Intelligent short message marketing carousel method and device and readable storage medium

Abstract

The invention provides an intelligent short message marketing carousel method and device and a readable storage medium, and the method comprises the steps: obtaining user multi-dimensional data which comprises user basic attributes, historical interaction behavior data and communication social data; according to the multi-dimensional data of the user, obtaining a dynamic interest vector of the user through a deep learning model; according to the multi-dimensional data of the user, predicting an easily-visited probability value of the user through an ensemble learning gradient boosting tree GBDT algorithm; determining a plurality of candidate marketing contents corresponding to each user from a marketing content library through a reinforcement learning algorithm on the basis of the dynamic interest vector and the user easy-to-visit probability value; and performing short message marketing carousel on the users according to the plurality of candidate marketing contents corresponding to the users. Through mining and analysis of user preferences and organic fusion with a short message marketing carousel strategy, the acceptability and satisfaction of the user to short message marketing are enhanced.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to an intelligent SMS marketing carousel method, device and readable storage medium. Background Art

[0002] SMS marketing, as a traditional and important marketing method in the communications industry, still occupies a unique position. It has the characteristics of low cost, fast transmission speed, wide coverage, etc., and can deliver information to a large number of users in a short time.

[0003] However, with increasingly fierce market competition and evolving consumer demands, SMS marketing faces numerous challenges. For example, consumers receive a large number of marketing text messages daily, leading to a growing indifference to standard SMS marketing content and significantly reducing marketing effectiveness. Furthermore, the excessive marketing and harassing text messages sent by some unscrupulous businesses have aroused user resentment and negatively impacted the overall reputation of SMS marketing.

[0004] Traditional SMS marketing often relies on large-scale, indiscriminate push notifications, which not only results in low conversion rates but also easily triggers user dissatisfaction and wastes resources. Therefore, how to improve the effectiveness of SMS marketing has become a pressing issue. Summary of the Invention

[0005] The technical problem to be solved by this application is to provide an intelligent SMS marketing carousel method, device and readable storage medium to solve the problems existing in the prior art.

[0006] In the first aspect, the present application provides an intelligent SMS marketing carousel method, the method

[0007] The law includes:

[0008] S1. Obtain multi-dimensional user data, including basic user attributes, historical interaction behavior data, and communication and social data;

[0009] S2. Obtaining the user's dynamic interest vector through a deep learning model based on the user's multi-dimensional data;

[0010] S3. Predicting the probability value of the user being interviewed by using an integrated learning gradient boosting tree (GBDT) algorithm based on the user's multi-dimensional data;

[0011] S4. Based on the dynamic interest vector and the user's probability of being interviewed, determine multiple candidate marketing contents corresponding to each user from the marketing content library using a reinforcement learning algorithm;

[0012] S5. Perform SMS marketing rotation for each user based on multiple candidate marketing contents corresponding to each user.

[0013] In some embodiments, S2 includes:

[0014] Extracting user behavior sequences and spatiotemporal contexts based on the multi-dimensional user data;

[0015] Based on the user behavior sequence and spatiotemporal context, feature extraction is performed through the Transformer encoder to obtain the user's long-term interests;

[0016] Based on the user behavior sequence and spatiotemporal context, feature extraction is performed through the gated recurrent unit (GRU) network to obtain the user's short-term dynamics;

[0017] The user's long-term interests and short-term dynamics are fused with features through the attention mechanism to obtain the user's dynamic interest vector.

[0018] In some embodiments, S3 includes:

[0019] Screening, converting and encoding the user's multi-dimensional data to obtain key features of the user;

[0020] Based on the key features of the user, the probability value of the user being easily interviewed is predicted by the integrated learning gradient boosting tree (GBDT) algorithm.

[0021] In some embodiments, the user multi-dimensional data is screened, converted, and encoded to obtain user key features, including:

[0022] Performing one-hot encoding on the category data in the user multi-dimensional data and converting it into a numerical form that can be processed by a computer;

[0023] Standardizing the numerical data in the user multi-dimensional data to obtain data in a standard format;

[0024] The multi-dimensional data of the user is analyzed by an analysis algorithm, and features with low correlation with the user interview prediction are screened out and removed, while features with high correlation with the user interview prediction are retained.

