of goods in a unified format, click data after removing customer errors, and keywords after filtering out spelling errors in search behavior.

[0041] Specifically, the characteristics of the optimized data include customers' browsing activity, customers' purchasing activity, and customers' search activity.

[0042] Specifically, the deep learning models are convolutional neural networks, recurrent neural networks, and Transformer architecture models.

[0043] Specifically, the prediction results include predicting the possibility of purchase, predicting the pages to be browsed, and predicting the search keywords.

[0044] Specifically, the product recommendation results include recommending products with high purchase probability, recommending products with a certain amount, and recommending possible browsing pages.

[0045] Specifically, the storage capacity of the database connection pool is the maximum number of database connections that can be accommodated in the connection pool at the same time.

[0046] Specifically, the learning rate of deep learning model training is a hyperparameter used to control the step size of model parameter updates, which determines the magnitude of parameter adjustments by the model according to the gradient of the loss function in each iteration.

[0047] Specifically, the similarity threshold between products is the maximum value used to measure the similarity between products.

[0048] In implementation, the system of the present invention determines the storage capacity of the database connection pool according to the proportion of lost data in customer behavior data by setting a data acquisition module, a data processing module, a model training module, a prediction and recommendation module, a storage module and a control module. Since the overall network traffic will increase significantly during the peak period of online shopping, a large number of users will conduct transactions and browse products at the same time, resulting in network congestion, which will cause data loss during transmission. By reducing the storage capacity of the database connection pool, the burden on the database server can be reduced, enabling it to more effectively process important transaction data-related operations, thereby reducing the risk of data loss. The learning rate of deep learning model training is determined according to the average delay time of recommendation updates of different products with the same data volume. Due to the different popularity of different products, some products are purchased more times and generate more data, while the number of products purchased less times is less. The amount of data is relatively small, resulting in uneven data distribution, which makes the model overly focused on the customer behavior patterns of popular products and overfitting. By reducing the learning rate of deep learning model training, the model can have more time to balance the learning of different product data during training, which helps the model to take into account the data of unpopular products while learning the data of popular products, so as to better learn the overall data pattern. The similarity threshold between products is determined according to the accuracy of product recommendations. Due to the lack of sufficient analysis of the characteristics of customer behavior data, the setting of recommendation algorithm parameters deviates from the actual situation, resulting in the algorithm finding too few similar users or similar products, resulting in a decrease in the accuracy of product recommendations. By reducing the similarity threshold between products, the range of similar products can be expanded, thereby increasing the range of products that can be selected by the recommendation system, improving the diversity of recommended products, and improving the accuracy of customer behavior recommendations.

[0049] Specifically, the data processing module also includes a preprocessing unit for preprocessing the customer behavior data to output the optimized data.

[0050] Specifically, the model training module also includes a model updating unit connected to the model training unit for updating the deep learning model.

[0051] Specifically, the prediction and recommendation module also includes a behavior prediction unit connected to the model training unit for predicting the customer's behavior to output a prediction result.

[0052] Specifically, the control module is used to obtain the amount of lost data and the total amount of customer behavior data in several collection cycles, and calculate the proportion of lost data of customer behavior data. If the proportion of lost data of the customer behavior data is greater than a preset first proportion, it is determined that the accuracy of customer behavior recommendation does not meet the requirements.

[0053] Specifically, the control module is used to preliminarily determine that the prediction accuracy of the customer behavior data does not meet the requirements when the proportion of lost data in the customer behavior data is greater than the preset first proportion and less than or equal to the preset second proportion, and to determine whether the prediction accuracy of the customer behavior data meets the requirements based on the average delay time for updating different product recommendations with the same data volume.

[0054] It can be understood that the three intervals corresponding to the preset first delay duration and the preset second delay duration correspond to three situations respectively:

[0055] The first interval is when the percentage of lost customer behavior data is less than or equal to the preset first percentage, and the accuracy of the recommendation of the corresponding customer behavior meets the requirements;

[0056] The second interval is when the proportion of lost data of customer behavior data is greater than the preset first proportion and less than or equal to the preset second proportion. This corresponds to the different popularity of different products, which leads to the fact that some products are purchased many times and generate a large amount of data, while the data of products purchased less often is relatively small, resulting in an uneven data distribution.

