Short message touch time determination method and device based on user portrait and storage medium

Abstract

The application provides a short message touch time determination method and device based on user portrait, equipment and storage medium, relates to the technical field of short message, and the method comprises the steps of constructing a multi-dimensional user portrait, classifying a to-be-sent short message task and extracting task features; inputting the multi-dimensional user portrait and the task features into a main prediction model to obtain a first opening rate prediction value of the to-be-sent short message of a user at least two time points, and inputting the multi-dimensional user portrait and the task features into a sub-prediction model determined according to a task type to obtain a second opening rate prediction value of the to-be-sent short message of the user at each time point; based on the target weight ratio, the confidence and the opening rate prediction value of the main prediction model and the sub-prediction model, generating a final opening rate prediction value of the to-be-sent short message of the user at each time point, and determining the short message touch time of the to-be-sent short message task. In the application, the short message sending time is accurately predicted and optimized, and the opening rate of the short message is improved.

Description

Technical Field

[0001] This invention relates to the field of SMS technology, and in particular to a method, apparatus, and storage medium for determining SMS delivery time based on user profiles. Background Technology

[0002] Current SMS sending tasks are generally conducted at fixed times or using random strategies, lacking personalization and intelligent considerations. This method ignores the differentiated responses of different users at different times under specific tasks, resulting in low SMS open rates, poor notification effectiveness, and wasted resources. Furthermore, operations require manually configuring SMS delivery times for each task, leading to high labor costs.

[0003] Optimizing SMS sending time decisions to improve user experience and SMS open rates has become a pressing technical problem in the industry. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and storage medium for determining SMS delivery time based on user profiles. It enables intelligent prediction of the open rate of SMS tasks to be sent by users at specific time points. Furthermore, it introduces a sub-prediction model for optimization based on the main prediction model to improve the accuracy and reliability of the model prediction. In turn, by accurately predicting and optimizing the SMS sending time, it improves the user open rate of SMS messages.

[0005] In a first aspect, the present invention provides a method for determining SMS delivery time based on user profiles, the method comprising the following steps: A multi-dimensional user profile is constructed based on user static attributes and dynamic behavioral characteristics. The task of sending SMS messages is classified and the task features of the SMS messages to be sent are extracted. The multidimensional user profile and the task features are input into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points. Then, a corresponding sub-prediction model is determined according to the task type of the SMS to be sent. The multidimensional user profile and the task features are input into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points. The main prediction model is a random forest or gradient boosting decision tree (GBDT). Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the user for the SMS to be sent at each of the time points is generated. Based on the predicted final open rate of the SMS message to be sent by the user at each of the time points, the SMS delivery time of the SMS message task to be sent is determined.

[0006] According to a method for determining SMS delivery time based on user profiles provided by the present invention, before generating the final open rate prediction value of the user for the SMS to be sent at each of the stated time points based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the method further includes: The initial weight ratio of the main prediction model and the sub-prediction model is calculated based on user behavior entropy and task user matching degree; the user behavior entropy is used to reflect the randomness of user behavior, and the higher the entropy value of the user behavior entropy, the more irregular the user behavior. The policy gradient reinforcement learning A2C algorithm is adopted, and the actual open rate of SMS messages is used as a reward signal to dynamically adjust the initial weight ratio, thereby obtaining the target weight ratio of the main prediction model and the sub-prediction model.

[0007] According to the present invention, a method for determining SMS delivery time based on user profiles, wherein determining the corresponding sub-prediction model according to the task type of the SMS to be sent includes: When the task type is a high-value decision-making task, the sub-prediction model is determined to be the attention long short-term memory recurrent neural network (LSTM) model. When the task type is an information notification task, the sub-prediction model is determined to be either the TextCNN text classification model based on a convolutional neural network or the Bidirectional Long Short-Term Memory (BiLSTM) model. When the task type is a periodic reminder task, the sub-prediction model is determined to be the Prophet model; In the case where the task type is a triggered interactive task, the sub-prediction model is determined to be a genetic neural network (GNN).

[0008] According to a method for determining SMS delivery time based on user profiles provided by the present invention, before generating the final open rate prediction value of the user for the SMS to be sent at each of the stated time points based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the method further includes: Determine the prediction variance of the main prediction model and the prediction variance of the sub-prediction model; The confidence level of the main prediction model is calculated based on the prediction variance of the main prediction model. The confidence level of the sub-prediction model is calculated based on the prediction variance of the sub-prediction model.

[0009] According to a method for determining SMS delivery time based on user profile provided by the present invention, the step of generating a final open rate prediction value for the user's SMS message at each of the specified time points, based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, includes: The predicted final open rate of the SMS message to be sent by the user at each of the time points is calculated using the following formula (1): (1) in, This represents the predicted final open rate of the SMS message to be sent by the user at each of the stated time points. This represents the weights of the main prediction model. This represents the predicted first open rate of the user for the SMS message to be sent at each of the time points, output by the main prediction model. This represents the confidence level of the main prediction model. This represents the weight of the sub-prediction model. This represents the second open rate prediction value of the user for the SMS message to be sent, output by the sub-prediction model at each of the stated time points. This represents the confidence level of the sub-prediction model.

[0010] According to the present invention, a method for determining SMS delivery time based on user profiles is provided, wherein the user's static attributes include at least one of the following: age, gender, and region; and the dynamic behavioral characteristics include at least one of the following: historical SMS opening conversion records, in-app behavior trajectory, distribution of active time of event tracking, historical active time periods, and historical active behavior sequences. The construction of a multi-dimensional user profile based on static user attributes and dynamic behavioral characteristics includes: Based on the user's static attributes and dynamic behavioral characteristics, data mining and machine learning techniques are used to construct a refined multidimensional user profile. The multidimensional user profile includes multiple dimensions of the user's interests, preferences, and behavioral patterns.