[0025] In some embodiments, S4 includes:

[0026] Based on the dynamic interest vector and the user's interview probability value, screening a first preset number of marketing contents from a marketing content library according to preset screening rules;

[0027] Using a reinforcement learning algorithm, the first preset number of marketing contents are sorted with the goal of maximizing the reward to obtain a sorting result;

[0028] A second preset number of candidate marketing contents corresponding to each user is obtained according to the sorting result, wherein the second preset number is smaller than the first preset number.

[0029] In some embodiments, S5 includes:

[0030] Obtain the user's click-through rate (CTR) and conversion rate for each marketing carousel content;

[0031] Adjust each user's marketing carousel content in real time based on the click-through rate (CTR) and conversion rate of each marketing carousel content.

[0032] In some embodiments, S5 further includes at least one of the following:

[0033] Parallel processing of marketing SMS sending tasks through multi-node distributed layout;

[0034] Provide intelligent Do Not Disturb protection, SMS frequency control, and sensitive content filtering for each user;

[0035] Get SMS marketing carousel result tracking data and display it in chart form.

[0036] In a second aspect, the present application provides an intelligent SMS marketing carousel device, the device comprising:

[0037] A data acquisition module configured to acquire multi-dimensional user data, including basic user attributes, historical interaction behavior data, and communication and social data;

[0038] A first processing module is configured to obtain a user's dynamic interest vector through a deep learning model based on the user's multi-dimensional data;

[0039] A second processing module is configured to predict the user's interview probability value through an integrated learning gradient boosting tree (GBDT) algorithm based on the user's multi-dimensional data;

[0040] a content determination module configured to determine, based on the dynamic interest vector and the user's probability of being interviewed, a plurality of candidate marketing contents corresponding to each user from a marketing content library using a reinforcement learning algorithm;

[0041] The marketing carousel module is configured to perform SMS marketing carousel for users based on multiple candidate marketing contents corresponding to each user.

[0042] In a third aspect, the present application provides an intelligent SMS marketing carousel device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the intelligent SMS marketing carousel method described in the first aspect above.

[0043] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the intelligent SMS marketing carousel method described in the first aspect is implemented.

[0044] The present application provides an intelligent SMS marketing carousel method, device, and readable storage medium, the method comprising: obtaining multi-dimensional user data, the multi-dimensional user data comprising basic user attributes, historical interaction behavior data, and communication and social data; obtaining the user's dynamic interest vector through a deep learning model based on the multi-dimensional user data; predicting the user's probability of being easily interviewed through an integrated learning gradient boosting tree (GBDT) algorithm based on the multi-dimensional user data; determining multiple candidate marketing contents corresponding to each user from a marketing content library based on the dynamic interest vector and the user's probability of being easily interviewed through a reinforcement learning algorithm; and performing SMS marketing carousels for the user based on the multiple candidate marketing contents corresponding to each user. The present application provides an intelligent SMS marketing carousel method, which improves the effectiveness of SMS marketing by mining and analyzing user preferences and organically integrating them with SMS marketing carousel strategies, and enhances user acceptance and satisfaction with SMS marketing by combining user preferences with SMS marketing carousel methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0046] Figure 1 A flowchart of an intelligent SMS marketing carousel method provided in an embodiment of the present application;

[0047] Figure 2 A schematic diagram of a specific application of the smart SMS marketing carousel provided in an embodiment of the present application;

[0048] Figure 3 A schematic diagram of the prediction of users who are likely to be interviewed by the intelligent SMS provided in an embodiment of the present application;

[0049] Figure 4 A schematic diagram of the structure of an intelligent SMS marketing carousel device provided in an embodiment of the present application;

[0050] Figure 5 A schematic structural diagram of another intelligent SMS marketing carousel device provided in an embodiment of the present application.

[0051] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the technical solution of the present application, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0053] It should be understood that the specific embodiments and drawings described herein are only used to explain the present application, rather than to limit the present application.

[0054] It can be understood that, in the absence of conflict, the various embodiments and features in the embodiments of the present application can be combined with each other.

[0055] It will be understood that, for the sake of ease of description, the drawings of this application only show the parts related to this application, while the parts not related to this application are not shown in the drawings.

[0056] It can be understood that each unit and module involved in the embodiments of the present application may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0057] It can be understood that the terms "first", "second", etc. in the embodiments of the present application are used to distinguish different objects, or to distinguish different processing of the same object, rather than to describe a specific order of objects.

