Enterprise customer data management system and management method

Abstract

The invention discloses a data management system and management method for enterprise clients, and relates to the technical field of intelligent advertisement putting, the system collects stay duration, inactive days, interaction times, advertisement conversion rate, preference vector and content label behavior data of clients through an embedded SDK and burying points, and performs cleaning, complementing and normalization processing on the data; modeling the customer portrait by using a convolutional neural network, and carrying out matching output on the advertisement content and the target customer; calculating a customer response potential coefficient KHX, comparing the customer response potential coefficient KHX with a preset threshold Q1, judging whether a loss risk exists or not, and triggering strategy adjustment; further calculating a comprehensive interest recall coefficient through an interest label drift coefficient, a similar content recall coefficient and a behavior response difference coefficient, comparing the comprehensive interest recall coefficient with a threshold Q2, and accurately identifying interest offset; behavior changes after customer strategies are executed are used for continuously training the optimization model, and closed-loop improvement of customer behavior and interest prediction is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent advertising delivery, and in particular to a data management system and a management method for enterprise customers. Background Art

[0002] With the development of digital marketing, companies are increasingly relying on a deep understanding of customer behavior and interests to deliver personalized recommendations and targeted advertising. Current customer data management systems on the market primarily focus on basic user attributes and static behavioral records. They lack the ability to identify and respond to dynamic changes in user interests in real time, making it difficult to effectively address the challenges of volatile customer behavior and frequent interest shifts. Furthermore, existing systems often rely on simple rules or shallow models to predict advertising response potential, failing to accurately reflect customers' true intent. This results in low ad conversion rates and increased customer churn.

[0003] Traditional data processing methods often rely on batch processing, and the cleaning, integration, and modeling of customer behavior data lag behind real-time behavioral changes, making it difficult to support frequent policy adjustments and immediate responses. Furthermore, in the customer profile modeling process, common feature extraction methods ignore high-order nonlinear correlations between behavioral data and lack the ability of deep learning models to automatically learn complex behavioral features.

[0004] In addition, when customers respond abnormally or conversion results are poor, existing systems are often unable to promptly identify the specific reasons for customer interest drift or abnormal behavior. They lack effective recall mechanisms and compensation strategies, resulting in content recommendations that do not match customers' actual interests, further reducing customer stickiness and platform activity. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention provides a data management system and a management method for enterprise customers to solve the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a data management system for enterprise customers, including a data acquisition module, a data processing module, a customer portrait analysis model establishment module, a behavior prediction analysis module, a customer interest recall module and a feedback optimization module;

[0007] The data collection module is used to collect customer behavior data and preference information through embedded SDK, tracking points and multi-source behavior tracking, including length of stay, number of inactive days, number of interactions, advertising conversion rate, interest preference vector, tag preference vector, advertising content tag and historical conversion records;

[0008] The data processing module is used to clean, complete, normalize and unify the format of the collected customer behavior data, interest vectors and advertising conversion information, covering the standardized organization of multiple types of information such as customer behavior indicators, preference vectors and conversion records;

[0009] The customer portrait analysis model building module is used to train and optimize based on convolutional neural networks, combined with customer behavior and preference data, to identify customer response potential and preference drift, and to match advertising content with target customers;

[0010] The behavior prediction analysis module is used to monitor the customer's behavior in the advertisement in real time, combine the multi-dimensional customer behavior data and preference feature data, calculate the customer response potential coefficient KHX, and compare and analyze it with the first threshold Q1 to determine whether the customer response degree is qualified. If not, a strategy is given;

[0011] The customer interest recall module is used to monitor the customer's recent behavior changes and interest tag deviations in real time when receiving the first warning instruction, and calculate the interest tag drift coefficient QPY of the i-th customer respectively. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , further calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , and compare and analyze it with the second threshold Q2 to determine whether the customer behavior is stable and whether the interest deviation is within the adjustment range. If it is unstable and does not fall within the adjustment range, a strategy is given;

[0012] The feedback optimization module is used to monitor the changes in customer behavior after executing the strategy, and feed them back into the customer portrait analysis model as training samples. It also combines historical and real-time data and uses convolutional neural networks to regularly update the training model to continuously optimize the accuracy of customer interest and behavior predictions.

[0013] Preferably, the data collection module is used to collect the average length of stay of customers through the embedded front-end SDK; collect the number of inactive days of customers through the log tracking system; collect the actual conversion rate of customers in each advertisement through the advertising delivery monitoring system; collect the number of times customers perform various interactive behaviors through the interactive behavior tracking module; collect the customer's preference feature vector for advertising content and the customer's interest preference vector through the advertising content adaptation record; collect the customer's recent customer label preference vector through the recent user behavior label extraction module; collect the customer's overall historical preference vector through the user historical behavior analysis system; collect the label vector of the customer's advertising content through the advertising content label management module; collect the content label vector of historical advertisements that the customer has received through the advertising delivery history database; and collect the customer's actual conversion value for historical advertisements through the conversion effect recording interface.

