Customer data driven precision marketing activity generation method and system

Abstract

The invention discloses a customer data-driven precision marketing activity generation method and system, and relates to the technical field of customer data processing and precision marketing, and the method comprises the steps: fusing text, image and behavior sequence data through a multi-modal feature extraction technology, and generating an enhanced customer portrait vector; in combination with a dynamic decision-making mechanism of a reinforcement learning agent, a customer state is mapped into a quantifiable marketing action tuple, and real-time accurate adaptation of a marketing strategy is realized; a closed-loop optimization architecture from strategy execution to feedback capture is constructed, and a value function is continuously updated by utilizing a customer response behavior, so that the defect of strategy adjustment lagging of a traditional system is overcome. And finally, the adaptability of a marketing strategy is enhanced while the customer portrait accuracy is improved, and the whole technical scheme naturally has a cross-channel collaborative execution capability, so that full-process automation support is provided for precise marketing activities.

Description

Technical Field

[0001] This invention relates to the field of customer data processing and precision marketing technology, and in particular to a method and system for generating precision marketing campaigns driven by customer data. Background Technology

[0002] With the development of big data and artificial intelligence technologies, customer data-driven precision marketing methods are gradually emerging. Traditional methods mainly rely on data mining algorithms to extract and transform key indicators from marketing data to achieve precise matching. However, existing technologies have shortcomings in processing multimodal data, making it difficult to effectively integrate customer feedback information from different modalities such as text and images, resulting in inaccurate customer profiles. In addition, existing technologies also have limitations in dynamically adjusting marketing strategies, failing to quickly optimize strategies based on real-time feedback, and thus struggling to meet complex and ever-changing marketing needs.

[0003] This invention addresses the shortcomings of existing technologies by proposing a customer data-driven method for generating precise marketing campaigns based on deep learning and reinforcement learning. This method enhances customer profiles through multimodal feature extraction and dynamically generates marketing strategies using reinforcement learning. It overcomes the limitations of existing technologies in multimodal data fusion and dynamic strategy adjustment, enabling more precise and flexible marketing campaigns. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a customer data-driven method for generating precision marketing campaigns to address the problems of insufficient multimodal data fusion in existing precision marketing methods, which leads to inaccurate customer profiles, weak dynamic strategy adjustment capabilities, and difficulty in quickly optimizing marketing strategies based on real-time feedback.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for generating precision marketing campaigns driven by customer data, characterized by comprising the following steps:

[0008] Raw customer data is collected from multiple data sources, cleaned and standardized, and a structured customer dataset is generated.

[0009] The structured customer dataset is input into a deep learning model for multimodal feature extraction, and text semantic vectors, image visual features and behavioral sequence embeddings are fused to generate an enhanced customer profile vector.

[0010] The enhanced customer profile vector is input into the reinforcement learning agent, and the optimal action value function in the preset marketing action space is calculated through the Q-learning algorithm. The output is a marketing strategy tuple containing the target product ID, the recommended channel type, and the incentive strength.

[0011] Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel interface to execute marketing actions and send marketing content that matches the target product ID and incentive level to the target customer terminal;

[0012] Capture customer terminal responses to marketing content in real time, extract response types and timestamps to generate feedback data streams;

[0013] The feedback data stream is input into the reinforcement learning agent to update the state value function, generating policy adjustment instructions to form a closed-loop optimization.

[0014] As a preferred embodiment of the customer data-driven precision marketing campaign generation method described in this invention, the steps of collecting raw customer data from multiple data sources, cleaning and standardizing the raw customer data, and generating a structured customer dataset are as follows:

[0015] We collect raw data such as customer browsing behavior, purchase records, comments, and social interactions from online platforms, offline stores, and social media.

[0016] The collected raw data is cleaned to remove duplicate data, abnormal data, and missing values;

[0017] The cleaned data is standardized to convert data of different formats and units into a unified structured data format;

[0018] Structured data is stored in a data warehouse, providing a data foundation for subsequent enhancement of customer profiles.

[0019] As a preferred embodiment of the customer data-driven precision marketing campaign generation method described in this invention, the steps include: inputting a structured customer dataset into a deep learning model for multimodal feature extraction, fusing text semantic vectors, image visual features, and behavioral sequence embeddings to generate an enhanced customer profile vector; and the specific steps are as follows.

[0020] The pre-trained BERT model is used to encode text data such as customer reviews and extract text semantic vectors;

[0021] Convolutional neural networks (CNNs) are used to extract features from image data such as customer-uploaded images or product images to obtain image visual features;

[0022] Recurrent neural networks (RNNs) are used to model customer browsing and purchasing behavior sequences, generating behavior sequence embeddings;

[0023] By fusing text semantic vectors, image visual features, and behavioral sequences, an enhanced customer profile vector is generated through a multimodal fusion algorithm.

[0024] As a preferred embodiment of the customer data-driven precision marketing campaign generation method described in this invention, the steps include: inputting the enhanced customer profile vector into the reinforcement learning agent, calculating the optimal action value function within a preset marketing action space using the Q-learning algorithm, and outputting a marketing strategy tuple containing the target product ID, recommendation channel type, and incentive level.

[0025] Initialize the reinforcement learning agent and set up a preset marketing action space, including different target product IDs, recommendation channel types, and incentive levels;

[0026] The enhanced customer profile vector is used as the input state of the reinforcement learning agent, and the state value function of each action is calculated through the Q-learning algorithm.

[0027] Select the optimal action based on the state value function, and output a marketing strategy tuple containing the target product ID, recommended channel type, and incentive level;

[0028] Store marketing strategy tuples in the strategy library to provide a strategic basis for subsequent marketing actions.

[0029] As a preferred embodiment of the customer data-driven precision marketing campaign generation method described in this invention, the step of calling the corresponding channel interface to execute marketing actions based on the recommended channel type in the marketing strategy tuple, and sending marketing content matching the target product ID and incentive level to the target customer terminal, specifically includes the following steps:

[0030] Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel's API interface, such as e-commerce platform interface, social media interface, etc.

[0031] Retrieve product details from the product database based on the target product ID, and generate personalized marketing content such as coupons and recommendation copy based on the incentive level;

[0032] The generated marketing content will be sent to the target customers' terminal devices, such as mobile phones and computers, through the channel interfaces invoked.