[0011] Secondly, the present invention also provides a device for determining SMS delivery time based on user profiles, the device comprising the following modules: The user profile building module is used to build a multi-dimensional user profile based on user static attributes and dynamic behavioral characteristics, classify the SMS task to be sent, and extract the task features of the SMS task to be sent. The prediction module is used to input the multi-dimensional user profile and the task features into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points, and to determine the corresponding sub-prediction model according to the task type of the SMS to be sent, and input the multi-dimensional user profile and the task features into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points; the main prediction model is a random forest or gradient boosting decision tree (GBDT); The determination module is used to generate the final open rate prediction value of the user for the SMS to be sent at each of the said time points based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value; and to determine the SMS delivery time of the SMS to be sent task based on the final open rate prediction value of the user for the SMS to be sent at each of the said time points.

[0012] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the SMS delivery time determination method based on user profile as described above.

[0013] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the SMS delivery time determination method based on user profile as described above.

[0014] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the SMS delivery time determination method based on user profile as described above.

[0015] The present invention provides a method, apparatus, device, and storage medium for determining SMS delivery time based on user profiles. First, a multi-dimensional user profile is constructed based on the user's static attributes and dynamic behavioral characteristics. The SMS delivery tasks are then classified and task features are extracted. Next, the multi-dimensional user profile and task features are input into a master prediction model to obtain first open rate predictions for the SMS messages to be sent at at least two time points. A corresponding sub-prediction model is determined based on the task type, and the multi-dimensional user profile and task features are input into the sub-prediction model to obtain second open rate predictions for the SMS messages to be sent at each time point. The master prediction model is either a random forest or a gradient boosting decision tree (GBDT). Further, based on the target weight ratio of the master and sub-prediction models, the confidence level of the master and sub-prediction models, the first open rate prediction, and the second open rate prediction, a final open rate prediction for the SMS messages to be sent at each time point is generated. Finally, based on the final open rate predictions for the SMS messages to be sent at each time point, the SMS delivery time for the SMS delivery task is determined.

[0016] In this invention, the main prediction model is a random forest or gradient boosting decision tree (GBDT). By inputting the constructed user profile and extracted task features into the main prediction model, the open rate of SMS messages to be sent by users at a specific time point is intelligently predicted. Furthermore, a sub-prediction model is introduced on the basis of the main prediction model for optimization to improve the accuracy and reliability of the model prediction. In turn, by accurately predicting and optimizing the SMS sending time, the user open rate of SMS messages is improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the method for determining SMS delivery time based on user profiles provided by this invention.

[0019] Figure 2 This is the second flowchart of the method for determining SMS delivery time based on user profiles provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the device for determining SMS delivery time based on user profiles provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first node can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0024] To more clearly understand the various embodiments provided by the present invention, the technical content involved in the present invention will first be described as follows: Optimizing SMS delivery timing decisions to improve user experience and notification effectiveness, and ultimately increase the return on investment (ROI) of SMS delivery, has become an urgent issue to be addressed.

[0025] Existing technologies typically employ a single machine learning prediction model to predict the open rate of SMS messages for specific tasks at different time points. However, a single model is difficult to adapt to different task characteristics (such as the timeliness of promotional SMS messages and the semantic importance of notification SMS messages). Some multi-model solutions use fixed weight fusion, which cannot be dynamically adjusted according to user behavior and task attributes, resulting in low open rates and conversion rates, and serious waste of resources.

[0026] Addressing the shortcomings of existing technologies, this invention aims to resolve the pain points of traditional SMS sending methods by constructing user profiles and combining them with task characteristics to achieve intelligent prediction of SMS sending time. Specifically, the technical problems to be solved include: how to comprehensively capture users' static and behavioral characteristics; how to accurately quantify task characteristics; how to construct an effective prediction model to estimate the open rate of specific tasks within a specific time period, thereby determining the optimal SMS sending time; how to achieve precise matching between task characteristics and model capabilities; how to optimize the multi-model fusion effect through a dynamic weighting mechanism; and how to maintain the long-term effectiveness of the model through closed-loop feedback.

[0027] The following is combined with Figures 1-4 The present invention describes a method, apparatus, device, and storage medium for determining SMS delivery time based on user profiles.

[0028] Figure 1 This is one of the flowcharts illustrating the method for determining SMS delivery time based on user profiles provided by this invention, such as... Figure 1 As shown, the method includes the following: Step 101: Construct a multi-dimensional user profile based on user static attributes and dynamic behavioral characteristics, classify the SMS sending tasks, and extract the task features of the SMS sending tasks.

[0029] It should be noted that the subject of this invention is an electronic device, used to accurately predict user open rates, and based on the prediction results, automatically select the optimal sending time to achieve personalized and intelligent SMS sending and improve user experience.

[0030] The technical approach of this invention includes: user profile construction, task scenario adaptation, intelligent prediction and optimization of user open rate, and multi-model fusion.

[0031] The user profile building process includes the following: building a multi-dimensional user profile based on the user's static attributes and dynamic behavioral characteristics.

[0032] User static attributes include age, gender, and region; dynamic behavioral characteristics include historical SMS opening conversion records, application activity habits, in-app behavior patterns, distribution of active time points, historical active time periods, and historical active behavior sequences.

[0033] Specifically, data mining and machine learning techniques are used to construct refined, multi-dimensional user profiles, which include multiple dimensions such as user interests and behavioral patterns, providing a basis for personalized prediction of SMS sending times.

[0034] The task scenario adaptation process includes the following: classifying the SMS sending task and extracting the task features of the SMS sending task.

[0035] A thorough analysis of the SMS sending task is conducted to extract key task characteristics such as timeliness, historical open rate, and conversion rate. These task characteristics reflect the task attributes and their impact on user open conversion. Based on user profiles and task characteristics, this embodiment uses machine learning or deep learning models to achieve personalized recommendation of SMS messages to be sent and the SMS delivery (sending) time.