[0014] Preferably, the data processing module is used to uniformly process the collected customer behavior data, interest vectors and advertising conversion information, including data cleaning, missing value filling, normalization, label de-redundancy, vector dimension alignment and anomaly detection operations; the processed data include: the average length of stay of customers, the number of inactive days, the actual conversion rate in advertising delivery, the number of interactive behaviors, the interest preference vector, the advertising preference feature vector, the recent customer label preference vector, the overall historical preference vector, the advertising content label vector, the historical advertising content label vector and the actual conversion value of historical advertisements.

[0015] Preferably, the customer portrait analysis model establishment module is used to utilize a convolutional neural network to construct an initial model of a convolutional neural network, and to train and test the initial model of the convolutional neural network using customer conversion rate data and historical conversion rate data, and to use the trained initial model of the convolutional neural network as a customer portrait analysis model, while using the intermediate layer output of customer behavior data and customer preference data as a feature vector to identify customer response potential and preference drift characteristics, and to train and test the customer portrait analysis model through the obtained feature vector, and to output the trained customer portrait analysis model as a match between advertising content and target customers.

[0016] Preferably, the behavior prediction and analysis module includes a first calculation unit and a first analysis unit;

[0017] The first calculation unit is used to monitor the customer's behavior in the advertisement in real time, and calculate the customer response potential coefficient KHX after dimensionless processing by combining multi-dimensional customer behavior data and preference feature data. The formula is as follows:

[0018]

[0019] Where, Iw represents the weighted average conversion rate of the i-th customer in several advertisements, i represents the i-th customer interaction index, T i represents the average length of stay of the i-th customer, δ i represents the number of inactive days of the i-th customer, P i represents the preference fit index of the i-th customer, η i represents the matching degree of the content of the i-th customer, w1, w2, w3 and w4 represent weight coefficients;

[0020]

[0021] Where n represents the number of times the i-th customer inspects the advertisement, C i,k represents the actual conversion rate of the i-th customer in the k-th advertising campaign, e represents a natural constant, and λ represents a weight decay parameter obtained by model training;

[0022]

[0023] Where θ j Indicates the behavior weight coefficient of the jth interactive behavior type, obtained by model training, I i,j represents the number of times the i-th customer has performed the j-th interactive behavior type;

[0024]

[0025] Where V i represents the interest preference vector of the i-th customer, A i Represents the feature vector of the i-th customer's preference for the content of the advertisement.

[0026] Preferably, the first analysis unit is used to preset a first threshold Q1 in advance, and compare and analyze the customer response potential coefficient KHX with the first threshold Q1, and obtaining the first evaluation result includes:

[0027] When the customer response potential coefficient KHX ≥ the first threshold Q1, it means that the customer response is qualified and there is no risk of customer loss, so continuous monitoring is required;

[0028] When the customer response potential coefficient KHX is less than the first threshold Q1, it indicates that the customer response is unsatisfactory and there is a risk of customer loss, triggering the first warning instruction and generating the first strategy: starting the interest recall mechanism; adjusting the content category, time period and interaction form.

[0029] Preferably, the customer interest recall module includes a second calculation unit, a third calculation unit and a second analysis unit;

[0030] The second calculation unit is used to monitor the customer's recent behavior changes and interest tag deviation in real time when receiving the first warning instruction, and calculate the interest tag drift coefficient QPY of the i-th customer after dimensionless processing based on the recent tag vector, historical preference vector, content tag, conversion effect and interactive behavior data. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , the formula is as follows:

[0031]

[0032] Where Urec i Represents the recent customer label preference vector of the i-th customer, Uhist i represents the overall historical preference vector of the i-th customer;

[0033]

[0034] Where N represents the number of recalled contents, Aa i Represents the advertising content label vector of the i-th customer, Al i,m Represents the content label vector of the historical advertisement m of the i-th customer, h i,m represents the actual conversion value of the i-th customer for ad m;

[0035]

[0036] Where, Iavg i Represents the overall historical interaction index mean of the i-th customer, Iw i represents the i-th customer interaction index, and ε is a constant to prevent the denominator from being zero.

[0037] Preferably, the third calculation unit is used to calculate and obtain the interest tag drift coefficient QPY of the i-th customer. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , after dimensionless processing, calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , the formula is as follows:

[0038] ZXY i =a1*QPY i +a2*SZH i +a3*WCY i ;

[0039] Where a1, a2 and a3 represent weight coefficients;

[0040] The second analysis unit is used to preset a second threshold Q2 in advance and calculate the comprehensive interest recall coefficient ZXY of the i-th customer. i Comparing and analyzing with the second threshold Q2 to obtain a second evaluation result includes:

[0041] When the comprehensive interest recall coefficient of the i-th customer ZXY i When ≤ the second threshold Q2, it means that the behavior of the i-th customer is stable and the interest deviation is within the adjustment range. No adjustment is made and continuous monitoring is performed;

[0042] When the comprehensive interest recall coefficient of the i-th customer ZXY i When the value is greater than the second threshold Q2, it indicates that the behavior of the i-th customer is unstable and the interest deviation is beyond the adjustment range, triggering the second warning instruction and generating the second strategy: re-extract the potential interest tags of the i-th customer in the latest behavior data and increase the probability of pushing candidate content containing new interest tags to 40%.