[0033] Record the execution log of marketing activities, including the sending time, target customer ID, and other information, to provide data support for subsequent performance evaluation.

[0034] As a preferred embodiment of the customer data-driven precision marketing campaign generation method described in this invention, the steps of real-time capture of customer terminal response behavior to marketing content, extraction of response type and timestamp to generate feedback data stream are as follows:

[0035] Set up listeners on the target customer's terminal device to capture the customer's response behavior to marketing content in real time, such as clicking, purchasing, ignoring, etc.

[0036] Extract the type of response behavior and the timestamp of the occurrence time to generate a feedback data stream;

[0037] The feedback data stream is transmitted to the server for initial parsing and storage;

[0038] The parsed feedback data is used as input to the reinforcement learning agent for policy adjustment.

[0039] As a preferred embodiment of the customer data-driven precision marketing campaign generation method described in this invention, the specific steps of inputting feedback data streams into the reinforcement learning agent to update the state value function and generate strategy adjustment instructions to form a closed-loop optimization are as follows:

[0040] The feedback data stream is input into the reinforcement learning agent, and the state value function is updated based on the customer's response behavior.

[0041] Based on the updated state value function, generate strategy adjustment instructions, such as adjusting the target product ID, recommendation channel type, or incentive level;

[0042] Apply the strategy adjustment command to the marketing strategy tuple in the strategy library to update the marketing strategy;

[0043] Establish a closed-loop optimization mechanism to continuously optimize marketing strategies and improve marketing effectiveness.

[0044] Secondly, the present invention provides a customer data-driven precision marketing campaign generation system, including a data collection and cleaning module, which collects raw customer data from multiple data sources, cleans and standardizes the raw customer data, and generates a structured customer dataset.

[0045] The feature extraction and profiling module inputs the structured customer dataset into the deep learning model for multimodal feature extraction, and integrates text semantic vectors, image visual features and behavioral sequence embeddings to generate enhanced customer profile vectors.

[0046] The strategy generation and optimization module inputs the enhanced customer profile vector into the reinforcement learning agent, calculates the optimal action value function in the preset marketing action space through the Q-learning algorithm, and outputs a marketing strategy tuple containing the target product ID, recommended channel type, and incentive strength.

[0047] The marketing execution and feedback module, based on the recommended channel type in the marketing strategy tuple, calls the corresponding channel interface to execute marketing actions and sends marketing content that matches the target product ID and incentive level to the target customer terminal;

[0048] The real-time feedback and capture module captures the client's response to marketing content in real time, extracts the response type and timestamp to generate a feedback data stream;

[0049] The policy adjustment and optimization module inputs the feedback data stream into the reinforcement learning agent to update the state value function and generates policy adjustment instructions to form a closed-loop optimization.

[0050] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the customer data-driven precision marketing campaign generation method as described in the first aspect of the present invention.

[0051] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the customer data-driven precision marketing campaign generation method as described in the first aspect of the present invention.

[0052] The beneficial effects of this invention are as follows: By effectively integrating text, image, and behavioral sequence data through multimodal feature extraction technology, it solves the shortcomings of existing methods in handling heterogeneous information and generates more representative customer profile vectors; combined with the dynamic decision-making mechanism of reinforcement learning agents, it maps customer states into quantifiable marketing action tuples, achieving real-time and accurate adaptation of marketing strategies; by constructing a closed-loop optimization architecture from strategy execution to feedback capture, it continuously updates the value function using customer response behavior, overcoming the drawback of lagging strategy adjustments in traditional systems; ultimately, it enhances the adaptability of marketing strategies while improving the accuracy of customer profiles, and the entire technical solution naturally possesses cross-channel collaborative execution capabilities, providing full-process automated support for precision marketing activities. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart of the customer data-driven precision marketing campaign generation method in Example 1. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0058] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for generating precision marketing campaigns driven by customer data, characterized by including the following steps:

[0059] Raw customer data is collected from multiple data sources, cleaned and standardized, and a structured customer dataset is generated.

[0060] The structured customer dataset is input into a deep learning model for multimodal feature extraction, and text semantic vectors, image visual features and behavioral sequence embeddings are fused to generate an enhanced customer profile vector.

[0061] The enhanced customer profile vector is input into the reinforcement learning agent, and the optimal action value function in the preset marketing action space is calculated through the Q-learning algorithm. The output is a marketing strategy tuple containing the target product ID, the recommended channel type, and the incentive strength.

[0062] Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel interface to execute marketing actions and send marketing content that matches the target product ID and incentive level to the target customer terminal;

[0063] Capture customer terminal responses to marketing content in real time, extract response types and timestamps to generate feedback data streams;

[0064] The feedback data stream is input into the reinforcement learning agent to update the state value function, generating policy adjustment instructions to form a closed-loop optimization.

[0065] It should be noted that the specific operation involves collecting raw data from multiple channels, including online platforms, offline stores, and social media, such as customer browsing behavior, purchase records, reviews, and social interactions. The collected raw data is then cleaned to remove duplicates, outliers, and missing values ​​to ensure accuracy and completeness. Next, the cleaned data undergoes standardization, converting data of different formats and units into a unified structured data format for subsequent processing and analysis. Finally, the structured data is stored in a data warehouse, providing a data foundation for future customer profiling enhancement.

[0066] Beneficial effects: By extensively collecting data from multiple channels and performing cleaning and standardization processes, high-quality structured customer datasets can be generated. This provides a solid data foundation for subsequent customer profiling enhancement and the generation of precision marketing strategies, ensuring data accuracy and consistency, thereby improving the precision and reliability of the entire marketing campaign.

[0067] The specific steps are as follows: A structured customer dataset is input into a deep learning model. A pre-trained BERT model is used to encode textual data such as customer reviews, extracting semantic vectors to understand the customer's true intentions and emotional inclinations. Simultaneously, a convolutional neural network (CNN) is applied to extract features from image data such as customer-uploaded pictures or product images, obtaining visual features to capture information such as the product's appearance and usage scenarios. Furthermore, a recurrent neural network (RNN) is used to model the customer's browsing and purchasing behavior sequences, generating behavioral sequence embeddings to reflect customer interests and purchasing habits. Finally, the textual semantic vectors, image visual features, and behavioral sequence embeddings are fused using a multimodal fusion algorithm to generate an enhanced customer profile vector.