[0057] The third interval is where the proportion of lost customer behavior data is greater than the second proportion. This corresponds to the fact that during the peak period of online shopping, the overall network traffic will increase significantly, and a large number of users will conduct transactions and browse products at the same time, causing network congestion, which will lead to data loss during transmission.

[0058] In practice, the preset first ratio is generally selected in the range of [0.02, 0.04], and the preset second ratio is generally selected in the range of [0.04, 0.06].

[0059] Preferably, the preferred embodiment of the preset first ratio is 0.03, and the preferred embodiment of the preset second ratio is 0.05.

[0060] Specifically, the proportion of lost data of customer behavior data is the ratio of the lost data of customer behavior data in several collection cycles to the total data volume of customer behavior data.

[0061] In implementation, the system of the present invention determines the accuracy of recommendations for customer behavior by setting a preset first ratio and a preset second ratio, thereby reducing the impact of decreased stability in recommendations for customer behavior due to inaccurate determination of the accuracy of recommendations for customer behavior, and further improving the accuracy of recommendations for customer behavior.

[0062] Specifically, the control module is used to reduce the storage capacity of the database connection pool when the proportion of lost data of the customer behavior data is greater than the second proportion;

[0063] The reduction extent of the storage capacity of the database connection pool is determined by the difference between the proportion of lost data of the customer behavior data and a preset second proportion.

[0064] Specifically, when the difference between the percentage of lost data of customer behavior data and the preset second percentage is within 0.05, the storage capacity of the database connection pool is reduced to 0.95 times of the original value; when the difference between the percentage of lost data of customer behavior data and the preset second percentage exceeds 0.05, the storage capacity of the database connection pool is reduced by 3 for every 0.02 exceeding the original value. For example, the percentage of lost data of customer behavior data and the preset second percentage is 0.09, the current storage capacity of the database connection pool is 80, and the reduced storage capacity of the database connection pool is 80×0.95-3×2=70.

[0065] In implementation, the system of the present invention determines the storage capacity of the database connection pool by setting a preset first ratio and a preset second ratio. Since the overall network traffic will increase significantly during the peak period of online shopping, a large number of users will conduct transactions and browse products at the same time, causing network congestion, thereby causing data loss during transmission. By reducing the storage capacity of the database connection pool, the burden on the database server can be reduced, enabling it to more effectively handle important transaction data-related operations, thereby reducing the risk of data loss and further improving the accuracy of customer behavior recommendations.

[0066] Specifically, the control module is used to obtain the delay time of the recommendation update of different commodities with the same data volume in several recommendation update cycles, and calculate the average delay time of the recommendation update of different commodities with the same data volume. If the average delay time of the recommendation update of different commodities with the same data volume is greater than the preset first delay time, it is determined that the prediction accuracy of the customer behavior data does not meet the requirements;

[0067] The control module is used to preliminarily determine that the recommendation accuracy of the product does not meet the requirements when the average delay time for updating the recommendations of different products with the same data volume is greater than a preset second delay time, and determine whether the recommendation accuracy of the product meets the requirements based on the accuracy of the product recommendations.

[0068] It can be understood that the three intervals corresponding to the preset first delay duration and the preset second delay duration correspond to three situations respectively:

[0069] The first interval is when the average delay time for updating recommendations of different products with the same amount of data is less than or equal to the preset first delay time, corresponding to the situation where the prediction accuracy of customer behavior data meets the requirements;

[0070] The second interval is the average delay time for updating the recommendation of different products with the same data volume, which is greater than the preset first delay time and less than or equal to the preset second delay time. This corresponds to the different popularity of different products, resulting in some products being purchased many times and generating a large amount of data, while the data volume of products purchased less frequently is relatively small, resulting in an uneven data distribution.

[0071] The third interval is when the average delay time for updating recommendations of different products with the same amount of data is longer than the preset second delay time. This corresponds to a lack of sufficient analysis of the characteristics of customer behavior data, which results in deviation from the actual situation when setting the recommendation algorithm parameters, causing the algorithm to find too few similar users or similar products.

[0072] In practice, the preset first delay time length is generally selected in the range of [3s, 5s], and the preset second delay time length is generally selected in the range of [5s, 7s].

[0073] Preferably, the preferred embodiment of the preset first delay time length is 4s, and the preferred embodiment of the preset second delay time length is 6s.

[0074] Specifically, the average delay time for updating recommendations of different commodities with the same data volume is the ratio of the total delay time for updating recommendations of different commodities with the same data volume in several recommendation update cycles to the number of recommendation update cycles.