[0043] Preferably, the feedback optimization module is used to use the changes in customer behavior after executing the strategy as training samples and feed them back into the customer portrait analysis model, so that the model can reflect the customer's new interests and preferences; combining historical data with real-time data, the customer portrait analysis model is regularly retrained through the convolutional neural network deep learning algorithm, and continuously optimized to enable the model to more accurately predict changes in customer behavior and interests.

[0044] Preferably, a data management method for enterprise customers includes the following steps:

[0045] Step 1: Collect customer behavior data and preference information through embedded SDKs, tracking points, and multi-source behavior tracking, including duration of stay, inactive days, number of interactions, ad conversion rate, interest preference vectors, tag preference vectors, ad content tags, and historical conversion records.

[0046] Step 2: Clean, complete, normalize, and format the collected customer behavior data, interest vectors, and ad conversion information. This includes standardized organization of multiple types of information, including customer behavior indicators, preference vectors, and conversion records.

[0047] Step 3: Based on a convolutional neural network, combined with customer behavior and preference data, it is trained and optimized to identify customer response potential and preference drift, and output matching advertising content with target customers;

[0048] Step 4: Monitor customer behavior in real time during the advertisement. Combine multi-dimensional customer behavior data with preference feature data to calculate the customer response potential coefficient KHX. Compare and analyze this coefficient with the first threshold Q1 to determine whether the customer response is qualified. If not, implement a strategy.

[0049] Step 5: When the first warning instruction is received, the customer's recent behavior changes and interest tag drift are monitored in real time, and the interest tag drift coefficient QPY of the i-th customer is calculated. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , further calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , and compare and analyze it with the second threshold Q2 to determine whether the customer behavior is stable and whether the interest deviation is within the adjustment range. If it is unstable and does not fall within the adjustment range, a strategy is given;

[0050] Step 6: Monitor changes in customer behavior after executing the strategy and use them as training samples to feed back into the customer profiling model. Combined with historical and real-time data, the training model is regularly updated using a convolutional neural network to continuously optimize the accuracy of customer interest and behavior predictions.

[0051] The present invention provides a data management system and management method for enterprise customers, which has the following beneficial effects:

[0052] This data management system and management method for corporate customers, based on a customer portrait analysis model established by a convolutional neural network, can accurately identify changes in customer interests and preference drift, ensure that advertising content is highly matched with customer needs, thereby improving advertising conversion rates and customer stickiness.

[0053] (2) This is a data management system and management method for corporate customers. It monitors and analyzes customer behavior in advertising in real time, and conducts dynamic evaluation based on the customer response potential coefficient and the set threshold. When the customer's response rate is lower than the standard, the system will automatically trigger the interest recall mechanism and adjust the strategy, thereby effectively reducing the risk of customer churn and improving the customer response rate.

[0054] (3) A data management system and management method for corporate customers. When the customer's interest deviates beyond the adjustment range, the system will automatically adjust the advertising push content and push strategy to ensure that the advertising content always meets the customer's latest interests and further optimize the effect of advertising delivery.

[0055] (4) This data management system and management method for corporate customers uses a feedback optimization module to feed back the behavioral changes of customers after they execute a strategy as training samples to the customer portrait analysis model, continuously optimizing the model so that the system can update customer interest and preference predictions in real time, improve the accuracy of customer behavior predictions, and enhance the system's adaptability to customer needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a system block diagram of a data management system for enterprise customers of the present invention;

[0057] Figure 2 The figure is a flow chart of a data management method for enterprise customers according to the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] Example 1

[0060] See also Figure 1 , the present invention provides a data management system and management method for enterprise customers, including a data acquisition module, a data processing module, a customer portrait analysis model establishment module, a behavior prediction analysis module, a customer interest recall module and a feedback optimization module;

[0061] The data collection module is used to collect customer behavior data and preference information through embedded SDK, tracking points and multi-source behavior tracking, including length of stay, number of inactive days, number of interactions, advertising conversion rate, interest preference vector, tag preference vector, advertising content tag and historical conversion records;

[0062] The data processing module is used to clean, complete, normalize and unify the format of the collected customer behavior data, interest vectors and advertising conversion information, covering the standardized organization of multiple types of information such as customer behavior indicators, preference vectors and conversion records;

[0063] The customer portrait analysis model building module is used to train and optimize based on convolutional neural networks, combined with customer behavior and preference data, to identify customer response potential and preference drift, and to match advertising content with target customers;