[0068] Beneficial effects: The system achieves multimodal feature extraction and fusion of customer data, generating richer and more accurate enhanced customer profile vectors. This enables the system to more comprehensively understand customer needs and preferences, providing a more accurate basis for generating subsequent precision marketing strategies, improving the personalization and targeting of marketing activities, and enhancing the interaction and stickiness between customers and the brand.

[0069] Specific operations: Initialize the reinforcement learning agent and set a preset marketing action space, which includes different target product IDs, recommendation channel types, and incentive levels. Use the enhanced customer profile vector as the input state of the reinforcement learning agent. Calculate the state value function for each action using the Q-learning algorithm to evaluate the potential value of different marketing actions in the current customer state. Select the optimal action based on the state value function and output a marketing strategy tuple containing the target product ID, recommendation channel type, and incentive level. Finally, store the marketing strategy tuple in a strategy library to provide a strategic basis for subsequent marketing action execution.

[0070] Beneficial effects: By dynamically generating optimal marketing strategy tuples using reinforcement learning algorithms, marketing actions can be dynamically adjusted based on customers' real-time status and behavior. This makes marketing campaigns more flexible and intelligent, enabling timely responses to changes in customer behavior, improving the adaptability and effectiveness of marketing strategies, and thus enhancing the overall effectiveness and return on investment of marketing campaigns.

[0071] Specific operations: Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel's API interface, such as e-commerce platform interfaces, social media interfaces, etc. Retrieve product details from the product database according to the target product ID, and generate personalized marketing content, such as coupons and recommendation copy, based on the incentive level. Send the generated marketing content to the target customer's terminal device, such as mobile phone, computer, etc., through the called channel interface. At the same time, record the execution log of the marketing action, including information such as sending time and target customer ID, to provide data support for subsequent performance evaluation.

[0072] Beneficial effects: It enables precise execution of multi-channel marketing actions, selecting the most suitable recommended channels based on customer preferences and behaviors, and generating personalized marketing content. This not only improves the attractiveness and relevance of marketing content but also provides data support for subsequent performance evaluation and strategy optimization through execution log recording, further enhancing the accuracy and effectiveness of marketing campaigns.

[0073] Specific operations: Listeners are set up on the target customer's terminal device to capture customer responses to marketing content in real time, such as clicks, purchases, and ignores. The type of response and timestamps of the occurrence time are extracted to generate a feedback data stream. This feedback data stream is transmitted to a server for initial parsing and storage. Finally, the parsed feedback data is used as input to the reinforcement learning agent for policy adjustments.

[0074] Beneficial effects: By capturing customer responses in real time and generating feedback data streams, it is possible to obtain timely and authentic customer feedback on marketing campaigns. This provides reinforcement learning agents with real-time feedback information, enabling them to dynamically adjust marketing strategies based on actual customer behavior. This achieves closed-loop optimization of marketing campaigns, improves the adaptability and effectiveness of marketing strategies, and further enhances the effectiveness of marketing campaigns and customer satisfaction.

[0075] Specific operation: Feedback data streams are input into the reinforcement learning agent, which updates the state value function based on customer response behavior, evaluating the new value of different marketing actions in the current customer state. Based on the updated state value function, strategy adjustment instructions are generated, such as adjusting the target product ID, recommended channel type, or incentive level. These instructions are then applied to marketing strategy tuples in the strategy library to update the marketing strategies. This ultimately forms a closed-loop optimization mechanism, continuously optimizing marketing strategies and improving marketing effectiveness.

[0076] Beneficial effects: It enables dynamic adjustment and closed-loop optimization of marketing strategies, allowing for timely adjustments based on real-time customer feedback. This enables marketing activities to continuously adapt to market changes and customer needs, constantly improving the accuracy and effectiveness of marketing strategies, thereby achieving long-term optimization of marketing activities and maximizing customer value.

[0077] Specifically, the process of collecting raw customer data from multiple data sources, cleaning and standardizing the raw customer data, and generating a structured customer dataset involves the following steps:

[0078] We collect raw data such as customer browsing behavior, purchase records, comments, and social interactions from online platforms, offline stores, social media, and other data sources.

[0079] The collected raw data is cleaned to remove duplicate data, abnormal data, and missing values;

[0080] The cleaned data is standardized to convert data of different formats and units into a unified structured data format;

[0081] Structured data is stored in a data warehouse to provide a data foundation for subsequent enhancement of customer profiles.

[0082] It should be noted that the specific operation involves collecting raw data from multiple channels, including online platforms, offline stores, and social media, such as customer browsing behavior, purchase records, reviews, and social interactions. The collected raw data is then cleaned to remove duplicates, outliers, and missing values ​​to ensure accuracy and completeness. Next, the cleaned data undergoes standardization, converting data of different formats and units into a unified structured data format for subsequent processing and analysis. Finally, the structured data is stored in a data warehouse, providing a data foundation for future customer profiling enhancement.

[0083] Beneficial effects: By extensively collecting data from multiple channels and performing cleaning and standardization processes, high-quality structured customer datasets can be generated. This provides a solid data foundation for subsequent customer profiling enhancement and precision marketing strategy generation, ensuring data accuracy and consistency, thereby improving the precision and reliability of the entire marketing campaign.

[0084] Specific steps: Data cleaning is performed on the collected raw data to remove duplicate data, outliers, and missing values. This process involves checking and filtering each data entry, using algorithms to identify and remove portions that do not meet data quality requirements, ensuring the accuracy and completeness of the data.

[0085] Beneficial effects: Data cleaning removes noise and errors from the data, improving its quality and usability. This provides a more accurate data foundation for subsequent data processing and analysis, avoiding analytical biases caused by data quality issues, thereby enhancing the stability and reliability of the entire system.

[0086] Specific procedures: The cleaned data undergoes standardization, converting data of different formats and units into a unified structured data format. This process includes data normalization and format unification to ensure consistency and comparability in subsequent processing.

[0087] Beneficial effects: By standardizing data, data from different sources is converted into a unified format, facilitating subsequent data fusion and analysis. This improves the efficiency and accuracy of data processing, enabling the system to utilize data resources more effectively and providing stronger support for precision marketing.