[0058] It is understandable that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present application may occur in an order different from that marked in the drawings.

[0059] It is understood that the flowcharts and block diagrams of the present application illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present application. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or by a combination of hardware and computer instructions.

[0060] It can be understood that the units and modules involved in the embodiments of the present application can be implemented by software or hardware, for example, the units and modules can be located in a processor.

[0061] Traditional SMS marketing techniques are as follows:

[0062] (1) Information communication technology solution: Utilize the SMS platform and have relevant marketing personnel operate to send the edited SMS content to the target customers’ mobile phones.

[0063] (2) Target customer setting: Marketing personnel determine the target users of SMS in advance, and use their experience or past business records to determine users who may be interested in the marketing content. Based on the existing customer database, they manually obtain relevant customer basic information data.

[0064] (3) Sending plan: This includes selecting the sending date (multiple dates are possible), setting the marketing product range and SMS template in advance. For example, a corresponding SMS content template can be developed for a certain product promotion season, and scheduled for mass sending on specific dates during the promotion period.

[0065] After research, it was found that traditional SMS marketing technology has the following problems:

[0066] 1. Single data dimension: Relying on static user attributes (such as age and region), lacking in-depth mining of dynamic behaviors (real-time clicks and social interactions);

[0067] 2. Insufficient model generalization: General classification algorithms (such as logistic regression) have poor modeling effects on high-dimensional sparse data (such as long-term and short-term user behavior characteristics), resulting in low prediction accuracy.

[0068] 3. Rigid strategy: Content carousels rely on manual rules and cannot adapt to changes in user preferences in real time, resulting in a decrease in click-through rate (CTR).

[0069] 4. System performance limitations: Traditional architectures cannot support millisecond-level responses for tens of millions of users and have insufficient concurrent processing capabilities.

[0070] In the current SMS marketing model, SMS, as a traditional communication method, still holds a significant position in the marketing field. Traditional SMS marketing often relies on large-scale, indiscriminate push notifications. This approach not only has low marketing conversion rates, but also easily triggers user resentment and results in a waste of resources.

[0071] Based on the above problems, this application aims to build a set of reinforcement learning intelligent SMS marketing carousel methods. By mining and analyzing user preferences and organically integrating them with SMS marketing carousel strategies, a complete technical architecture system is constructed. By combining user preferences with SMS marketing carousel methods, the effectiveness of SMS marketing is improved and user acceptance and satisfaction with SMS marketing are enhanced. The main concepts and specific beneficial effects of this application are as follows:

[0072] 1. Accurately identify user preferences: Utilizing a comprehensive array of data analysis techniques, we conduct in-depth analysis of user communication behaviors, consumption histories, browsing history, and other data, building a comprehensive and accurate user preference model. This model accurately identifies user preferences for products, service types, service times, and message content, providing a solid data foundation for subsequent SMS marketing.

[0073] 2. Optimize SMS marketing carousel strategies: Develop personalized SMS marketing carousel strategies based on identified user preferences. This includes rationally arranging the timing, frequency, and order of SMS messages to ensure they reach users at the most appropriate time with content that best meets their needs.

[0074] 3. Improve SMS marketing effectiveness: By implementing a SMS marketing carousel based on user preferences, key SMS marketing metrics, such as open rates, click-through rates, and conversion rates, can be significantly improved. For example, through targeted SMS marketing, users can be guided to click on links in SMS messages to access product pages, increasing product views and purchases, thereby boosting sales performance.

[0075] 4. Enhance user satisfaction: This approach changes the one-size-fits-all approach of traditional SMS marketing, focusing on user preferences to provide users with truly interesting and valuable information. This reduces the frequency of irrelevant or harassing SMS messages received, improving the user experience and perception of SMS marketing.

[0076] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0077] The present application provides an intelligent SMS marketing carousel method, the working process of which can be implemented by electronic devices, such as computers, handheld smart terminals, etc. For the convenience of explanation, the embodiments of the present application are described with the method execution subject being a computer.