[0064] The behavior prediction analysis module is used to monitor the customer's behavior in the advertisement in real time, combine the multi-dimensional customer behavior data and preference feature data, calculate the customer response potential coefficient KHX, and compare and analyze it with the first threshold Q1 to determine whether the customer response degree is qualified. If not, a strategy is given;

[0065] The customer interest recall module is used to monitor the customer's recent behavior changes and interest tag deviations in real time when receiving the first warning instruction, and calculate the interest tag drift coefficient QPY of the i-th customer respectively. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i, further calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , and compare and analyze it with the second threshold Q2 to determine whether the customer behavior is stable and whether the interest deviation is within the adjustment range. If it is unstable and does not fall within the adjustment range, a strategy is given;

[0066] The feedback optimization module is used to monitor the changes in customer behavior after executing the strategy, and feed them back into the customer portrait analysis model as training samples. It also combines historical and real-time data and uses convolutional neural networks to regularly update the training model to continuously optimize the accuracy of customer interest and behavior predictions.

[0067] In this embodiment, by real-time monitoring and dynamically adjusting the customer interest recall mechanism, the system can accurately identify shifts in customer interests and behavioral changes, and adjust advertising content and delivery strategies based on actual customer needs. This not only improves ad conversion rates, but also effectively prevents customer churn and increases customer responsiveness to advertising content, thereby optimizing the effectiveness of corporate advertising and enhancing customer stickiness and long-term value.

[0068] Example 2

[0069] This embodiment is an explanation of the embodiment 1. Specifically, the data collection module is used to collect the average length of stay of customers through the embedded front-end SDK; collect the number of inactive days of customers through the log tracking system; collect the actual conversion rate of customers in each advertisement through the advertising delivery monitoring system; collect the number of times customers perform various types of interactive behaviors through the interactive behavior tracking module; collect the customer's preference feature vector for advertising content and the customer's interest preference vector through the advertising content adaptation record; collect the customer's recent customer label preference vector through the recent user behavior label extraction module; collect the customer's overall historical preference vector through the user historical behavior analysis system; collect the customer's advertising content label vector through the advertising content label management module; collect the content label vector of historical advertisements that the customer has received through the advertising delivery history database; and collect the customer's actual conversion value for historical advertisements through the conversion effect recording interface.

[0070] In this embodiment, through multi-channel data collection, the system can comprehensively and accurately collect customer behavior data, interest preferences, and advertising conversion results, forming a complete customer profile. The integration and analysis of this data helps to gain a deeper understanding of customer interest changes and behavior patterns, providing a more accurate basis for decision-making in advertising strategies, thereby significantly improving the accuracy and conversion rate of advertising, optimizing the customer experience, and enhancing advertising effectiveness.

[0071] Example 3

[0072] This embodiment is an explanation of Embodiment 2. Specifically, the data processing module is used to uniformly process the collected customer behavior data, interest vectors and advertising conversion information, including data cleaning, missing value filling, normalization, label de-redundancy, vector dimension alignment and anomaly detection operations; the processed data include: the average length of stay of customers, the number of inactive days, the actual conversion rate in advertising delivery, the number of interactive behaviors, the interest preference vector, the advertising preference feature vector, the recent customer label preference vector, the overall historical preference vector, the advertising content label vector, the historical advertising content label vector and the actual conversion value of historical advertisements.

[0073] In this embodiment, the system ensures high quality and consistency of customer behavior data through a unified data processing process, including cleaning, missing value filling, normalization, label deduplication, and anomaly detection. This standardized processing not only improves data availability and accuracy, but also eliminates noise and bias in the data, making subsequent analysis and model training more reliable, thereby improving the accuracy of customer behavior predictions and the effectiveness of improving advertising conversion rates.

[0074] Example 4

[0075] This embodiment is an explanation of Embodiment 3. Specifically, the customer portrait analysis model building module is used to utilize a convolutional neural network to construct an initial model of a convolutional neural network, and to train and test the initial model of the convolutional neural network using customer conversion rate data and historical conversion rate data. The trained initial model of the convolutional neural network is used as a customer portrait analysis model, and the intermediate layer output of customer behavior data and customer preference data is used as a feature vector to identify customer response potential and preference drift characteristics. The customer portrait analysis model is trained and tested through the obtained feature vector, and the trained customer portrait analysis model is output as advertising content matching with target customers.

[0076] In this embodiment, by building a customer profiling model using a convolutional neural network (CNN), the system can accurately identify customers' response potential and preference drift characteristics. By training the intermediate-layer feature vectors of customer behavior and preference data, the model can accurately model customer needs, effectively optimizing the matching of advertising content with target customers, and improving the accuracy and conversion rate of advertising delivery.