[0088] Specific steps: Store structured data in a data warehouse to provide a data foundation for subsequent customer profiling enhancement. A data warehouse is a centralized data storage system that can efficiently store and manage large amounts of structured data, facilitating subsequent querying and analysis.

[0089] Beneficial effects: By storing structured data in a data warehouse, centralized data management and efficient storage are achieved. This provides stable data support for subsequent customer profiling enhancement, enabling the system to quickly respond to data query and analysis requests, and improving the overall performance and efficiency of the system.

[0090] Specifically, the process involves inputting the structured customer dataset into a deep learning model for multimodal feature extraction, fusing text semantic vectors, image visual features, and behavioral sequence embeddings to generate an enhanced customer profile vector. The specific steps are as follows:

[0091] The pre-trained BERT model is used to encode text data such as customer reviews and extract text semantic vectors;

[0092] Convolutional neural networks (CNNs) are used to extract features from image data such as customer-uploaded images or product images to obtain image visual features;

[0093] Recurrent neural networks (RNNs) are used to model customer browsing and purchasing behavior sequences, generating behavior sequence embeddings;

[0094] By fusing text semantic vectors, image visual features, and behavioral sequences, an enhanced customer profile vector is generated through a multimodal fusion algorithm.

[0095] It should be noted that a pre-trained BERT model is used to encode text data such as customer reviews and extract semantic vectors. Specifically, the customer review text is first segmented into individual word units. Then, using the pre-trained parameters of the BERT model, these word units are vectorized to generate a semantic vector for each word. Next, the BERT model's multi-layer encoder structure is used to model the context of the word semantic vectors, capturing the semantic relationships and contextual information between words. Finally, the semantic information of the entire text is integrated into a fixed-length vector, i.e., the text semantic vector.

[0096] By using a pre-trained BERT model, the semantic information and contextual relationships in text data can be deeply captured, generating high-quality text semantic vectors. This enables the system to more accurately understand customer comments and sentiment, providing rich textual feature information for subsequent customer profiling enhancement, thereby improving the accuracy of customer profiling.

[0097] Convolutional Neural Networks (CNNs) are used to extract features from image data, such as customer-uploaded images or product images, to obtain visual features. The specific process includes: first, preprocessing the image data, such as adjusting image size and normalizing pixel values, to adapt to the input requirements of the CNN model. Then, the preprocessed image is input into the CNN model, where features are extracted layer by layer through convolutional layers, pooling layers, and other structures. Convolutional layers use convolutional kernels to slide across the image, extracting local features; pooling layers downsample local features, reducing feature dimensionality. After multiple layers of convolution and pooling operations, a high-level feature representation of the image, i.e., the image's visual features, is obtained.

[0098] CNN models can automatically learn visual features in images, capturing information such as texture, shape, and color to generate distinctive visual features. This allows the system to understand customer preferences and needs for products from a visual perspective, enriching the multimodal features of customer profiles and further improving their comprehensiveness and accuracy.

[0099] Recurrent Neural Networks (RNNs) are used to model customer browsing and purchasing behavior sequences, generating behavior sequence embeddings. First, customer browsing and purchasing behaviors are arranged chronologically to form a behavior sequence. Then, each behavior is represented as a vector, which serves as the input to the RNN model. The RNN model, through its recurrent structure, processes the behavior sequence step by step, remembering information from previous behaviors and combining it with the current behavior to generate the behavior features at each time step. Finally, the features of the entire behavior sequence are integrated into a fixed-length vector, i.e., the behavior sequence embedding.

[0100] By modeling customer behavior sequences using RNN models, the time dependence and dynamic changes of customer behavior can be captured, generating behavioral sequence embeddings that reflect customer interests, preferences, and purchasing habits. This allows the system to gain a deep understanding of customer needs from a behavioral perspective, providing crucial behavioral feature support for precision marketing and enhancing the personalization and targeting of marketing strategies.

[0101] This method fuses textual semantic vectors, image visual features, and behavioral sequence embeddings to generate an enhanced customer profile vector using a multimodal fusion algorithm. Specifically, the three features are first normalized to ensure comparability. Then, a suitable fusion algorithm, such as weighted summation, concatenation, or attention mechanisms, is selected to fuse the textual semantic vectors, image visual features, and behavioral sequence embeddings. The fused vector comprehensively reflects the customer's textual, visual, and behavioral information, forming a more comprehensive and richer customer profile vector.

[0102] By integrating features from text, image, and behavioral modalities through a multimodal fusion algorithm, the generated enhanced customer profile vector more comprehensively reflects customer characteristics and needs. This enables the system to understand customers from multiple perspectives, providing more accurate and richer customer information for precision marketing, thereby improving the effectiveness of marketing campaigns and customer satisfaction.

[0103] Specifically, the enhanced customer profile vector is input into the reinforcement learning agent, and the optimal action value function within the preset marketing action space is calculated using the Q-learning algorithm. The output is a marketing strategy tuple containing the target product ID, recommendation channel type, and incentive level. The specific steps are as follows:

[0104] Initialize the reinforcement learning agent and set up a preset marketing action space, including different target product IDs, recommendation channel types, and incentive levels;

[0105] The enhanced customer profile vector is used as the input state of the reinforcement learning agent, and the state value function of each action is calculated through the Q-learning algorithm.

[0106] Select the optimal action based on the state value function, and output a marketing strategy tuple containing the target product ID, recommended channel type, and incentive level;

[0107] Store marketing strategy tuples in the strategy library to provide a strategic basis for subsequent marketing actions.

[0108] It should be noted that the initialization of the reinforcement learning agent involves setting up a pre-defined marketing action space, including different target product IDs, recommendation channel types, and incentive levels. Specifically, the initial state of the reinforcement learning agent is first defined, including its initial parameters and hyperparameters such as the learning rate. Then, based on the specific needs of the marketing campaign, the pre-defined marketing action space is clarified. This space covers all possible marketing actions, such as different target product IDs, recommendation channel types (e.g., email, social media, SMS), and incentive levels (e.g., discounts, points, gifts). In this way, a clear operational framework is provided for the reinforcement learning agent, enabling it to dynamically select the optimal marketing action based on customer profiles and real-time feedback during subsequent learning.