[0075] In implementation, the system of the present invention determines the prediction accuracy of customer behavior data by setting a preset first delay period and a preset second delay period, thereby reducing the impact of a decrease in the accuracy of customer behavior recommendations due to inaccurate determination of the accuracy of customer behavior prediction data, and further improving the accuracy of customer behavior recommendations.

[0076] Specifically, the control module is used to reduce the learning rate of deep learning model training when the average delay time of updating different commodity recommendations with the same data volume is greater than the preset first delay time and less than or equal to the preset second delay time;

[0077] Among them, the reduction range of the learning rate of the deep learning model training is determined by the difference between the average delay time of updating different product recommendations with the same data volume and the preset first delay time.

[0078] Specifically, when the difference between the average delay time for updating different product recommendations with the same data volume and the preset first delay time is within 3s, the learning rate of the deep learning model training is reduced to 0.9 times the original; when the difference between the average delay time for updating different product recommendations with the same data volume and the preset first delay time exceeds 3s, the learning rate of the deep learning model training is reduced by 0.002 for every 2s exceeding it. For example, the difference between the average delay time for updating different product recommendations with the same data volume and the preset first delay time is 7s. The current learning rate of the deep learning model training is 0.009, and the reduced learning rate of the deep learning model training is 0.009×0.9-0.002×2=0.0041.

[0079] In implementation, the system of the present invention determines the learning rate of deep learning model training by setting a preset first delay time and a preset second delay time. Due to the different popularity of different commodities, some commodities are purchased many times and generate a large amount of data, while the amount of data for commodities purchased less times is relatively small, resulting in uneven data distribution, so that the model pays too much attention to the customer behavior patterns of popular commodities and overfitting occurs. By reducing the learning rate of deep learning model training, the model can have more time to balance the learning of different commodity data during training, which helps the model to take into account the data of unpopular commodities while learning the data of popular commodities, so as to better learn the overall data pattern and further improve the accuracy of customer behavior recommendation.

[0080] Specifically, the control module is used to obtain the number of products that meet the customer's needs and the total number of recommended products, and calculate the accuracy of the product recommendations. If the accuracy of the product recommendations is less than the preset accuracy, it is determined that the accuracy of the product recommendations does not meet the requirements, and the similarity threshold between the products is reduced.

[0081] It is understandable that the two intervals corresponding to the preset accuracy rate correspond to two situations:

[0082] The first interval is when the accuracy of product recommendations is lower than the preset accuracy. This corresponds to the situation where the characteristics of customer behavior data are not fully analyzed, resulting in deviation from the actual situation when setting the recommendation algorithm parameters, and the algorithm finds too few similar users or similar products.

[0083] The second interval is when the accuracy of product recommendations is greater than or equal to the preset accuracy, and the recommendation accuracy of the corresponding products meets the requirements.

[0084] In practice, the preset accuracy is generally selected in the range of [94%, 96%].

[0085] Preferably, the preferred embodiment of the preset accuracy is 95%.

[0086] Specifically, the accuracy of product recommendations is the ratio of the number of products that match the customer to the total number of recommended products.

[0087] In implementation, the system of the present invention determines the recommendation accuracy of goods by setting a preset accuracy rate, thereby reducing the impact of decreased recommendation accuracy on customer behavior due to inaccurate determination of the recommendation accuracy of goods, and further improving the recommendation accuracy on customer behavior.

[0088] Specifically, the reduction range of the similarity threshold between the commodities is determined by the difference between the preset accuracy and the accuracy of the commodity recommendation.

[0089] Specifically, when the difference between the preset accuracy and the accuracy of product recommendations is within 3%, the similarity threshold between products is reduced to 0.9 times the original value; when the difference between the preset accuracy and the accuracy of product recommendations exceeds 3%, the similarity threshold between products is reduced by 0.05 for every 1% increase. For example, the difference between the preset accuracy and the accuracy of product recommendations is 5%, the current similarity threshold between products is 0.8, and the reduced similarity threshold between products is 0.8×0.9-0.05×2=0.62.

[0090] In implementation, the system of the present invention determines the similarity threshold between products by setting a preset accuracy rate. Since the characteristics of customer behavior data are not fully analyzed, the recommendation algorithm parameters deviate from the actual situation, resulting in the algorithm finding too few similar users or similar products, thereby causing the accuracy of product recommendations to decrease. By reducing the similarity threshold between products, the range of similar products can be expanded, thereby increasing the range of products that can be selected by the recommendation system, improving the diversity of recommended products, and further improving the accuracy of customer behavior recommendations.