[0078] Figure 1 This is a schematic diagram of the intelligent SMS marketing carousel method provided in an embodiment of the present application. Figure 2 This is a schematic diagram of a specific application of the smart SMS marketing carousel provided in the embodiment of this application. Figure 3 This is a schematic diagram of the prediction of users who are likely to be interviewed by the intelligent SMS provided in the embodiment of the present application, as shown in FIG. Figures 1 to 3 As shown, the present application provides an intelligent SMS marketing carousel method, which includes S1-S5, as follows:

[0079] S1. Obtain multi-dimensional user data, including basic user attributes, historical interaction behavior data, and communication and social data;

[0080] Specifically, in order to fully understand users' potential interview preferences, the collection of multi-dimensional user data is fundamental. This multi-dimensional user data mainly covers the following key information:

[0081] (1) Basic user attributes: basic demographic information such as age, gender, region, and occupation. Different age and gender groups have different preferences for product and service needs. Regions reflect differences in consumption capacity and habits. Occupations are related to industry demand characteristics. For example, young office workers in first-tier cities may be more interested in convenient commuting products and services such as smart text messaging.

[0082] (2) Historical interaction behavior data: past interaction records with smart SMS (or traditional SMS marketing content), including the number of times the link is clicked, the frequency of replying to SMS, whether or not the user participated in the event, etc., directly reflects the user's historical enthusiasm for responding to interviews; it also includes browsing, searching, and purchasing behaviors on mobile applications and websites, deeply reflecting interest preferences and consumption habits. Users who have frequently searched for beauty products are more likely to respond to beauty-related smart SMS.

[0083] (3) Communication and social data: The characteristics of contacts in the address book (occupational distribution, industry attributes of social circles) can indirectly reflect the user's social ecology and potential needs. The interaction between friends on social software and the activeness of group participation indicate interest circles. Users who are keen on participating in fitness group chats are more likely to respond to fitness-related smart text messages.

[0084] S2. Obtaining the user's dynamic interest vector through a deep learning model based on the user's multi-dimensional data;

[0085] In some embodiments, S2 includes:

[0086] Extracting user behavior sequences and spatiotemporal contexts based on the user multi-dimensional data;

[0087] Based on the user behavior sequence and spatiotemporal context, feature extraction is performed through the Transformer encoder to obtain the user's long-term interests;

[0088] Based on the user behavior sequence and spatiotemporal context, feature extraction is performed through the gated recurrent unit (GRU) network to obtain the user's short-term dynamics;

[0089] The user's long-term interests and short-term dynamics are fused with features through the attention mechanism to obtain the user's dynamic interest vector.

[0090] In this application, user behavior sequences (clicks, purchases, social interactions, etc.) and spatiotemporal context (timestamps, geographic location, etc.) are extracted based on the multi-dimensional user data. The deep learning model uses a Transformer encoder and a GRU (Gated Recurrent Unit) network. The GRU is a variant of a recurrent neural network (RNN) designed to process sequential data. The Transformer encoder extracts long-term interests, while the GRU network captures short-term dynamics, fusing features through an attention mechanism.

[0091] Specifically, the input of the deep learning model includes: (1) user behavior sequence data, which covers the click behavior records generated by users on the platform, and details of the page elements and product details clicked; (2) purchase behavior information, including key data such as the type, quantity, and amount of purchased goods; and social interaction related data, such as social actions such as likes, comments, and shares between users. (3) It also receives spatiotemporal context data, in which the timestamp accurately records the specific time when the user behavior occurs, accurate to the second level, so as to facilitate the subsequent analysis of behavior patterns in different time periods; geographic location data is specific to longitude and latitude, and combined with map information, it can analyze the differences in user behavior in different regions.

[0092] The model construction and processing include: (1) Using the Transformer encoder, the core of which is the self-attention mechanism. By performing multiple rounds of feature extraction and integration on the input user behavior sequence, it can effectively learn the long-term interest patterns in user behavior and explore the long-distance dependencies between behaviors. (2) With the help of the GRU network, due to its special gated structure, it can efficiently capture the dynamic changes of user behavior in a short period of time and respond to emerging behavioral trends in a timely manner. (3) In the feature fusion stage, the attention mechanism is used to assign different weights to the long-term interest features extracted by the Transformer encoder and the short-term dynamic features captured by the GRU network. The weights are adaptively adjusted according to different behaviors and contextual factors, so as to more accurately fuse the two types of features.

[0093] The output consists of a final 128-dimensional dynamic interest vector, which represents the user's current interest. To ensure the model can reflect changes in user interests in real time, this vector is updated hourly. This update process re-extracts, fuses, and calculates features based on the user's behavior sequence and spatiotemporal context data from the most recent hour to generate a new dynamic interest vector.