[0109] By initializing the reinforcement learning agent and setting a preset marketing action space, a basic framework is provided for the dynamic generation and optimization of the entire marketing strategy. This allows the system to learn and make decisions within a clear scope, avoiding the uncertainty of blindly selecting marketing actions and improving the targeting and effectiveness of the marketing strategy. At the same time, the preset marketing action space can be flexibly adjusted according to different marketing objectives and customer groups, enhancing the system's adaptability and flexibility.

[0110] The enhanced customer profile vector is used as the input state of the reinforcement learning agent, and the state value function of each action is calculated using the Q-learning algorithm. Specifically, the enhanced customer profile vector is input into the reinforcement learning agent, which then uses the Q-learning algorithm to evaluate the potential value of each action in a pre-defined marketing action space based on the current customer profile state. The Q-learning algorithm learns the relationship between the customer profile state and marketing actions, dynamically updating the state value function of each action, thereby providing a basis for selecting the optimal marketing action.

[0111] By using enhanced customer profile vectors as input states, the reinforcement learning agent can make precise marketing decisions based on detailed customer characteristics. Through Q-learning algorithms to calculate the state value function for each action, the system can dynamically evaluate the potential effectiveness of different marketing actions in the current customer state, thereby selecting the optimal marketing action. This approach not only improves the accuracy of marketing strategies but also dynamically adjusts strategies based on real-time customer behavior and feedback, enhancing the adaptability and flexibility of marketing activities.

[0112] The optimal action is selected based on the state value function, and a marketing strategy tuple containing the target product ID, recommended channel type, and incentive level is output. Specifically, the reinforcement learning agent selects the action with the highest value from a pre-defined marketing action space based on the calculated state value function. This action corresponds to a specific target product ID, recommended channel type, and incentive level, forming a marketing strategy tuple. This tuple details the optimal marketing strategy for the current customer, including the recommended product, recommended channel, and specific incentive measures.

[0113] By selecting the optimal action through a state-value function, a marketing strategy tuple containing the target product ID, recommended channel type, and incentive level is generated, enabling precise customization of marketing strategies. This method dynamically generates the most suitable marketing strategy for each customer based on their real-time state and behavior, improving the personalization and targeting of marketing campaigns. Simultaneously, the output marketing strategy tuple provides clear guidance for subsequent marketing actions, ensuring the efficient implementation of marketing activities.

[0114] Marketing strategy tuples are stored in a strategy repository to provide a strategic basis for subsequent marketing actions. Specifically, the generated marketing strategy tuples are stored in a centralized storage system that holds all generated marketing strategies. When a marketing action needs to be executed, the system can retrieve the corresponding marketing strategy tuple from the strategy repository and execute the specific marketing action based on the target product ID, recommended channel type, and incentive level. This process ensures the consistency and traceability of marketing strategies and also provides data support for subsequent strategy adjustments and optimizations.

[0115] Storing marketing strategy tuples in a strategy repository provides a stable and reliable strategic basis for subsequent marketing actions. This not only ensures the efficient implementation of marketing strategies but also facilitates the monitoring and evaluation of marketing activities. Through centralized management of the strategy repository, the system can quickly retrieve and apply optimal marketing strategies, improving the efficiency and effectiveness of marketing activities. Simultaneously, the data in the strategy repository also supports subsequent strategy adjustments and optimizations, further enhancing the accuracy and adaptability of marketing activities.

[0116] Specifically, based on the recommended channel type in the marketing strategy tuple, the corresponding channel interface is called to execute marketing actions, sending marketing content matching the target product ID and incentive level to the target customer terminal. The specific steps are as follows:

[0117] Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel's API interface, such as e-commerce platform interface, social media interface, etc.

[0118] Retrieve product details from the product database based on the target product ID, and generate personalized marketing content such as coupons and recommendation copy based on the incentive level;

[0119] The generated marketing content will be sent to the target customers' terminal devices, such as mobile phones and computers, through the channel interfaces invoked.

[0120] Record the execution log of marketing activities, including the sending time, target customer ID, and other information, to provide data support for subsequent performance evaluation.

[0121] It should be noted that the system calls the corresponding channel's API interface based on the recommendation channel type in the marketing strategy tuple. Specifically, the system first parses the marketing strategy tuple to identify the recommendation channel type, such as e-commerce platforms, social media platforms, or email platforms. Then, the system calls the corresponding channel's API interface via network request to establish a data communication connection with that channel. During the call, the system transmits necessary authentication information and request parameters to ensure the legality and accuracy of the API call. For example, for e-commerce platform interfaces, the system transmits parameters such as the target product ID and user identity information; for social media interfaces, the system transmits parameters such as recommendation text and images. In this way, the system can seamlessly integrate with different marketing channels, providing technical support for subsequent marketing content delivery.

[0122] By calling the corresponding channel's API interface, an effective connection between marketing strategies and specific marketing channels was achieved. This allows the system to flexibly select and connect to multiple marketing channels based on different marketing strategies, broadening the coverage of marketing activities and increasing their diversity and flexibility. At the same time, API calls ensure efficient and accurate data transmission, providing technical support for the smooth implementation of marketing activities.

[0123] The system retrieves product details from the product database based on the target product ID and generates personalized marketing content according to the incentive level. Specifically, the system queries the product database using the target product ID from the marketing strategy tuple to obtain detailed product information, including product name, price, image, specifications, and inventory. Then, the system combines the incentive level in the marketing strategy, such as discounts, gifts, or points, to generate personalized marketing content. For example, if the incentive is a 20% discount, the system will highlight "Limited-time 20% discount" in the marketing copy; if the incentive is a gift, the system will clearly state "Purchase this product and receive a free gift." Furthermore, the system will personalize the marketing content based on the target customer's preferences and behavioral history, such as selecting appropriate language and image styles, to enhance the attractiveness and relevance of the marketing content.

[0124] By obtaining product details and generating personalized marketing content, the targeting and appeal of marketing campaigns are enhanced. Personalized marketing content better meets the needs and preferences of diverse customers, thereby increasing customer attention and engagement. Simultaneously, clear incentives can stimulate customers' desire to purchase, prompting them to take action and improving the conversion rate and effectiveness of marketing campaigns.