[0036] Step 102: Input the multidimensional user profile and task features into the main prediction model to obtain the first open rate prediction value of the user's unsent SMS at at least two time points, and determine the corresponding sub-prediction model according to the task type. Input the multidimensional user profile and task features into the sub-prediction model to obtain the second open rate prediction value of the user's unsent SMS at each time point. The main prediction model is a random forest or gradient boosting decision tree (GBDT).

[0037] Among them, the main prediction model is, for example, random forest or gradient boosting decision tree. Random forest (RT) and gradient boosting decision tree (GBDT) are two mainstream ensemble learning algorithms, which are widely used in user profiling, recommendation systems, financial risk control and other fields.

[0038] Input data (multidimensional user profiles and task features) is fed into a random forest, where multiple trees are generated in parallel. Voting or averaging yields the predicted first open rate of the user's unsent SMS messages at at least two time points, i.e., the predicted initial open rate of the user's specific task at a specific time point.

[0039] Input data (multidimensional user profiles and task features) is fed into a gradient boosting tree, which is then sequentially generated. The fitted residuals and weighted sums are used to obtain the first open rate prediction for the user's unsent SMS messages at each time point. In other words, the initial open rate prediction for the user's specific task at a specific time point.

[0040] Furthermore, to further improve the accuracy of predictions, this invention introduces a sub-prediction model for optimization, while ensuring the accuracy and stability of the main prediction model. The sub-prediction model can be optimized for specific task types, further improving prediction accuracy by capturing subtle differences or specific patterns that the main prediction model might overlook.

[0041] The selection and construction of sub-prediction models require in-depth data analysis and model experiments tailored to the task type. Based on decision complexity and response patterns, SMS tasks to be sent are categorized into four types: high-value decision-making, information notification, periodic reminder, and trigger-based interaction. Core features such as timeliness, text semantics, and periodic patterns are extracted for each type.

[0042] Furthermore, by inputting multi-dimensional user profiles and task features into the sub-prediction model, the second open rate prediction value of the SMS messages to be sent is obtained at each time point. After processing the results of the main prediction model in the task dimension by the sub-prediction model, an optimal SMS sending time strategy can be formulated for each user and each task, thereby improving the overall notification effect.

[0043] Step 103: Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, generate the final open rate prediction value of the SMS to be sent for the user at each time point.

[0044] The target weight ratio of the main prediction model and the sub-prediction model is used to characterize the ratio of the importance of the main prediction model and the sub-prediction model in the final prediction, that is, the importance of the main prediction model and the importance of the sub-prediction model in the final prediction.

[0045] After obtaining the first open rate prediction value output by the main prediction model and the second open rate prediction value output by the sub-prediction model, the final open rate prediction value of the user's unsent SMS at each time point is obtained by using a weighted average with confidence (confidence is inversely proportional to the prediction variance).

[0046] Optionally, when the divergence between the main and sub-models exceeds a threshold, the Transformer arbitration model is activated to make a decision, inputting a more granular sequence of user behavior (such as the responses to the three most recent similar tasks) and outputting the final decision (the final predicted open rate).

[0047] By processing the results of the main prediction model at the task dimension through the sub-prediction model, an optimal SMS sending time strategy can be formulated for each user and each task, thereby improving the overall notification effect.

[0048] Step 104: Based on the predicted final open rate of the SMS messages to be sent at each time point, determine the SMS delivery time for the SMS message task to be sent.

[0049] Specifically, for example, the time point corresponding to the maximum value among the predicted final open rates of the SMS messages to be sent at each time point is determined as the SMS delivery time of the SMS task to be sent.

[0050] For example, user A's profile shows "active at night and sensitive to high discounts." The main prediction model recommends sending at 8:00 PM, while the sub-prediction model (Attention Long Short Term Memory (LSTM) recurrent neural network), which considers business activities and ends after 24 hours, recommends sending at 7:00 PM the next day, 3 hours before the end of the recommendation period. Due to high user behavior entropy, the sub-model has a weight of 50%, and the final fused recommendation is sent at 7:30 PM, with an actual open rate of 85%. The system then uses reinforcement learning to adjust the weight of the sub-model for this type of user to 55%.

[0051] The method provided in this embodiment first constructs a multi-dimensional user profile based on the user's static attributes and dynamic behavioral characteristics, classifies the SMS sending task, and extracts the task features of the SMS sending task. Then, the multi-dimensional user profile and task features are input into the main prediction model to obtain the first open rate prediction value of the SMS to be sent at at least two time points. The corresponding sub-prediction model is determined according to the task type, and the multi-dimensional user profile and task features are input into the sub-prediction model to obtain the second open rate prediction value of the SMS to be sent at each time point. The main prediction model is a random forest or gradient boosting decision tree (GBDT). Further, based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the SMS to be sent at each time point is generated. Finally, based on the final open rate prediction value of the SMS to be sent at each time point, the SMS delivery time of the SMS task to be sent is determined.

[0052] In this invention, the main prediction model is a random forest or gradient boosting decision tree (GBDT). By inputting the constructed user profile and extracted task features into the main prediction model, intelligent prediction of the open rate of SMS tasks to be sent at a specific time point is achieved. Furthermore, a sub-prediction model is introduced on the basis of the main prediction model for optimization to improve the accuracy and reliability of the model prediction. In this way, by accurately predicting and optimizing the SMS sending time, the user open rate can be improved.

[0053] It should be noted that each implementation method of this application can be freely combined, rearranged, or executed individually, and does not need to rely on or depend on a fixed execution order.