[0094] S3. Predicting the probability value of the user being interviewed by using an integrated learning gradient boosting tree (GBDT) algorithm based on the user's multi-dimensional data;

[0095] In some embodiments, S3 includes:

[0096] Screening, converting and encoding the user's multi-dimensional data to obtain key features of the user;

[0097] Based on the key features of the user, the probability value of the user being easily interviewed is predicted by the integrated learning gradient boosting tree (GBDT) algorithm.

[0098] In some embodiments, the user multi-dimensional data is screened, converted, and encoded to obtain user key features, including:

[0099] Performing one-hot encoding on the category data in the user multi-dimensional data and converting it into a numerical form that can be processed by a computer;

[0100] Standardizing the numerical data in the user multi-dimensional data to obtain data in a standard format;

[0101] The multi-dimensional data of the user is analyzed by an analysis algorithm, and features with low correlation with the user interview prediction are screened out and removed, while features with high correlation with the user interview prediction are retained.

[0102] In this application, data is aggregated from multiple channels to ensure data integrity and accuracy. The SMS marketing platform records past SMS messages and interaction details, and collaboratively builds a comprehensive user profile data pool.

[0103] Specifically, first, during the feature engineering phase, the collected data is screened, converted, and encoded to extract key features. Among them, categorical data (such as occupation and region) is uniquely encoded and converted into a numerical form that can be processed by computers. Numerical data (such as age and number of clicks) is standardized and dimensionalized to avoid the impact of data magnitude differences on model training. For example, age is normalized to a specific range to facilitate comprehensive analysis. Using correlation analysis, principal component analysis, and other methods, redundant and low-correlation features are screened and removed, retaining features that contribute greatly to the prediction target (easily interviewed or not), such as core feature combinations such as historical SMS click counts and recent related product search frequency.

[0104] In addition, in the selection and construction of the model, the nonlinearity, complexity and prediction task characteristics of the data are comprehensively considered, and the integrated learning gradient boosting decision tree (GBDT) algorithm is adopted as the core model architecture, and some advantages of neural networks are integrated for improvement. GBDT uses serial training of multiple decision trees and gradient boosting to efficiently capture the combination logic and nonlinear relationships of data features, and has a good effect on modeling structured user data; the embedding layer of the neural network is introduced to process category features, better represent semantic information and potential associations, and map categories such as user occupations to low-dimensional dense vector space, thereby improving the model's ability to understand and learn complex category features. The overall architecture of the model includes an input layer that receives standardized feature vectors, encodes category features through the embedding layer, enters an integrated module composed of multiple layers of GBDT weak learners, and finally outputs an output layer that predicts the probability value of the user's susceptibility to interview.

[0105] Furthermore, during model training and optimization, the dataset is proportionally divided into training, validation, and test sets. The training set is used for learning and fitting model parameters, the validation set is used for parameter optimization to avoid overfitting, and the test set is used for performance evaluation. A cross-entropy loss is defined as the loss function, which is appropriate for probabilistic prediction tasks and measures the deviation between the predicted value and the true label. An adaptive learning rate strategy is employed, dynamically adjusting the step size based on the training progress. Combined with early stopping, validation set metrics are monitored and training is terminated early if performance stops improving to prevent worsening overfitting. Regular incremental training with new data is performed to update the model to adapt to changes in user behavior and continuously optimize and improve prediction accuracy.

[0106] Optionally, this application uses FocalLoss instead of the cross entropy loss function. Focal Loss dynamically adjusts the weights of samples of different categories in the loss calculation by introducing a modulation factor. For samples that are easy to classify, their weights are reduced so that the model focuses more on samples that are difficult to classify, especially minority class samples. In practical applications, the parameters of the modulation factor are adaptively adjusted according to the proportion of samples of each category in the data set, thereby effectively alleviating the problem of category imbalance and improving the model's prediction accuracy for responses of each category.

[0107] Optionally, this application uses a feature enhancement feature: an embedded attention layer. During data preprocessing, all features are thoroughly analyzed to identify key features, such as "number of clicks in the last seven days." The attention layer then assigns different weights to each feature based on its relevance to the target response.

[0108] Optionally, this application uses dynamic pruning: As data dimensions increase, the amount of computation increases exponentially, and redundant dimensions interfere with the model's learning. This application implements dynamic pruning, calculating the importance of each feature in real time during model training. The contribution of features to the target response is assessed using methods such as the Gini index. Once the feature importance of certain dimensions is found to be below a set threshold, it is immediately removed from the model.