[0125] The generated marketing content is sent to the target customers' devices via the called channel interfaces. Specifically, the system encapsulates the generated marketing content into a data format that conforms to the requirements of the corresponding channel, such as JSON or XML. Then, the system sends the marketing content to the target customers' devices, such as mobile phones and computers, through the previously called API interfaces. During the sending process, the system selects an appropriate sending method and time based on the characteristics of different channels. For example, for social media channels, the system sends marketing content during peak user activity times; for email channels, the system sets appropriate email subjects and sending frequencies. Simultaneously, the system monitors the sending status in real time to ensure that the marketing content is successfully delivered to the target customers.

[0126] By sending marketing content directly to the target customers' devices, effective marketing information delivery is achieved. This allows customers to receive personalized marketing messages immediately, improving the timeliness and effectiveness of marketing campaigns. Simultaneously, the system optimizes delivery strategies based on channel characteristics, further increasing the delivery rate and read rate of marketing messages, enhancing the impact and effectiveness of marketing campaigns.

[0127] The system records the execution logs of marketing actions, including information such as the sending time and target customer ID. Specifically, the system generates a detailed execution log each time a marketing action is executed. The log content includes information such as the sending time, target customer ID, marketing content summary, sending channel, and sending status. This log information is stored in the system's log database for subsequent querying and analysis. The system also performs regular backups and archiving of the log data to ensure its security and integrity. Furthermore, the system conducts preliminary statistical analysis of the log data, generating an overview report of the marketing action execution to provide data support for subsequent performance evaluation.

[0128] By recording execution logs for marketing activities, detailed data support is provided for subsequent performance evaluation. This enables the system to comprehensively monitor and analyze the execution of marketing campaigns, promptly identifying problems and shortcomings. Simultaneously, statistical analysis of the log data provides a basis for optimizing marketing strategies, helping the system continuously improve and enhance the effectiveness and quality of marketing activities.

[0129] Specifically, the real-time capture of customer terminal response behavior to marketing content, extraction of response type and timestamp to generate feedback data stream, involves the following steps:

[0130] Set up listeners on the target customer's terminal device to capture the customer's response behavior to marketing content in real time, such as clicking, purchasing, ignoring, etc.

[0131] Extract the type of response behavior and the timestamp of the occurrence time to generate a feedback data stream;

[0132] The feedback data stream is transmitted to the server for initial parsing and storage;

[0133] The parsed feedback data is used as input to the reinforcement learning agent for policy adjustment.

[0134] It should be noted that listeners are set up on the target customer's terminal devices to capture customer responses to marketing content in real time. Specifically, the system deploys a listener program on the customer's terminal device (such as a mobile phone or computer), which can monitor the customer's interaction with the marketing content in real time. The listener program records various customer responses to the marketing content, including clicks, purchases, and ignores. To ensure the accuracy and completeness of the data, the listener program generates a timestamp for each response, recording the specific time the action occurred. These responses and their timestamps are encapsulated into a feedback data stream for subsequent processing and analysis.

[0135] By placing listeners on target customers' devices, the system can obtain real-time feedback on marketing content. This allows the system to understand customer interests and behavioral patterns promptly, providing real-time data support for subsequent marketing strategy adjustments. The real-time feedback mechanism enhances the flexibility and adaptability of marketing campaigns, better meeting customers' personalized needs.

[0136] The system extracts the type and timestamp of each response behavior to generate a feedback data stream. Specifically, the system categorizes customer response behaviors captured by the listener, identifying the specific type of each behavior, such as click, purchase, or ignore. Simultaneously, the system extracts the timestamp of each behavior, recording the exact time it occurred. Then, the system combines these response behaviors and their timestamps into a feedback data stream containing all customer feedback information regarding the marketing content. This feedback data stream is stored in a structured format for easy subsequent transmission and processing.

[0137] By extracting the type and timestamp of response behavior to generate a feedback data stream, the system can record and analyze customer feedback behavior in detail. This allows the system to more accurately evaluate the effectiveness of marketing campaigns and promptly identify problems and shortcomings in marketing strategies. The feedback data stream provides direct evidence for optimizing marketing strategies, helping to improve the accuracy and effectiveness of marketing campaigns.

[0138] The feedback data stream is transmitted to the server for initial parsing and storage. Specifically, the system transmits the generated feedback data stream to the server via the network, where it receives and stores the data. Before storage, the server performs initial parsing of the feedback data stream, extracting key information such as response behavior type, timestamp, and customer ID. The parsed data is then stored in the server's database for further analysis and processing. The server regularly backs up the stored data to ensure its security and integrity.

[0139] By transmitting feedback data streams to the server for initial parsing and storage, the system can centrally manage and analyze customer feedback behavior. This enables the system to respond quickly to customer feedback and adjust marketing strategies promptly. Simultaneously, centralized data storage facilitates subsequent in-depth analysis and mining, helping the system continuously optimize the effectiveness of marketing campaigns.

[0140] The parsed feedback data is used as input to the reinforcement learning agent for policy adjustment. Specifically, the system transmits the parsed feedback data from the server to the reinforcement learning agent, which uses this data as input to evaluate the effectiveness of the current marketing strategy. Based on customer feedback behavior, the agent updates the state value function using the Q-learning algorithm, calculating the potential value of each marketing action. Then, based on the updated state value function, the agent generates policy adjustment instructions to optimize the marketing strategy. The optimized strategy is stored in a policy library, providing a new strategic basis for subsequent marketing actions.

[0141] By using feedback data as input to the reinforcement learning agent, the system can dynamically adjust marketing strategies based on real-time customer feedback. This allows marketing campaigns to continuously adapt to market changes and customer needs, improving the adaptability and effectiveness of marketing strategies. The application of reinforcement learning algorithms further enhances the system's learning and optimization capabilities, enabling marketing campaigns to continuously improve and achieve better marketing results.

[0142] Specifically, the steps for inputting the feedback data stream into the reinforcement learning agent to update the state value function and generate policy adjustment instructions to form a closed-loop optimization are as follows:

[0143] The feedback data stream is input into the reinforcement learning agent, and the state value function is updated based on the customer's response behavior.