[0054] According to the method for determining SMS delivery time based on user profile provided by the present invention, before generating the final open rate prediction value of the SMS to be sent at each time point based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the method further includes: The initial weight ratio of the main prediction model and the sub-prediction model is calculated based on user behavior entropy and task user matching degree. User behavior entropy is used to reflect the randomness of user behavior. The higher the entropy value of user behavior, the more irregular the user behavior. The policy gradient reinforcement learning A2C algorithm is adopted, and the actual open rate of SMS messages is used as a reward signal to dynamically adjust the initial weight ratio, thereby obtaining the target weight ratio of the main prediction model and the sub-prediction model.

[0055] Specifically, in some embodiments, before calculating the final predicted open rate, the target weight ratio of the main prediction model and the sub-prediction model is also calculated. An example of this process is as follows: First, the initial weight ratio of the main prediction model and the sub-prediction model is calculated based on user behavior entropy and task user matching degree.

[0056] This embodiment flexibly allocates the initial weight ratio of the main prediction model and the sub-prediction model based on the regularity of user behavior and the degree of matching between the task and the user.

[0057] Specifically, user behavior entropy reflects the randomness of user behavior; a higher entropy value indicates less predictable patterns in user behavior. The more random the user behavior, the higher the weight of the sub-prediction model.

[0058] For users with highly random behavior, the initial weight of the sub-prediction model is increased to 40%-50%, as these users require more reliance on individual patterns for prediction. Conversely, for users with low behavioral entropy and strong behavioral patterns, the initial weight of the main prediction model is increased to 60%-70%, where leveraging group patterns yields better prediction results. Furthermore, when the task-user match rate reaches 0.8 or higher, the weight of the sub-model is increased by an additional 10% to enhance targeted prediction for that task. Additionally, a forgetting factor can be introduced to reduce the impact of stale data.

[0059] Furthermore, the Advantage Actor-Critic (A2C) algorithm is adopted, and the initial weight ratio is dynamically adjusted with the actual open rate of the SMS as a reward signal, so as to obtain the target weight ratio of the main prediction model and the sub-prediction model.

[0060] For example, the actual open rate and conversion rate of SMS messages are used as the basis for judgment, and the initial weight ratio is continuously and dynamically updated. When the prediction error of the sub-prediction model is smaller than that of the main prediction model for three consecutive times, the weight of the sub-prediction model will be increased by 5%-10%.

[0061] Meanwhile, in order to adapt to changes in user behavior over time, the system introduces a forgetting factor of 0.9-0.95 to reduce the impact of outdated data, such as data from 30 days ago, on the weights.

[0062] The method provided in this embodiment first calculates the initial weight ratio of the main prediction model and the sub-prediction model based on user behavior entropy and task-user matching degree. User behavior entropy reflects the randomness of user behavior; the higher the entropy value, the more irregular the user behavior. Then, the policy gradient reinforcement learning A2C algorithm is used, with the actual SMS open rate as the reward signal, to dynamically adjust the initial weight ratio, thereby obtaining the target weight ratio of the main prediction model and the sub-prediction model. Next, the output results of the main prediction model and the sub-prediction model are integrated using a confidence-weighted average method to obtain the final open rate prediction value. Introducing the sub-model for optimization based on the main model improves the accuracy and reliability of the prediction. Furthermore, based on the multi-model fusion results, an optimal SMS sending time strategy can be formulated for each user and each task, improving user experience and significantly increasing the user open rate.

[0063] According to the present invention, a method for determining SMS delivery time based on user profiles is provided, which determines a corresponding sub-prediction model according to the task type, including: When the task type is a high-value decision-making task, the sub-prediction model is determined to be the attention long short-term memory recurrent neural network (LSTM) model. When the task type is information notification task, the sub-prediction model is determined to be either the TextCNN text classification model based on convolutional neural network or the Bi-directional long short-term memory network BiLSTM model. When the task type is a periodic reminder task, the sub-prediction model is determined to be the Prophet model; In the case of a trigger-based interactive task, the sub-prediction model is determined to be a genetic neural network (GNN).

[0064] Specifically, in some embodiments, the process of determining the sub-prediction model in step 102 is as follows: It should be noted that, according to decision complexity and response mode, the task types in this embodiment can be divided into high-value decision-making tasks, information notification tasks, periodic reminder tasks, and trigger-based interactive tasks.

[0065] For high-value decision-making tasks, the sub-prediction model uses attention LSTM to capture temporal dependencies.

[0066] For information notification tasks (such as bill reminders), the sub-prediction model uses a text classification model based on a convolutional neural network (TextCNN) and a bidirectional long short-term memory (BiLSTM) network to enhance text semantics. The semantic vector of the input task text and the user's historical response speed to notification SMS messages are used to strengthen the impact of text urgency on opening time.

[0067] For periodic reminder tasks, the sub-prediction model uses the Prophet model to match the user's biological clock. It takes the user's historical response cycle and task cycle characteristics as input to predict the degree of fit between the user's biological clock and the reminder time. Prophet is an open-source time series forecasting algorithm.

[0068] For triggered interactive events (such as questionnaire invitations), the sub-prediction model uses a genetic neural network (GNN) to associate preceding behaviors, modeling the user's preceding behaviors (such as clicking on an activity link) and opening SMS messages as node relationships, capturing the temporal dependencies of the behavior chain.

[0069] The method provided in this embodiment not only focuses on the user's basic attributes and behavioral patterns, but also deeply analyzes the specific characteristics of tasks in various scenarios, accurately reflects their impact on user decisions, and achieves precise matching between users and tasks in two dimensions. Furthermore, it classifies task types into four categories according to decision complexity and response mode: high-value decision-making, information notification, periodic reminder, and trigger-based interaction. For different task types, different sub-prediction models are used to extract core features such as timeliness, text semantics, and periodic patterns, which facilitates the introduction of sub-prediction models for optimization on the basis of the main prediction model, thereby improving the accuracy and reliability of the prediction. Based on the results of multi-model fusion, it can formulate the optimal SMS sending time strategy for each user and each task.