[0077] Example 5

[0078] This embodiment is explained in Example 4. Specifically, the behavior prediction and analysis module includes a first calculation unit and a first analysis unit;

[0079] The first calculation unit is used to monitor the customer's behavior in the advertisement in real time, and calculate the customer response potential coefficient KHX after dimensionless processing by combining multi-dimensional customer behavior data and preference feature data. The formula is as follows:

[0080]

[0081] Where, Iw represents the weighted average conversion rate of the i-th customer in several advertisements, i represents the i-th customer interaction index, T i represents the average length of stay of the i-th customer, δ i represents the number of inactive days of the i-th customer, P i represents the preference fit index of the i-th customer, η i represents the matching degree of the content of the i-th customer, w1, w2, w3 and w4 represent weight coefficients, 0≤w1≤1, 0≤w2≤1, 0≤w3≤1, 0≤w4≤1 and w1+w2+w3+w4=1;

[0082]

[0083] Where n represents the number of times the i-th customer inspects the advertisement, C i,k represents the actual conversion rate of the i-th customer in the k-th advertising campaign, e represents a natural constant, and λ represents a weight decay parameter obtained by model training;

[0084]

[0085] Where θ j Indicates the behavior weight coefficient of the jth interactive behavior type, obtained by model training, I i,j represents the number of times the i-th customer has performed the j-th interactive behavior type;

[0086]

[0087] Where V i represents the interest preference vector of the i-th customer, A i The feature vector representing the preference of the i-th customer for the content of the advertisement.

[0088] In this embodiment, by combining multi-dimensional customer behavior data with preference profile data and using dimensionless processing to calculate the customer response potential coefficient KHX, this module accurately assesses customer interaction performance and conversion potential in advertising. By comprehensively calculating indicators such as weighted average conversion rate, interaction index, dwell time, and inactive days, it enables real-time monitoring and prediction of customer ad responsiveness, thereby optimizing advertising delivery strategies and improving conversion efficiency and customer match.

[0089] Example 6

[0090] This embodiment is an explanation of Embodiment 5. Specifically, the first analysis unit is configured to preset a first threshold value Q1 in advance, and compare and analyze the customer response potential coefficient KHX with the first threshold value Q1. Obtaining a first evaluation result includes:

[0091] When the customer response potential coefficient KHX ≥ the first threshold Q1, it means that the customer response is qualified and there is no risk of customer loss, so continuous monitoring is required;

[0092] When the customer response potential coefficient KHX is less than the first threshold Q1, it indicates that the customer response is unsatisfactory and there is a risk of customer loss, triggering the first warning instruction and generating the first strategy: starting the interest recall mechanism; adjusting the content category, time period and interaction form.

[0093] In this embodiment, by comparing and analyzing the customer response potential coefficient KHX with a preset first threshold Q1, this module can identify whether customer responsiveness is acceptable in real time, promptly identifying customer churn risks and triggering an early warning mechanism. This mechanism not only optimizes advertising delivery strategies but also triggers interest recall, adjusting ad content, delivery timing, and interaction methods, thereby increasing customer engagement and retention, and reducing customer churn.

[0094] Example 7

[0095] This embodiment is explained in Example 6. Specifically, the customer interest recall module includes a second calculation unit, a third calculation unit and a second analysis unit;

[0096] The second calculation unit is used to monitor the customer's recent behavior changes and interest tag deviation in real time when receiving the first warning instruction, and calculate the interest tag drift coefficient QPY of the i-th customer after dimensionless processing based on the recent tag vector, historical preference vector, content tag, conversion effect and interactive behavior data. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , the formula is as follows:

[0097]

[0098] Where Urec i Represents the recent customer label preference vector of the i-th customer, Uhist i represents the overall historical preference vector of the i-th customer;

[0099]

[0100] Where N represents the number of recalled contents, Aa i Represents the advertising content label vector of the i-th customer, Al i,m Represents the content label vector of the historical advertisement m of the i-th customer, h i,m represents the actual conversion value of the i-th customer for ad m;

[0101]

[0102] Where, Iavg i Represents the overall historical interaction index mean of the i-th customer, Iw i represents the i-th customer interaction index, and ε is a constant to prevent the denominator from being zero.

[0103] In this embodiment, by monitoring customer behavior changes and interest tag shifts in real time, and combining multi-dimensional data including recent tags, historical preferences, ad content tags, conversion results, and interactive behavior data, the Customer Interest Recall Module can accurately calculate the degree of customer interest drift, the relevance of content recall, and behavioral response differences, thereby providing customers with more accurate ad content recommendations. This mechanism helps improve the personalized matching of ads, promote customer engagement and conversion rates, enhance advertising effectiveness, and reduce the risk of customer churn.