[0144] Based on the updated state value function, generate strategy adjustment instructions, such as adjusting the target product ID, recommendation channel type, or incentive level;

[0145] Apply the strategy adjustment command to the marketing strategy tuple in the strategy library to update the marketing strategy;

[0146] Establish a closed-loop optimization mechanism to continuously optimize marketing strategies and improve marketing effectiveness.

[0147] It should be noted that the feedback data stream is input into the reinforcement learning agent, and the state value function is updated based on the customer's response behavior. Specifically, the system first receives a feedback data stream captured from the customer's terminal device. This data stream contains the type of customer's response behavior (such as click, purchase, ignore, etc.) and the corresponding timestamp. Then, the system uses this feedback data as input to pass it to the reinforcement learning agent. The reinforcement learning agent uses this data to update the state value function through the Q-learning algorithm. Specifically, the agent calculates the action value for each state based on the customer's response behavior and updates the state value function based on these values, thereby better evaluating the effectiveness of different marketing strategies.

[0148] By feeding feedback data streams into the reinforcement learning agent and updating the state value function, the system can learn and adapt to customer feedback behavior in real time. This enables the system to more accurately evaluate the effectiveness of different marketing strategies, providing data support for subsequent strategy adjustments, thereby improving the accuracy and adaptability of marketing strategies.

[0149] Based on the updated state value function, policy adjustment instructions are generated, such as adjusting the target product ID, recommendation channel type, or incentive strength. Specifically, after updating the state value function, the reinforcement learning agent selects the optimal action based on the new value assessment results. These actions may include changing the recommended product (adjusting the target product ID), changing the recommendation channel (adjusting the recommendation channel type), or adjusting the strength of incentives (such as discounts, gifts, etc.). The agent generates specific policy adjustment instructions based on these optimal actions and transmits these instructions to the policy library.

[0150] By generating strategy adjustment instructions, the system can dynamically adjust marketing strategies based on real-time customer feedback. This allows marketing campaigns to adapt more flexibly to market changes and customer needs, improving the targeting and effectiveness of marketing strategies, thereby enhancing the overall impact of marketing campaigns.

[0151] The system updates marketing strategies by applying adjustment instructions to the marketing strategy tuples in the strategy library. Specifically, after receiving adjustment instructions from the reinforcement learning agent, the system applies these instructions to the corresponding marketing strategy tuples in the strategy library. The strategy library is a centralized database storing all marketing strategies. The system modifies relevant parameters in the strategy library based on the adjustment instructions, such as target product ID, recommendation channel type, and incentive level. The updated marketing strategy tuples are then re-stored in the strategy library for subsequent marketing actions.

[0152] By applying strategy adjustment instructions to the strategy library and updating marketing strategies, the system ensures continuous optimization of marketing strategies. This allows the system to adjust marketing strategies promptly based on customer feedback, improving the adaptability and flexibility of marketing activities, thereby better meeting customer needs and enhancing customer satisfaction.

[0153] A closed-loop optimization mechanism is established to continuously optimize marketing strategies and improve marketing effectiveness. Specifically, the system continuously receives customer feedback data, updates the state value function, and generates strategy adjustment instructions, forming a closed-loop optimization mechanism. In this mechanism, every adjustment by the system is based on the latest customer feedback, thereby ensuring that marketing strategies can be continuously optimized. The system regularly evaluates the effectiveness of marketing strategies and further adjusts the strategies based on the evaluation results to maximize marketing effectiveness.

[0154] By establishing a closed-loop optimization mechanism, the system can continuously optimize marketing strategies, ensuring the long-term effectiveness and adaptability of marketing activities. This mechanism enables the system to continuously learn and improve, thereby enhancing the accuracy and effectiveness of marketing activities and ultimately maximizing marketing results.

[0155] This embodiment also provides a customer data-driven precision marketing campaign generation system, including:

[0156] The data acquisition and cleaning module collects raw customer data from multiple data sources, cleans and standardizes the raw customer data, and generates a structured customer dataset.

[0157] The feature extraction and profiling module inputs the structured customer dataset into the deep learning model for multimodal feature extraction, and integrates text semantic vectors, image visual features and behavioral sequence embeddings to generate enhanced customer profile vectors.

[0158] The strategy generation and optimization module inputs the enhanced customer profile vector into the reinforcement learning agent, calculates the optimal action value function in the preset marketing action space through the Q-learning algorithm, and outputs a marketing strategy tuple containing the target product ID, recommended channel type, and incentive strength.

[0159] The marketing execution and feedback module, based on the recommended channel type in the marketing strategy tuple, calls the corresponding channel interface to execute marketing actions and sends marketing content that matches the target product ID and incentive level to the target customer terminal;

[0160] The real-time feedback and capture module captures the client's response to marketing content in real time, extracts the response type and timestamp to generate a feedback data stream;

[0161] The policy adjustment and optimization module inputs the feedback data stream into the reinforcement learning agent to update the state value function and generates policy adjustment instructions to form a closed-loop optimization.

[0162] This embodiment also provides a computer device applicable to the customer data-driven precision marketing campaign generation method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the customer data-driven precision marketing campaign generation method proposed in the above embodiment.

[0163] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0164] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for generating customer data-driven precision marketing campaigns as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0165] In summary, this invention addresses the shortcomings of existing methods in handling heterogeneous information by effectively fusing text, image, and behavioral sequence data through multimodal feature extraction technology, generating more representative customer profile vectors. It also maps customer states to quantifiable marketing action tuples using a reinforcement learning agent's dynamic decision-making mechanism, achieving real-time and accurate adaptation of marketing strategies. Furthermore, by constructing a closed-loop optimization architecture from strategy execution to feedback capture, it continuously updates the value function using customer response behavior, overcoming the lag in traditional system strategy adjustments. Ultimately, this invention improves the accuracy of customer profiles while enhancing the adaptability of marketing strategies. Moreover, the entire technical solution inherently possesses cross-channel collaborative execution capabilities, providing full-process automated support for precision marketing campaigns.