[0070] According to the method for determining SMS delivery time based on user profile provided by the present invention, before generating the final open rate prediction value of the SMS to be sent at each time point based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the method further includes: Determine the prediction variance of the main prediction model and the prediction variance of the sub-prediction models; The confidence level of the main prediction model is calculated based on the prediction variance of the main prediction model. The confidence level of the sub-prediction model is calculated based on the prediction variance of the sub-prediction model.

[0071] Specifically, in some embodiments, before step 103 calculates the predicted final open rate of the SMS message to be sent at each time point, it also includes the calculation of the confidence level of the main prediction model and the sub-prediction model. The higher the confidence level, the more stable and reliable the prediction results of the model are, and the greater the role it plays in the final prediction.

[0072] The process of calculating confidence level is illustrated below: First, calculate the prediction variance of the main prediction model and the prediction variance of the sub-prediction models. Model prediction variance is a core metric for measuring model stability in machine learning. It reflects the degree of fluctuation in the model's output when the same input data undergoes small changes; physically, it represents the dispersion of the model's prediction results across multiple training iterations.

[0073] Furthermore, the confidence level of the main prediction model is calculated based on the prediction variance of the main prediction model, and the confidence level of the sub-prediction model is calculated based on the prediction variance of the sub-prediction model.

[0074] In this embodiment, the confidence level is inversely proportional to the prediction variance, that is, confidence level = 1 / prediction variance. The lower the variance, the higher the weight. In other words, the smaller the variance of the prediction result, the more reliable the prediction, the higher the confidence level, and the greater the weight it occupies in the weighted average.

[0075] The method provided in this embodiment calculates the confidence level of the main prediction model based on the prediction variance of the main prediction model. The confidence level ensures the reliability of the model prediction. Then, the output results of the main prediction model and the sub-prediction model are integrated based on the weighted average method with confidence level to obtain the final open rate prediction value of the user's unsent SMS at each time point, thereby improving the prediction accuracy.

[0076] According to the present invention, a method for determining SMS delivery time based on user profiles generates the final open rate prediction value of the SMS to be sent by the user at each time point based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value. The method includes: The predicted final open rate of the SMS messages to be sent at each time point is calculated using the following formula (1): (1) in, This represents the predicted final open rate of text messages to be sent at each point in time. Indicates the weights of the main prediction model. This represents the predicted first open rate of the SMS messages to be sent for each user at each time point, as output by the main prediction model. The symbol represents the confidence level of the main prediction model. This represents the weights of the sub-prediction model. This represents the second open rate prediction value for users at each time point output by the sub-prediction model. This indicates the confidence level of the sub-prediction model.

[0077] Specifically, in some embodiments, step 103 is implemented in the following manner: By integrating the outputs of the main prediction model and the sub-prediction model using a weighted average method with confidence levels, the final open rate prediction value of the SMS messages to be sent is obtained at each time point.

[0078] The predicted final open rate of the SMS messages to be sent at each time point is calculated using the following formula (1): (1) in, This represents the predicted final open rate of the SMS message to be sent at each point in time. It is the predicted open rate of the SMS message for a specific task at a specific point in time obtained by combining the prediction results of the main model and the sub-model. This value is the key basis for determining the optimal SMS sending time. This represents the weight of the main prediction model. This weight is determined according to a dynamic weight allocation mechanism and will be dynamically adjusted as user behavior and task characteristics change, reflecting the proportion of the main model in the final prediction. This represents the predicted first open rate of the SMS message to be sent for the user at each time point, as output by the main prediction model. In other words, it is the result calculated by the main model based on user profile and task characteristics. This indicates the confidence level of the main prediction model. The more stable and reliable the prediction results of the main model, the higher the confidence level, and the greater its role in the final prediction. The weights of the sub-prediction models are determined by a dynamic weight allocation mechanism, which is adjusted according to different task types and user characteristics, reflecting the importance of the sub-models in the final prediction. This represents the second open rate prediction value of the user's unsent SMS messages at each time point output by the sub-prediction model. In other words, it is the prediction result calculated by the sub-model for a specific task type, combined with the user profile. This represents the confidence level of the sub-prediction model, used to measure the reliability of the sub-model's prediction results.

[0079] The method provided in this embodiment integrates the outputs of the main prediction model and the sub-prediction model using a weighted average method with confidence levels to obtain the final open rate prediction. By introducing the sub-model for optimization based on the main model, the accuracy and reliability of the prediction are improved. Furthermore, based on the multi-model fusion results, an optimal SMS sending time strategy can be formulated for each user and each task, improving user experience, significantly increasing the user open rate, and adaptively generating estimated times for different scenarios, thereby improving the overall return on investment (ROI).

[0080] According to the present invention, a method for determining SMS delivery time based on user profile is provided, wherein the user's static attributes include at least one of the following: age, gender, and region; and the dynamic behavioral characteristics include at least one of the following: historical SMS opening conversion records, in-application behavior trajectory, distribution of active time of event tracking, historical active time period, and historical active behavior sequence. Multidimensional user profiles are constructed based on static user attributes and dynamic behavioral characteristics, including: Based on users' static attributes and dynamic behavioral characteristics, data mining and machine learning techniques are used to construct refined multidimensional user profiles. Among them, the multi-dimensional user profile includes multiple dimensions of user interests, preferences, and behavioral patterns.

[0081] Specifically, in some embodiments, the static attributes of users used to construct user profiles include at least one of the following: age, gender, and region; dynamic behavioral characteristics include at least one of the following: historical SMS opening conversion records, in-app behavior trajectory, distribution of active time of event tracking, historical active time periods, and historical active behavior sequences.