[0104] Example 8

[0105] This embodiment is explained in Example 7. Specifically, the third calculation unit is used to calculate and obtain the interest tag drift coefficient QPY of the i-th customer. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , after dimensionless processing, calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , the formula is as follows:

[0106] ZXY i =a1*QPY i +a2*SZH i +a3*WCY i ;

[0107] Where a1, a2, and a3 represent weight coefficients, 0≤a1≤1, 0≤a2≤1, 0≤a3≤1, and a1+a2+a3=1;

[0108] The second analysis unit is used to preset a second threshold Q2 in advance and calculate the comprehensive interest recall coefficient ZXY of the i-th customer. i Comparing and analyzing with the second threshold Q2 to obtain a second evaluation result includes:

[0109] When the comprehensive interest recall coefficient of the i-th customer ZXY i When ≤ the second threshold Q2, it means that the behavior of the i-th customer is stable and the interest deviation is within the adjustment range. No adjustment is made and continuous monitoring is performed;

[0110] When the comprehensive interest recall coefficient of the i-th customer ZXY i When the value is greater than the second threshold Q2, it indicates that the behavior of the i-th customer is unstable and the interest deviation is beyond the adjustment range, triggering the second warning instruction and generating the second strategy: re-extract the potential interest tags of the i-th customer in the latest behavior data and increase the probability of pushing candidate content containing new interest tags to 40%.

[0111] In this embodiment, by constructing a comprehensive interest recall coefficient and comparing and analyzing it with the preset threshold Q2, it is possible to accurately identify whether the customer interest deviation exceeds the adjustable range, and to achieve timely warning and strategy generation for the unstable state of customer behavior, thereby improving the matching degree of content recommendations to the customer's latest interests, improving advertising click-through rate and conversion efficiency, and effectively reducing the probability of customer churn due to interest drift.

[0112] Example 9

[0113] This embodiment is an explanation of Embodiment 8. Specifically, the feedback optimization module is used to use the changes in customer behavior after executing the strategy as training samples, and feed them back into the customer portrait analysis model to enable the model to reflect the customer's new interests and preferences; combining historical data with real-time data, the customer portrait analysis model is regularly retrained through a convolutional neural network deep learning algorithm, and continuously optimized to enable the model to more accurately predict changes in customer behavior and interests.

[0114] In this embodiment, by feeding back the customer's behavioral changes after executing the strategy into the customer portrait analysis model and regularly conducting deep learning training based on historical and real-time data, the model's recognition accuracy of customer interests and behavioral changes can be continuously optimized, thereby achieving dynamic adaptive updates of advertising recommendation strategies and improving customer response rates and personalized recommendation effects.

[0115] Example 10

[0116] A data management method for enterprise customers, please refer to Figure 2 , including the following steps:

[0117] Step 1: Collect customer behavior data and preference information through embedded SDKs, tracking points, and multi-source behavior tracking, including duration of stay, inactive days, number of interactions, ad conversion rate, interest preference vectors, tag preference vectors, ad content tags, and historical conversion records.

[0118] Step 2: Clean, complete, normalize, and format the collected customer behavior data, interest vectors, and ad conversion information. This includes standardized organization of multiple types of information, including customer behavior indicators, preference vectors, and conversion records.

[0119] Step 3: Based on a convolutional neural network, combined with customer behavior and preference data, it is trained and optimized to identify customer response potential and preference drift, and output matching advertising content with target customers;

[0120] Step 4: Monitor customer behavior in real time during the advertisement. Combine multi-dimensional customer behavior data with preference feature data to calculate the customer response potential coefficient KHX. Compare and analyze this coefficient with the first threshold Q1 to determine whether the customer response is qualified. If not, implement a strategy.

[0121] Step 5: When the first warning instruction is received, the customer's recent behavior changes and interest tag drift are monitored in real time, and the interest tag drift coefficient QPY of the i-th customer is calculated. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , further calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , and compare and analyze it with the second threshold Q2 to determine whether the customer behavior is stable and whether the interest deviation is within the adjustment range. If it is unstable and does not fall within the adjustment range, a strategy is given;

[0122] Step 6: Monitor changes in customer behavior after executing the strategy and use them as training samples to feed back into the customer profiling model. Combined with historical and real-time data, the training model is regularly updated using a convolutional neural network to continuously optimize the accuracy of customer interest and behavior predictions.

[0123] In this embodiment, by building a closed-loop customer behavior data collection, portrait modeling, response evaluation, interest recall and strategy feedback optimization process, it is possible to achieve real-time identification of changes in customer interest preferences and dynamic adjustment of personalized strategies, significantly improving the accuracy of advertising content matching and customer conversion rate, enhancing user stickiness and reducing the risk of customer churn.