[0109] S4. Based on the dynamic interest vector and the user's probability of being interviewed, determine multiple candidate marketing contents corresponding to each user from the marketing content library using a reinforcement learning algorithm;

[0110] Among them, users whose user probability of being easily interviewed is greater than a preset threshold are identified as target users for SMS marketing carousel, that is, users are screened by the user probability of being easily interviewed to obtain target users for SMS marketing carousel.

[0111] In some embodiments, S4 includes:

[0112] Based on the dynamic interest vector and the user's interview probability value, screening a first preset number of marketing contents from a marketing content library according to preset screening rules;

[0113] Using a reinforcement learning algorithm, the first preset number of marketing contents are sorted with the goal of maximizing the reward to obtain a sorting result;

[0114] A second preset number of candidate marketing contents corresponding to each user is obtained according to the sorting result, wherein the second preset number is smaller than the first preset number.

[0115] When determining candidate marketing content, this application first constructs the state space, including:

[0116] (1) User interest vector acquisition: Through in-depth analysis of user historical behavior data, such as browsing history, purchase history, and search keywords, deep learning models (such as Transformer and LSTM) are used to extract user interest features and construct a user interest vector. This vector can comprehensively and accurately reflect the user's interest preferences and potential needs.

[0117] (2) Marketing content feature extraction: Based on the product type of marketing content, each product is classified and coded, such as electronic products coded as 001 and clothing coded as 002, so that the model can identify different product types. For the discount strength, it is quantified into a specific value, such as the discount rate is directly used as a numerical feature, and the full-discount activity is converted into a corresponding numerical representation according to the full-discount rules. Then, the user interest vector is spliced with the content features to form a complete state space, providing comprehensive input information for the reinforcement learning model.

[0118] Then the action space is defined, including:

[0119] (1) Candidate content screening: A certain number of candidate contents are screened out from the content library according to certain rules (such as popularity, new product recommendations, etc.), and then these candidate contents are sorted.

[0120] (1) TOPX ranking strategy: Using a reinforcement learning algorithm, with the goal of maximizing rewards, candidate content is ranked and the final TOPX (e.g., X=5) content order displayed in the carousel is determined. During the ranking process, the matching degree between the user interest vector and the content features, as well as the complementarity between different content, is fully considered to ensure that the displayed content can both meet the user's interests and provide a variety of choices.

[0121] S5. Perform SMS marketing rotation for each user based on multiple candidate marketing contents corresponding to each user.

[0122] In some embodiments, S5 includes:

[0123] Obtain the user's click-through rate (CTR) and conversion rate for each marketing carousel content;

[0124] Adjust each user's marketing carousel content in real time based on the click-through rate (CTR) and conversion rate of each marketing carousel content.

[0125] In this application, weight distribution is used for SMS marketing carousel, including:

[0126] (1) CTR calculation: The number of times users click on the content in the carousel is counted in real time and divided by the number of times it is displayed to obtain the click-through rate (CTR). To ensure the stability and accuracy of the data, a sliding window method is used to calculate the CTR over the most recent period.

[0127] (2) Conversion rate calculation: Count the number of times users complete a purchase or other targeted behavior after clicking on the content, and divide it by the number of clicks to get the conversion rate. The sliding window method is also used to calculate the conversion rate.

[0128] (3) Weight Allocation and Calculation: Based on business needs and goals, CTR is weighted 70% and conversion rate is weighted 30%. The reward function is reward value = CTR * 0.7 + conversion rate * 0.3. In this way, the reinforcement learning model learns a carousel strategy that both attracts user clicks and promotes user conversions.

[0129] In some embodiments, S5 further includes at least one of the following:

[0130] Parallel processing of marketing SMS sending tasks through multi-node distributed layout;

[0131] Provide intelligent Do Not Disturb protection, SMS frequency control, and sensitive content filtering for each user;

[0132] Get SMS marketing carousel result tracking data and display it in chart form.