[0082] The historical SMS conversion records document the user's behavioral conversion path after receiving an SMS / notification. The analysis includes: SMS delivery rate → open rate → click-through rate → final conversion rate (e.g., payment, registration); in-app behavior trajectory visualizes the complete operation path of a user's single app usage; the distribution of active time points reveals the time patterns of user behavior statistically analyzed through tracking points; the difference between historical active time periods and tracked activity lies in focusing on long-term patterns rather than single behaviors; historical active behavior sequences are essentially the DNA of user behavior patterns across time. For example, historical SMS conversion records show a 40% increase in the open rate of discount SMS pushed at 19:00, and the behavior trajectory reveals that users often browse > compare prices > favorite during evening peak hours; the active time distribution confirms 20:00 as the peak time for orders; historical active time periods can be segmented into "evening active user groups"; and the behavior sequence identifies the "order after browsing 3 times" pattern.

[0083] The multidimensional user profile includes multiple dimensions of user interests and behavioral patterns. Interests reflect what users like, including content interests and product preferences; behavioral patterns reveal how users act, such as identifying user time patterns (e.g., active between 8-10 PM, 10 visits of less than 5 minutes per day, and a 300% increase in activity on weekends), path patterns, and consumption behavior DNA (decision speed, payment methods), etc.

[0084] The implementation process of step 101 includes the following steps: (1) Static feature vectorization Gender → [0,1] (One-Hot) Geographic level → Embedding (2) Deep processing of dynamic behavior

[0085] (3) Model layer Machine learning techniques are used to generate user tags, which include interests, preferences, and behavioral patterns.

[0086] The method provided in this embodiment constructs a refined multi-dimensional user profile based on user static attributes and dynamic behavioral characteristics, using data mining and machine learning techniques. The multi-dimensional user profile includes multiple dimensions of user interests and behavioral patterns, which facilitates subsequent prediction of the final open rate of SMS messages using the main prediction model and sub-prediction models in combination with task characteristics, thereby optimizing SMS sending time and improving the user open rate of SMS messages.

[0087] Optionally, the prediction model can be continuously optimized based on the actual response (such as open rate, conversion rate, etc.) after each task SMS is sent, to ensure that the model can automatically adapt to changes in user behavior and adjustments to task characteristics.

[0088] Figure 2 This is the second flowchart illustrating the method for determining SMS delivery time based on user profiles provided by this invention. Figure 2 As shown, the method includes the following steps: Step 201: Based on users' static attributes and dynamic behavioral characteristics, use data mining and machine learning techniques to construct a refined multidimensional user profile; the multidimensional user profile includes multiple dimensions of users' interests, preferences, and behavioral patterns. Step 202: Calculate the initial weight ratio of the main prediction model and the sub-prediction model based on user behavior entropy and task user matching degree; user behavior entropy is used to reflect the randomness of user behavior. The higher the entropy value of user behavior, the more irregular the user behavior. Step 203: Use the policy gradient reinforcement learning A2C algorithm, with the actual open rate of SMS messages as the reward signal, to dynamically adjust the initial weight ratio, thereby obtaining the target weight ratio of the main prediction model and the sub-prediction model; Step 204: Determine the prediction variance of the main prediction model and the prediction variance of the sub-prediction models; Step 205: Calculate the confidence level of the main prediction model based on the prediction variance of the main prediction model, and calculate the confidence level of the sub-prediction model based on the prediction variance of the sub-prediction model. Step 206: When the task type is a high-value decision-making task, determine the sub-prediction model as an LSTM model; when the task type is an information notification task, determine the sub-prediction model as a TextCNN or BiLSTM model; when the task type is a periodic reminder task, determine the sub-prediction model as a Prophet model; when the task type is a trigger-based interactive task, determine the sub-prediction model as a GNN model. Step 207: Input the multi-dimensional user profile and task features into the main prediction model to obtain the first open rate prediction value of the user's unsent SMS at at least two time points, and determine the corresponding sub-prediction model according to the task type. Input the multi-dimensional user profile and task features into the sub-prediction model to obtain the second open rate prediction value of the user's unsent SMS at each time point. Step 208: Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, generate the final open rate prediction value of the SMS to be sent for the user at each time point. Step 209: Determine the SMS delivery time for the SMS task to be sent based on the predicted final open rate of the SMS to be sent at each time point.

[0089] The method provided in this embodiment, for a text message task to be sent, uses machine learning algorithms and deep learning algorithms (such as random forest, gradient boosting tree, LSTM, etc.) based on user profiles and task characteristics, combined with historical text message sending records and corresponding user feedback data (such as whether the message was opened, whether it was converted, etc.), to construct a predictive model that can predict the open rate of text messages for a specific task at different time points. The time point with the highest expected ROI is selected as the optimal sending time. Through cross-validation and parameter tuning, the accuracy and stability of the model are ensured, and personalized optimization of text message sending time is achieved.

[0090] The following describes the SMS delivery time determination device based on user profile provided by the present invention. The SMS delivery time determination device based on user profile described below and the SMS delivery time determination method based on user profile described above can be referred to in correspondence.

[0091] Figure 3 This is a schematic diagram of the SMS delivery time determination device based on user profile provided by the present invention, as shown below. Figure 3 As shown, the SMS delivery time determination device 300 based on user profile includes the following modules: User profile building module 310 is used to build a multi-dimensional user profile based on user static attributes and dynamic behavioral characteristics, classify the SMS task to be sent, and extract the task features of the SMS task to be sent. The prediction module 320 is used to input the multi-dimensional user profile and the task features into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points, and to determine the corresponding sub-prediction model according to the task type of the SMS to be sent, and input the multi-dimensional user profile and the task features into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points; the main prediction model is a random forest or gradient boosting decision tree (GBDT); The determining module 330 is used to generate the final open rate prediction value of the user for the SMS to be sent at each of the time points based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value; and to determine the SMS delivery time of the SMS to be sent task based on the final open rate prediction value of the user for the SMS to be sent at each of the time points.

