Short message channel intelligent recommendation system and method established based on intelligent marketing system

Abstract

The invention relates to the field of short message channel optimization, and discloses a short message channel intelligent recommendation system and method established based on an intelligent marketing system, and the short message channel intelligent recommendation system established based on the intelligent marketing system comprises a data collection module, a feature extraction module, a multi-objective optimization module, a recommendation algorithm module and a feedback learning module. The short message channel intelligent recommendation method established based on the intelligent marketing system comprises the following steps: through data acquisition, feature extraction, multi-objective optimization, a recommendation algorithm and a feedback learning module, precise evaluation, dynamic recommendation and adaptive optimization of a short message channel are realized, the intelligent level of channel selection is improved, and the short message delivery effect is improved. According to the method, multi-objective optimization, feature extraction, feedback learning and recommendation algorithms are combined with multiple optimization methods, accurate evaluation, real-time adjustment and intelligent recommendation of short message channels are achieved, and compared with a traditional scheme, the adaptability and stability of channel selection and the marketing putting effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of SMS channel optimization, and specifically to an SMS channel intelligent recommendation system and method based on an intelligent marketing system. Background Art

[0002] In the existing SMS channel recommendation technologies, most solutions rely on fixed rules or simple scoring mechanisms, usually setting thresholds based on historical statistical data to screen each SMS channel; however, this approach has significant limitations - although historical data can reflect the long-term performance of the channel, it cannot adapt to the dynamic changes of the SMS channel status in real time; once the channel quality fluctuates, such as a decrease in the delivery rate or an increase in latency, the static scoring mechanism often cannot adjust in time, resulting in the recommended channel may no longer be the optimal choice, affecting the actual delivery effect of marketing.

[0003] In addition, there are obvious deficiencies in the feature extraction and evaluation methods of the existing technologies for SMS channels; most methods only use basic metrics, such as the delivery rate and cost, but ignore the multi-dimensional factors of channel quality; for example, different business scenarios have different sensitivities to metrics such as conversion rate and latency, and traditional methods cannot perform personalized optimization for specific requirements; this makes the adaptability of channel recommendation poor, especially in high-demand precision marketing scenarios, it is difficult to obtain the best combination of SMS channels.

[0004] Furthermore, traditional SMS channel recommendation schemes lack an effective feedback adjustment mechanism; most systems only perform one-time optimization at the initial stage of recommendation, and the subsequent recommendation strategies are basically fixed, and do not fully utilize the actual behavior data of users for adaptive adjustment; this results in the system being unable to continuously optimize the decision-making logic, especially when the quality of the SMS channel changes over time, the recommendation effect often gradually decreases over time; without a feedback learning mechanism, the recommendation system is like "one-way thinking" and cannot effectively respond to changes in the environment, ultimately affecting the overall delivery quality of SMS marketing; therefore, the present invention proposes an SMS channel intelligent recommendation system and method based on an intelligent marketing system to solve the deficiencies of the existing technologies. Summary of the Invention

[0005] In view of the deficiencies of the existing technologies, the present invention provides an SMS channel intelligent recommendation system and method based on an intelligent marketing system, which solves the problems of difficult real-time optimization of SMS channel selection, insufficient feature extraction dimensions, and lack of adaptive adjustment of recommendation strategies, ensures the accurate recommendation and dynamic optimization of SMS channels, and improves the marketing delivery effect.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: An SMS channel intelligent recommendation system based on an intelligent marketing system, comprising: A data collection module, which is used to collect historical data and real-time data of each SMS channel, and transmit the data to the feature extraction module and the multi-objective optimization module; A feature extraction module, which is used to extract key features of each SMS channel from the data and generate feature vectors, and transmit the feature vectors to the multi-objective optimization module for comprehensively evaluating the effectiveness of each channel; A multi-objective optimization module, which is used to calculate the comprehensive score of each channel according to the feature vectors, and combine with a multi-objective optimization algorithm. The multi-objective optimization module generates an optimal SMS channel combination through interaction with the recommendation algorithm module; A recommendation algorithm module, which is used to generate a recommended list of SMS channels based on the output of the multi-objective optimization module and real-time feedback data, and feedback the result to the feedback learning module to optimize the recommendation strategy; A feedback learning module, which is used to dynamically adjust the strategy of the recommendation algorithm module according to the real-time marketing effect and user feedback data.