[0133] Specifically, this application also includes relevant management measures for high-concurrency reach, including: focusing on distributed architecture to build efficient mass messaging projects, and overcoming the problem that traditional SMS mass messaging systems are prone to freezes, delays, and even paralysis when faced with massive sending requests. With the help of a multi-node distributed layout, each node is equipped with an independent computing unit and storage module, which are interconnected through a high-speed internal network to form a powerful computing cluster. When the number of SMS sending requests is large, the system processes tasks in parallel by city and prefecture, and achieves the simultaneous sending of thousands of SMS messages per second. At the level of intelligent management of the sending queue, a multi-dimensional sorting and grouping algorithm is designed. The tail number of the recipient's number is accurately used for grouping, which is in line with the underlying logic of network transmission, reduces the resource loss caused by frequent switching of communication links, greatly improves the overall sending efficiency, and ensures that each SMS message can reach the target audience efficiently and accurately under high concurrency. This application is based on Kafka's optimized sharding strategy and supports dynamic capacity expansion (peak throughput of 1,000 messages / second).

[0134] In addition, this application also includes management measures related to compliance control, including:

[0135] (1) Do Not Disturb (DND) intelligent protection mechanism: To protect users from spam messages, an intelligent SMS filtering system is built to fully demonstrate the DND feature. The DND record of the sender number is retrieved in real time to determine whether the user of that number has requested DND.

[0136] (2) Frequency Control and Precision Limitation: Incorporating sophisticated frequency control technology, this technology is specifically designed to control the SMS sending rhythm, comply with communications regulations, and enhance user experience. Fully automated, with refined frequency thresholds. For example, marketing SMS messages can be limited to no more than three messages sent to the same mobile phone number per day.

[0137] (3) Compliance control mechanism: Sensitive content filtering mechanism, using the BERT model to detect illegal keywords in real time.

[0138] In addition, this application also includes data tracking views, including: providing SMS push conversion effects and detailed tracking views. By analyzing SMS open rates, click-through rates, conversion rates and other data, various data indicators and analysis results are intuitively displayed in the form of charts (such as bar charts, line charts, pie charts, etc.) and tables. This can help understand user reactions and preferences to SMS messages, and then optimize SMS content and marketing strategies.

[0139] This application provides an intelligent SMS marketing carousel method, which improves the effectiveness of SMS marketing by mining and analyzing user preferences and organically integrating them with SMS marketing carousel strategies, and enhances user acceptance and satisfaction with SMS marketing by combining user preferences with SMS marketing carousel methods.

[0140] It should be understood that, although the various steps in the flowcharts of the above embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they may be performed in other orders. Moreover, at least a portion of the steps in the figure may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but may be performed at different times, and their execution order is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of sub-steps or stages of other steps.

[0141] Figure 4 This is a schematic diagram of the intelligent SMS marketing carousel device provided in the embodiment of the present application, as shown in FIG. Figure 4 As shown, the present application provides an intelligent SMS marketing carousel device, the device comprising:

[0142] A data acquisition module 11 is configured to acquire multi-dimensional user data, including basic user attributes, historical interaction behavior data, and communication and social data;

[0143] A first processing module 12 is configured to obtain a user's dynamic interest vector through a deep learning model based on the user's multi-dimensional data;

[0144] The second processing module 13 is configured to predict the user's probability of being interviewed by using an integrated learning gradient boosting tree (GBDT) algorithm based on the user's multi-dimensional data;

[0145] A content determination module 14 is configured to determine a plurality of candidate marketing contents corresponding to each user from a marketing content library using a reinforcement learning algorithm based on the dynamic interest vector and the user's probability of being susceptible to an interview;

[0146] The marketing carousel module 15 is configured to perform SMS marketing carousel for each user based on a plurality of candidate marketing contents corresponding to each user.

[0147] Regarding the limitation of the intelligent SMS marketing carousel device, please refer to the limitation of the intelligent SMS marketing carousel method in the above embodiments of this application, which will not be repeated here.

[0148] Figure 5 Another schematic diagram of the intelligent SMS marketing carousel device provided in the embodiment of the present application is as follows Figure 5 As shown, the device includes a memory 22 and a processor 21, the memory stores a computer program, and the processor is configured to run the computer program to execute the methods in the above embodiments of the present application.

[0149] The memory is connected to the processor, the memory may be a flash memory, a read-only memory or other memory, and the processor may be a central processing unit or a single-chip microcomputer.

[0150] In some embodiments, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the methods in the above embodiments of the present application are implemented.