[0092] The apparatus provided in this embodiment firstly, the user profile construction module 310 constructs a multi-dimensional user profile based on the user's static attributes and dynamic behavioral characteristics, classifies the SMS task to be sent, and extracts the task features of the SMS task to be sent; then, the prediction module 320 is used to input the multi-dimensional user profile and task features into the main prediction model to obtain the first open rate prediction value of the SMS to be sent at at least two time points, and determines the corresponding sub-prediction model according to the task type, inputs the multi-dimensional user profile and task features into the sub-prediction model to obtain the second open rate prediction value of the SMS to be sent at each time point, wherein the main prediction model is a random forest or gradient boosting decision tree GBDT; further, the determination module 330 is used to generate the final open rate prediction value of the SMS to be sent at each time point based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value; then, based on the final open rate prediction value of the SMS to be sent at each time point, the SMS delivery time of the SMS task to be sent is determined.

[0093] In this invention, the main prediction model is a random forest or gradient boosting decision tree (GBDT). By inputting the constructed user profile and extracted task features into the main prediction model, the open rate of SMS messages to be sent by users at a specific time point is intelligently predicted. Furthermore, a sub-prediction model is introduced on the basis of the main prediction model for optimization to improve the accuracy and reliability of the model prediction. In turn, by accurately predicting and optimizing the SMS sending time, the user open rate of SMS messages is improved.

[0094] According to the present invention, a user profile-based SMS delivery time determination device 300 is provided, wherein the prediction module 320 is further configured to: The initial weight ratio of the main prediction model and the sub-prediction model is calculated based on user behavior entropy and task user matching degree; the user behavior entropy is used to reflect the randomness of user behavior, and the higher the entropy value of the user behavior entropy, the more irregular the user behavior. The policy gradient reinforcement learning A2C algorithm is adopted, and the actual open rate of SMS messages is used as a reward signal to dynamically adjust the initial weight ratio, thereby obtaining the target weight ratio of the main prediction model and the sub-prediction model.

[0095] According to the present invention, a device 300 for determining SMS delivery time based on user profile is provided, wherein the prediction module 320 is specifically used for: When the task type is a high-value decision-making task, the sub-prediction model is determined to be the attention long short-term memory recurrent neural network (LSTM) model. When the task type is an information notification task, the sub-prediction model is determined to be either TextCNN, a text classification model based on a convolutional neural network, or Bidirectional Long Short-Term Memory (BiLSTM) network. When the task type is a periodic reminder task, the sub-prediction model is determined to be the Prophet model; In the case where the task type is a triggered interactive task, the sub-prediction model is determined to be a genetic neural network (GNN).

[0096] According to the present invention, a user profile-based SMS delivery time determination device 300 is provided, wherein the prediction module 320 is further configured to: Determine the prediction variance of the main prediction model and the prediction variance of the sub-prediction model; The confidence level of the main prediction model is calculated based on the prediction variance of the main prediction model. The confidence level of the sub-prediction model is calculated based on the prediction variance of the sub-prediction model.

[0097] According to the present invention, a device 300 for determining SMS delivery time based on user profile is provided, wherein the determining module 330 is specifically used for: The predicted final open rate of the SMS message to be sent by the user at each of the time points is calculated using the following formula (1): (1) in, This represents the predicted final open rate of the SMS message to be sent by the user at each of the stated time points. This represents the weights of the main prediction model. This represents the predicted first open rate of the user for the SMS message to be sent at each of the time points, output by the main prediction model. This represents the confidence level of the main prediction model. This represents the weight of the sub-prediction model. This represents the second open rate prediction value of the user for the SMS message to be sent, output by the sub-prediction model at each of the stated time points. This represents the confidence level of the sub-prediction model.

[0098] According to the present invention, a user profile-based SMS delivery time determination device 300 is provided, wherein the user static attributes include at least one of the following: age, gender, and region; and the dynamic behavioral characteristics include at least one of the following: historical SMS opening conversion records, in-application behavior trajectory, distribution of active time of data tracking, historical active time periods, and historical active behavior sequences. The user profile building module 310 is specifically used for: Based on the user's static attributes and dynamic behavioral characteristics, data mining and machine learning techniques are used to construct a refined multidimensional user profile. The multidimensional user profile includes multiple dimensions of the user's interests, preferences, and behavioral patterns.

[0099] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a method for determining SMS delivery time based on user profiles, the method including: A multi-dimensional user profile is constructed based on user static attributes and dynamic behavioral characteristics. The task of sending SMS messages is classified and the task features of the SMS messages to be sent are extracted. The multidimensional user profile and the task features are input into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points. Then, a corresponding sub-prediction model is determined according to the task type of the SMS to be sent. The multidimensional user profile and the task features are input into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points. The main prediction model is a random forest or gradient boosting decision tree (GBDT). Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the user for the SMS to be sent at each of the time points is generated. Based on the predicted final open rate of the SMS message to be sent by the user at each of the time points, the SMS delivery time of the SMS message task to be sent is determined.

[0100] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the SMS delivery time determination method based on user profile provided by the above methods, the method comprising: A multi-dimensional user profile is constructed based on user static attributes and dynamic behavioral characteristics. The task of sending SMS messages is classified and the task features of the SMS messages to be sent are extracted. The multidimensional user profile and the task features are input into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points. Then, a corresponding sub-prediction model is determined according to the task type of the SMS to be sent. The multidimensional user profile and the task features are input into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points. The main prediction model is a random forest or gradient boosting decision tree (GBDT). Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the user for the SMS to be sent at each of the time points is generated. Based on the predicted final open rate of the SMS message to be sent by the user at each of the time points, the SMS delivery time of the SMS message task to be sent is determined.