[0166] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating precision marketing campaigns driven by customer data, characterized in that, Includes the following steps: Raw customer data is collected from multiple data sources, cleaned and standardized, and a structured customer dataset is generated. The structured customer dataset is input into a deep learning model for multimodal feature extraction, and text semantic vectors, image visual features and behavioral sequence embeddings are fused to generate an enhanced customer profile vector. The enhanced customer profile vector is input into the reinforcement learning agent, and the optimal action value function in the preset marketing action space is calculated through the Q-learning algorithm. The output is a marketing strategy tuple containing the target product ID, the recommended channel type, and the incentive strength. Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel interface to execute marketing actions and send marketing content that matches the target product ID and incentive level to the target customer terminal; Capture customer terminal responses to marketing content in real time, extract response types and timestamps to generate feedback data streams; The feedback data stream is input into the reinforcement learning agent to update the state value function, generating policy adjustment instructions to form a closed-loop optimization.

2. The customer data-driven precision marketing campaign generation method as described in claim 1, characterized in that: The process involves collecting raw customer data from multiple data sources, cleaning and standardizing the raw customer data, and generating a structured customer dataset. The specific steps are as follows: We collect raw data such as customer browsing behavior, purchase records, comments, and social interactions from online platforms, offline stores, and social media. The collected raw data is cleaned to remove duplicate data, abnormal data, and missing values; The cleaned data is standardized to convert data of different formats and units into a unified structured data format; Structured data is stored in a data warehouse, providing a data foundation for subsequent enhancement of customer profiles.

3. The customer data-driven precision marketing campaign generation method as described in claim 2, characterized in that: The process involves inputting a structured customer dataset into a deep learning model for multimodal feature extraction, fusing text semantic vectors, image visual features, and behavioral sequence embeddings to generate an enhanced customer profile vector. The specific steps are as follows: The pre-trained BERT model is used to encode text data such as customer reviews and extract text semantic vectors; Convolutional neural networks (CNNs) are used to extract features from image data such as customer-uploaded images or product images to obtain image visual features; Recurrent neural networks (RNNs) are used to model customer browsing and purchasing behavior sequences, generating behavior sequence embeddings; By fusing text semantic vectors, image visual features, and behavioral sequences, an enhanced customer profile vector is generated through a multimodal fusion algorithm.

4. The customer data-driven precision marketing campaign generation method as described in claim 3, characterized in that: The process involves inputting an enhanced customer profile vector into a reinforcement learning agent, calculating the optimal action value function within a preset marketing action space using the Q-learning algorithm, and outputting a marketing strategy tuple containing the target product ID, recommendation channel type, and incentive level. The specific steps are as follows: Initialize the reinforcement learning agent and set up a preset marketing action space, including different target product IDs, recommendation channel types, and incentive levels; The enhanced customer profile vector is used as the input state of the reinforcement learning agent, and the state value function of each action is calculated through the Q-learning algorithm. Select the optimal action based on the state value function, and output a marketing strategy tuple containing the target product ID, recommended channel type, and incentive level; Store marketing strategy tuples in the strategy library to provide a strategic basis for subsequent marketing actions.

5. The customer data-driven precision marketing campaign generation method as described in claim 4, characterized in that: The steps involve calling the corresponding channel interface to execute marketing actions based on the recommended channel type in the marketing strategy tuple, and sending marketing content matching the target product ID and incentive level to the target customer terminal. Based on the recommended channel type in the marketing strategy tuple, call the corresponding channel's API interface, such as e-commerce platform interface, social media interface, etc. Retrieve product details from the product database based on the target product ID, and generate personalized marketing content such as coupons and recommendation copy based on the incentive level; The generated marketing content will be sent to the target customers' terminal devices, such as mobile phones and computers, through the channel interfaces invoked. Record the execution log of marketing activities, including the sending time, target customer ID, and other information, to provide data support for subsequent performance evaluation.

6. The customer data-driven precision marketing campaign generation method as described in claim 5, characterized in that: The specific steps for capturing customer terminal responses to marketing content in real time, extracting response types and timestamps to generate feedback data streams are as follows: Set up listeners on the target customer's terminal device to capture the customer's response behavior to marketing content in real time, such as clicking, purchasing, ignoring, etc. Extract the type of response behavior and the timestamp of the occurrence time to generate a feedback data stream; The feedback data stream is transmitted to the server for initial parsing and storage; The parsed feedback data is used as input to the reinforcement learning agent for policy adjustment.

7. The customer data-driven precision marketing campaign generation method as described in claim 6, characterized in that: The specific steps for inputting the feedback data stream into the reinforcement learning agent to update the state value function and generate policy adjustment instructions to form a closed-loop optimization are as follows: The feedback data stream is input into the reinforcement learning agent, and the state value function is updated based on the customer's response behavior. Based on the updated state value function, generate strategy adjustment instructions, such as adjusting the target product ID, recommendation channel type, or incentive level; Apply the strategy adjustment command to the marketing strategy tuple in the strategy library to update the marketing strategy; Establish a closed-loop optimization mechanism to continuously optimize marketing strategies and improve marketing effectiveness.

8. A customer data-driven precision marketing campaign generation system, based on the customer data-driven precision marketing campaign generation method according to any one of claims 1 to 7, characterized in that: include, The data acquisition and cleaning module collects raw customer data from multiple data sources, cleans and standardizes the raw customer data, and generates a structured customer dataset. The feature extraction and profiling module inputs the structured customer dataset into the deep learning model for multimodal feature extraction, and integrates text semantic vectors, image visual features and behavioral sequence embeddings to generate enhanced customer profile vectors. The strategy generation and optimization module inputs the enhanced customer profile vector into the reinforcement learning agent, calculates the optimal action value function in the preset marketing action space through the Q-learning algorithm, and outputs a marketing strategy tuple containing the target product ID, recommended channel type, and incentive strength. The marketing execution and feedback module, based on the recommended channel type in the marketing strategy tuple, calls the corresponding channel interface to execute marketing actions and sends marketing content that matches the target product ID and incentive level to the target customer terminal; The real-time feedback and capture module captures the client's response to marketing content in real time, extracts the response type and timestamp to generate a feedback data stream; The policy adjustment and optimization module inputs the feedback data stream into the reinforcement learning agent to update the state value function and generates policy adjustment instructions to form a closed-loop optimization.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the customer data-driven precision marketing campaign generation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the customer data-driven precision marketing campaign generation method according to any one of claims 1 to 7.

Images