[0151] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0152] It is understood that the above embodiments are merely exemplary embodiments for illustrating the principles of the present application, and the present application is not limited thereto. Those skilled in the art may make various modifications and improvements without departing from the spirit and substance of the present application, and such modifications and improvements are also considered to be within the scope of protection of the present application.

Claims

1. An intelligent SMS marketing carousel method, characterized in that: The method comprises: S1. Obtain multi-dimensional user data, including basic user attributes, historical interaction behavior data, and communication and social data; S2. Obtaining the user's dynamic interest vector through a deep learning model based on the user's multi-dimensional data; S3. Predicting the probability value of the user being interviewed by using an integrated learning gradient boosting tree (GBDT) algorithm based on the user's multi-dimensional data; S4. Based on the dynamic interest vector and the user's probability of being interviewed, determine multiple candidate marketing contents corresponding to each user from the marketing content library using a reinforcement learning algorithm; S5. Perform SMS marketing rotation for each user based on multiple candidate marketing contents corresponding to each user.

2. The intelligent SMS marketing carousel method according to claim 1, characterized in that: S2, including: Extracting user behavior sequences and spatiotemporal contexts based on the multi-dimensional user data; Based on the user behavior sequence and spatiotemporal context, feature extraction is performed through the Transformer encoder to obtain the user's long-term interests; Based on the user behavior sequence and spatiotemporal context, feature extraction is performed through the gated recurrent unit (GRU) network to obtain the user's short-term dynamics; The user's long-term interests and short-term dynamics are fused with features through the attention mechanism to obtain the user's dynamic interest vector.

3. The intelligent SMS marketing carousel method according to claim 1, characterized in that: S3, including: Screening, converting and encoding the user's multi-dimensional data to obtain key features of the user; Based on the key features of the user, the probability value of the user being easily interviewed is predicted by the integrated learning gradient boosting tree (GBDT) algorithm.

4. The intelligent SMS marketing carousel method according to claim 3, characterized in that: The user's multi-dimensional data is screened, converted, and encoded to obtain key user features, including: Performing one-hot encoding on the category data in the user multi-dimensional data and converting it into a numerical form that can be processed by a computer; Standardizing the numerical data in the user multi-dimensional data to obtain data in a standard format; The multi-dimensional data of the user is analyzed by an analysis algorithm, and features with low correlation with the user interview prediction are screened out and removed, while features with high correlation with the user interview prediction are retained.

5. The intelligent SMS marketing carousel method according to claim 1, characterized in that: S4, including: Based on the dynamic interest vector and the user's interview probability value, screening a first preset number of marketing contents from a marketing content library according to preset screening rules; Using a reinforcement learning algorithm, the first preset number of marketing contents are sorted with the goal of maximizing the reward to obtain a sorting result; A second preset number of candidate marketing contents corresponding to each user is obtained according to the sorting result, wherein the second preset number is smaller than the first preset number.

6. The intelligent SMS marketing carousel method according to claim 1, characterized in that: S5, including: Obtain the user's click-through rate (CTR) and conversion rate for each marketing carousel content; Adjust each user's marketing carousel content in real time based on the click-through rate (CTR) and conversion rate of each marketing carousel content.

7. The intelligent SMS marketing carousel method according to claim 1, characterized in that: S5, also includes at least one of the following: Parallel processing of marketing SMS sending tasks through multi-node distributed layout; Provide intelligent Do Not Disturb protection, SMS frequency control, and sensitive content filtering for each user; Get SMS marketing carousel result tracking data and display it in chart form.

8. An intelligent SMS marketing carousel device, characterized in that: The device comprises: A data acquisition module configured to acquire multi-dimensional user data, including basic user attributes, historical interaction behavior data, and communication and social data; A first processing module is configured to obtain a user's dynamic interest vector through a deep learning model based on the user's multi-dimensional data; A second processing module is configured to predict the user's interview probability value through an integrated learning gradient boosting tree (GBDT) algorithm based on the user's multi-dimensional data; a content determination module configured to determine, based on the dynamic interest vector and the user's probability of being interviewed, a plurality of candidate marketing contents corresponding to each user from a marketing content library using a reinforcement learning algorithm; The marketing carousel module is configured to perform SMS marketing carousel for users based on multiple candidate marketing contents corresponding to each user.

9. An intelligent SMS marketing carousel device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the intelligent SMS marketing carousel method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent SMS marketing carousel method according to any one of claims 1 to 7.

Images