[0124] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by technicians in this field for each set of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0125] The above formulas are obtained by collecting a large amount of data and performing software simulation, and a formula close to the actual value is selected. The coefficients in the formula are set by those skilled in the art according to actual conditions. The above is only a preferred specific implementation method of the present invention, but the protection scope of the present invention is not limited to this. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solution and inventive concept of the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A data management system for corporate customers, characterized in that: It includes data collection module, data processing module, customer portrait analysis model building module, behavior prediction analysis module, customer interest recall module and feedback optimization module; The data collection module is used to collect customer behavior data and preference information through embedded SDK, tracking points and multi-source behavior tracking, including length of stay, number of inactive days, number of interactions, advertising conversion rate, interest preference vector, tag preference vector, advertising content tag and historical conversion records; The data processing module is used to clean, complete, normalize and unify the format of the collected customer behavior data, interest vectors and advertising conversion information, covering the standardized organization of multiple types of information such as customer behavior indicators, preference vectors and conversion records; The customer portrait analysis model building module is used to train and optimize based on convolutional neural networks, combined with customer behavior and preference data, to identify customer response potential and preference drift, and to match advertising content with target customers; The behavior prediction analysis module is used to monitor the customer's behavior in the advertisement in real time, combine the multi-dimensional customer behavior data and preference feature data, calculate the customer response potential coefficient KHX, and compare and analyze it with the first threshold Q1 to determine whether the customer response degree is qualified. If not, a strategy is given; The customer interest recall module is used to monitor the customer's recent behavior changes and interest tag deviations in real time when receiving the first warning instruction, and calculate the interest tag drift coefficient QPY of the i-th customer respectively. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , further calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , and compare and analyze it with the second threshold Q2 to determine whether the customer behavior is stable and whether the interest deviation is within the adjustment range. If it is unstable and does not fall within the adjustment range, a strategy is given; The feedback optimization module is used to monitor the changes in customer behavior after executing the strategy, and feed them back into the customer portrait analysis model as training samples. It also combines historical and real-time data and uses convolutional neural networks to regularly update the training model to continuously optimize the accuracy of customer interest and behavior predictions.

2. The data management system for enterprise customers according to claim 1, characterized in that: The data collection module is used to collect the average length of stay of customers through the embedded front-end SDK; collect the number of inactive days of customers through the log tracking system; collect the actual conversion rate of customers in each advertisement through the advertising delivery monitoring system; collect the number of times customers engage in various interactive behaviors through the interactive behavior tracking module; collect the customer's preference feature vector for advertising content and the customer's interest preference vector through the advertising content adaptation record; collect the customer's recent customer label preference vector through the recent user behavior label extraction module; collect the customer's overall historical preference vector through the user historical behavior analysis system; collect the label vector of the customer's advertising content through the advertising content label management module; collect the content label vector of historical advertisements that the customer has received through the advertising delivery history database; and collect the customer's actual conversion value for historical advertisements through the conversion effect recording interface.

3. The data management system for enterprise customers according to claim 2, characterized in that: The data processing module is used to uniformly process the collected customer behavior data, interest vectors and advertising conversion information, including data cleaning, missing value filling, normalization, label de-redundancy, vector dimension alignment and anomaly detection operations; the processed data includes: the average length of customer stay, number of inactive days, actual conversion rate in advertising delivery, number of interactive behaviors, interest preference vector, advertising preference feature vector, recent customer label preference vector, overall historical preference vector, advertising content label vector, historical advertising content label vector and actual conversion value of historical advertisements.

4. The data management system for enterprise customers according to claim 3, characterized in that: The customer portrait analysis model establishment module is used to use a convolutional neural network to construct an initial convolutional neural network model, and train and test the initial convolutional neural network model with customer conversion rate data and historical conversion rate data, and use the trained initial convolutional neural network model as the customer portrait analysis model. At the same time, the intermediate layer output of customer behavior data and customer preference data is used as a feature vector to identify customer response potential and preference drift characteristics, and the customer portrait analysis model is trained and tested through the obtained feature vector, and the trained customer portrait analysis model is output as advertising content and target customer matching.

5. The data management system for enterprise customers according to claim 4, characterized in that: The behavior prediction and analysis module includes a first calculation unit and a first analysis unit; The first calculation unit is used to monitor the customer's behavior in the advertisement in real time, and calculate the customer response potential coefficient KHX after dimensionless processing by combining multi-dimensional customer behavior data and preference feature data. The formula is as follows: Where, Iw represents the weighted average conversion rate of the i-th customer in several advertisements, i represents the i-th customer interaction index, T i represents the average length of stay of the i-th customer, δ i represents the number of inactive days of the i-th customer, P i represents the preference fit index of the i-th customer, η i represents the matching degree of the content of the i-th customer, w1, w2, w3 and w4 represent weight coefficients; Where n represents the number of times the i-th customer inspects the advertisement, C i,k represents the actual conversion rate of the i-th customer in the k-th advertising campaign, e represents a natural constant, and λ represents a weight decay parameter obtained by model training; Where θ j Indicates the behavior weight coefficient of the jth interactive behavior type, obtained by model training, I i,j represents the number of times the i-th customer has performed the j-th interactive behavior type; Where V i represents the interest preference vector of the i-th customer, A i The feature vector representing the preference of the i-th customer for the content of the advertisement.