[0091] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A customer behavior prediction and recommendation system based on AI big model, characterized by: include: The data collection module is used to collect customer behavior data on the e-commerce group buying platform and store it in the database; A data processing module, which is connected to the data acquisition module and includes a feature extraction unit for extracting features of optimized data outputted by preprocessing the customer behavior data; A model training module, which is connected to the data processing module, and includes a model training unit connected to the feature extraction unit for training a deep learning model; A prediction and recommendation module, which is connected to the model training module and includes a product recommendation unit for recommending products based on the prediction results output by predicting the customer's behavior; A storage module, which is respectively connected to the data acquisition module, the data processing module, the model training module and the prediction and recommendation module, and is used to store the customer behavior data, the optimization data, the features of the optimization data, the deep learning model, the prediction results and the product recommendation results respectively; A control module, which is respectively connected to the data acquisition module, the data processing module, the model training module, the prediction and recommendation module and the storage module, and is used to determine the storage capacity of the database connection pool according to the proportion of lost data of the customer behavior data, or to determine the learning rate of the deep learning model training according to the average delay time of updating different product recommendations with the same data volume, and to determine the similarity threshold between products according to the accuracy of product recommendations; The control module is used to determine the recommendation accuracy of the customer behavior according to the percentage of the amount of lost data of the customer behavior data, and if the percentage of the amount of lost data of the customer behavior data is greater than a preset first percentage, it is determined that the recommendation accuracy of the customer behavior does not meet the requirements; The control module is used to preliminarily determine that the prediction accuracy of the customer behavior data does not meet the requirements when the proportion of the amount of lost data of the customer behavior data is greater than the preset first proportion and less than or equal to the preset second proportion, and determine whether the prediction accuracy of the customer behavior data meets the requirements based on the average delay time of updating different product recommendations with the same amount of data; The control module is used to reduce the storage capacity of the database connection pool when the proportion of lost data of the customer behavior data is greater than the second proportion; The reduction extent of the storage capacity of the database connection pool is determined by the difference between the proportion of lost data of the customer behavior data and a preset second proportion.

2. The customer behavior prediction and recommendation system based on AI big model according to claim 1 is characterized in that: The data processing module also includes a preprocessing unit for preprocessing the customer behavior data to output the optimized data.

3. The customer behavior prediction and recommendation system based on AI big model according to claim 1 is characterized in that: The model training module also includes a model updating unit connected to the model training unit for updating the deep learning model.

4. The customer behavior prediction and recommendation system based on AI big model according to claim 1 is characterized in that: The prediction and recommendation module also includes a behavior prediction unit connected to the model training unit for predicting the customer's behavior to output a prediction result.

5. The customer behavior prediction and recommendation system based on AI big model according to claim 4 is characterized in that: The control module is used to determine the prediction accuracy of the customer behavior data according to the average delay time of updating the recommendation of different products with the same data volume, and if the average delay time of updating the recommendation of different products with the same data volume is greater than the preset first delay time, it is determined that the prediction accuracy of the customer behavior data does not meet the requirements; The control module is used to preliminarily determine that the recommendation accuracy of the product does not meet the requirements when the average delay time for updating the recommendations of different products with the same data volume is greater than a preset second delay time, and determine whether the recommendation accuracy of the product meets the requirements based on the accuracy of the product recommendations.

6. The customer behavior prediction and recommendation system based on AI big model according to claim 5 is characterized in that: The control module is used to reduce the learning rate of deep learning model training when the average delay time of updating different commodity recommendations with the same data volume is greater than the preset first delay time and less than or equal to the preset second delay time; Among them, the reduction range of the learning rate of the deep learning model training is determined by the difference between the average delay time of updating different product recommendations with the same data volume and the preset first delay time.

7. The customer behavior prediction and recommendation system based on AI big model according to claim 6 is characterized in that: The control module is used to determine the recommendation accuracy of the product according to the accuracy of the product recommendation. If the accuracy of the product recommendation is less than the preset accuracy, it is determined that the recommendation accuracy of the product does not meet the requirements, and the similarity threshold between the products is reduced; The reduction range of the similarity threshold between the commodities is determined by the difference between the preset accuracy and the accuracy of the commodity recommendation.

Images