[0102] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for determining SMS delivery time based on user profiles provided by the above methods, the method comprising: A multi-dimensional user profile is constructed based on user static attributes and dynamic behavioral characteristics. The task of sending SMS messages is classified and the task features of the SMS messages to be sent are extracted. The multidimensional user profile and the task features are input into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points. Then, a corresponding sub-prediction model is determined according to the task type of the SMS to be sent. The multidimensional user profile and the task features are input into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points. The main prediction model is a random forest or gradient boosting decision tree (GBDT). Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the user for the SMS to be sent at each of the time points is generated. Based on the predicted final open rate of the SMS message to be sent by the user at each of the time points, the SMS delivery time of the SMS message task to be sent is determined.

[0103] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining SMS delivery time based on user profiles, characterized in that, include: A multi-dimensional user profile is constructed based on user static attributes and dynamic behavioral characteristics. The task of sending SMS messages is classified and the task features of the SMS messages to be sent are extracted. The multidimensional user profile and the task features are input into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points. Then, a corresponding sub-prediction model is determined according to the task type of the SMS to be sent. The multidimensional user profile and the task features are input into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points. The main prediction model is a random forest or gradient boosting decision tree (GBDT). Based on user behavior entropy and task user matching degree, the initial weight ratio of the main prediction model and the sub-prediction model is calculated; The user behavior entropy is used to reflect the randomness of user behavior. The higher the entropy value of the user behavior entropy, the more irregular the user behavior is. The policy gradient reinforcement learning A2C algorithm is adopted, and the actual open rate of SMS messages is used as a reward signal to dynamically adjust the initial weight ratio, thereby obtaining the target weight ratio of the main prediction model and the sub-prediction model. Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the user for the SMS to be sent at each of the time points is generated. Based on the predicted final open rate of the SMS message to be sent by the user at each of the time points, the SMS delivery time of the SMS message task to be sent is determined.

2. The method for determining SMS delivery time based on user profiles according to claim 1, characterized in that, The step of determining the corresponding sub-prediction model based on the task type of the SMS message to be sent includes: When the task type is a high-value decision-making task, the sub-prediction model is determined to be the attention long short-term memory recurrent neural network (LSTM) model. When the task type is an information notification task, the sub-prediction model is determined to be either TextCNN, a text classification model based on a convolutional neural network, or Bidirectional Long Short-Term Memory (BiLSTM) network. When the task type is a periodic reminder task, the sub-prediction model is determined to be the Prophet model; In the case where the task type is a triggered interactive task, the sub-prediction model is determined to be a genetic neural network (GNN).

3. The method for determining SMS delivery time based on user profiles according to claim 1, characterized in that, Before generating the final open rate prediction value for the user's SMS message at each of the stated time points based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the method further includes: Determine the prediction variance of the main prediction model and the prediction variance of the sub-prediction model; The confidence level of the main prediction model is calculated based on the prediction variance of the main prediction model. The confidence level of the sub-prediction model is calculated based on the prediction variance of the sub-prediction model.

4. The method for determining SMS delivery time based on user profiles according to claim 1, characterized in that, The step of generating the final open rate prediction value for the user's SMS message at each of the aforementioned time points, based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, includes: The predicted final open rate of the SMS message to be sent by the user at each of the time points is calculated using the following formula (1): （1） in, This represents the predicted final open rate of the SMS message to be sent by the user at each of the stated time points. This represents the weights of the main prediction model. This represents the predicted first open rate of the user for the SMS message to be sent at each of the time points, output by the main prediction model. This represents the confidence level of the main prediction model. This represents the weight of the sub-prediction model. This represents the second open rate prediction value of the user for the SMS message to be sent, output by the sub-prediction model at each of the stated time points. This represents the confidence level of the sub-prediction model.

5. The method for determining SMS delivery time based on user profiles according to any one of claims 1-4, characterized in that, The user's static attributes include at least one of the following: age, gender, and region; the dynamic behavioral characteristics include at least one of the following: historical SMS opening conversion records, in-app behavior trajectory, distribution of active time of event tracking, historical active time periods, and historical active behavior sequences. The construction of a multi-dimensional user profile based on static user attributes and dynamic behavioral characteristics includes: Based on the user's static attributes and dynamic behavioral characteristics, data mining and machine learning techniques are used to construct a refined multidimensional user profile. The multidimensional user profile includes multiple dimensions of the user's interests, preferences, and behavioral patterns.

6. A device for determining SMS delivery time based on user profile, characterized in that, include: The user profile building module is used to build a multi-dimensional user profile based on user static attributes and dynamic behavioral characteristics, classify the SMS task to be sent, and extract the task features of the SMS task to be sent. The prediction module is used to input the multi-dimensional user profile and the task features into the main prediction model to obtain the first open rate prediction value of the user for the SMS to be sent at at least two time points, and to determine the corresponding sub-prediction model according to the task type of the SMS to be sent, and input the multi-dimensional user profile and the task features into the sub-prediction model to obtain the second open rate prediction value of the user for the SMS to be sent at each of the time points; the main prediction model is a random forest or gradient boosting decision tree (GBDT); The determination module is used to calculate the initial weight ratio of the main prediction model and the sub-prediction model based on user behavior entropy and task user matching degree. The user behavior entropy is used to reflect the randomness of user behavior. The higher the entropy value, the less regular the user behavior. The policy gradient reinforcement learning A2C algorithm is adopted, and the actual opening rate of SMS messages is used as a reward signal to dynamically adjust the initial weight ratio to obtain the target weight ratio of the main prediction model and the sub-prediction model. Based on the target weight ratio of the main prediction model and the sub-prediction model, the confidence level of the main prediction model, the confidence level of the sub-prediction model, the first open rate prediction value, and the second open rate prediction value, the final open rate prediction value of the user for the SMS to be sent at each of the time points is generated. Based on the predicted final open rate of the SMS message to be sent by the user at each of the time points, the SMS delivery time of the SMS message task to be sent is determined.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the SMS delivery time determination method based on user profile as described in any one of claims 1 to 5.

8. A non-transitory 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 SMS delivery time determination method based on user profile as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the SMS delivery time determination method based on user profile as described in any one of claims 1 to 5.

Images