6. The data management system for enterprise customers according to claim 5, characterized in that: The first analysis unit is used to preset a first threshold Q1 in advance, and compare and analyze the customer response potential coefficient KHX with the first threshold Q1 to obtain a first evaluation result including: When the customer response potential coefficient KHX ≥ the first threshold Q1, it means that the customer response is qualified and there is no risk of customer loss, so continuous monitoring is required; When the customer response potential coefficient KHX is less than the first threshold Q1, it indicates that the customer response is unsatisfactory and there is a risk of customer loss, triggering the first warning instruction and generating the first strategy: starting the interest recall mechanism; adjusting the content category, time period and interaction form.

7. The data management system for enterprise customers according to claim 6, characterized in that: The customer interest recall module includes a second calculation unit, a third calculation unit and a second analysis unit; The second calculation unit is used to monitor the customer's recent behavior changes and interest tag deviation in real time when receiving the first warning instruction, and calculate the interest tag drift coefficient QPY of the i-th customer after dimensionless processing based on the recent tag vector, historical preference vector, content tag, conversion effect and interactive behavior data. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , the formula is as follows: Where Urec i Represents the recent customer label preference vector of the i-th customer, Uhist i represents the overall historical preference vector of the i-th customer; Where N represents the number of recalled contents, Aa i Represents the advertising content label vector of the i-th customer, Al i,m Represents the content label vector of the historical advertisement m of the i-th customer, h i,m represents the actual conversion value of the i-th customer for ad m; Where, Iavg i Represents the overall historical interaction index mean of the i-th customer, Iw i represents the i-th customer interaction index, and ε is a constant to prevent the denominator from being zero.

8. The data management system for enterprise customers according to claim 7, characterized in that: The third calculation unit is used to calculate and obtain the interest tag drift coefficient QPY of the i-th customer. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , after dimensionless processing, calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , the formula is as follows: <h2 style=";text-align:left;direction:ltr">ZXY<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> =a1*QPY<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> +a2*SZH<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> +a3*WCY<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ; Where a1, a2 and a3 represent weight coefficients; The second analysis unit is used to preset a second threshold Q2 in advance and calculate the comprehensive interest recall coefficient ZXY of the i-th customer. i Comparing and analyzing with the second threshold Q2 to obtain a second evaluation result includes: When the comprehensive interest recall coefficient of the i-th customer ZXY i When ≤ the second threshold Q2, it means that the behavior of the i-th customer is stable and the interest deviation is within the adjustment range. No adjustment is made and continuous monitoring is performed; When the comprehensive interest recall coefficient of the i-th customer ZXY i When the value is greater than the second threshold Q2, it indicates that the behavior of the i-th customer is unstable and the interest deviation is beyond the adjustment range, triggering the second warning instruction and generating the second strategy: re-extract the potential interest tags of the i-th customer in the latest behavior data and increase the probability of pushing candidate content containing new interest tags to 40%.

9. The data management system for enterprise customers according to claim 8, characterized in that: The feedback optimization module is used to use the changes in customer behavior after executing the strategy as training samples and feed them back into the customer portrait analysis model, so that the model can reflect the customer's new interests and preferences; combining historical data with real-time data, the customer portrait analysis model is regularly retrained through the convolutional neural network deep learning algorithm, and continuously optimized to enable the model to more accurately predict changes in customer behavior and interests.

10. A data management method for enterprise customers, applied to a data management system for enterprise customers according to any one of claims 1 to 9, characterized in that: The management method comprises the following steps: Step 1: Collect customer behavior data and preference information through embedded SDKs, tracking points, and multi-source behavior tracking, including duration of stay, inactive days, number of interactions, ad conversion rate, interest preference vectors, tag preference vectors, ad content tags, and historical conversion records. Step 2: Clean, complete, normalize, and format the collected customer behavior data, interest vectors, and ad conversion information. This includes standardized organization of multiple types of information, including customer behavior indicators, preference vectors, and conversion records. Step 3: Based on a convolutional neural network, combined with customer behavior and preference data, it is trained and optimized to identify customer response potential and preference drift, and output matching advertising content with target customers; Step 4: Monitor customer behavior in real time during the advertisement. Combine multi-dimensional customer behavior data with preference feature data to calculate the customer response potential coefficient KHX. Compare and analyze this coefficient with the first threshold Q1 to determine whether the customer response is qualified. If not, implement a strategy. Step 5: When the first warning instruction is received, the customer's recent behavior changes and interest tag drift are monitored in real time, and the interest tag drift coefficient QPY of the i-th customer is calculated. i , Similar content recall coefficient SZH i and behavioral response difference coefficient WCY i , further calculate and obtain the comprehensive interest recall coefficient ZXY of the i-th customer i , and compare and analyze it with the second threshold Q2 to determine whether the customer behavior is stable and whether the interest deviation is within the adjustment range. If it is unstable and does not fall within the adjustment range, a strategy is given; Step 6: Monitor changes in customer behavior after executing the strategy and use them as training samples to feed back into the customer profiling model. Combined with historical and real-time data, the training model is regularly updated using a convolutional neural network to continuously optimize the accuracy of customer interest and behavior predictions.

Images