[0007] Preferably, the data collection module includes: A historical data collection unit, which is used to collect the delivery rate, conversion rate, delay, and cost metrics of each channel in historical marketing activities; a real-time data collection unit, which is used to collect the status information of the SMS channel in real time, including delivery situation, failure rate, and user feedback.

[0008] Preferably, the feature extraction module includes: A feature analysis unit, which is used to analyze the historical data of each channel, extract key factors affecting the marketing effect, and generate channel feature vectors.

[0009] Preferably, the multi-objective optimization module includes: An objective function unit, which is used to construct and calculate an optimization objective function. The objective function considers the comprehensive effects of delivery rate, conversion rate, cost, and delay to determine the optimal SMS channel combination; An optimization algorithm unit, which is used to calculate the optimal SMS channel combination according to the multi-objective optimization algorithm.

[0010] Preferably, the recommendation algorithm module includes: A deep learning unit, which is used to optimize the selection strategy of the SMS channel by training a deep neural network based on the feature vectors of the SMS channel; A strategy adjustment unit, which is used to adjust the parameters of the deep learning unit according to the real-time feedback data and marketing effect.

[0011] Preferably, the feedback learning module includes: A real-time data feedback unit, which is used to collect user feedback data and feedback it to the recommendation algorithm module; A model adjustment unit for adjusting the parameters of the recommendation algorithm module according to real-time data.

[0012] Preferably, the objective function in the multi-objective optimization module includes: Where Z is the comprehensive scoring result, w1, w2, w3, w4 are the weight coefficients of delivery rate, conversion rate, cost, and latency, R i , A i , C i , T i are respectively the delivery rate, conversion rate, cost, and latency of the i-th SMS channel, x i is the selection decision variable, and n is the total number of SMS channels.

[0013] Preferably, the optimization algorithm unit in the multi-objective optimization module uses the Lagrange multiplier method, and the Lagrangian function is: Where L is the Lagrangian function; x i is the decision variable; n is the total number of SMS channels; R i , A i , C i , T i are respectively the delivery rate, conversion rate, cost, and latency of the i-th SMS channel; w1, w2, w3, w4 are the weight coefficients of delivery rate, conversion rate, cost, and latency; λ is the Lagrange multiplier; k is the target value in the constraint condition.

[0014] Preferably, the recommendation algorithm module uses a deep Q-network for policy optimization, and the Q-value update formula is: Where Q(s t , a t ) is the Q-value of performing action a t in state s t ; α is the learning rate; R t is the immediate reward obtained after performing action a t in state s t ; γ is the discount factor; s t+1 is the new state transferred to after performing action a t ; a ′ is the action with the maximum Q-value among all the actions available in the new state s t +1; is the possible optimal Q-value in the new state s t+1 , that is, the maximum future return under the optimal policy.

[0015] The present invention also provides a method for intelligent recommendation of SMS channels based on the construction of an intelligent marketing system, including the following steps: S1. Collect historical data and real-time data of each SMS channel through the data collection module. The historical data includes the delivery success rate, conversion rate, latency, and cost of each channel. S2. Extract key features of each SMS channel from the data through the feature extraction module and generate feature vectors. S3. Through the multi-objective optimization module, calculate the comprehensive score of each channel according to the feature vectors, combined with the multi-objective optimization algorithm, and determine the optimal SMS channel combination according to the comprehensive score. S4. Generate a recommended list of SMS channels through the recommendation algorithm module. The recommended list is based on the output of the multi-objective optimization module and real-time feedback data. S5. Through the feedback learning module, dynamically adjust the strategy of the recommendation algorithm module according to the real-time marketing effect and user feedback data to optimize the recommendation effect.

[0016] The present invention provides an intelligent recommendation system and method for SMS channels based on an intelligent marketing system, having the following beneficial effects: 1. The present invention adopts a technical solution of combining a multi-objective optimization model with real-time feedback of SMS channels, achieving the technical effect of dynamically adjusting the selection of SMS channels under different business requirements. Compared with the prior art solutions that only recommend channels based on historical statistical data or fixed rules, the present invention can perform real-time optimization according to the actual marketing effect and changes in channel status, avoiding problems such as lagging channel selection and poor adaptability, and improving the accuracy and effect of SMS delivery.

[0017] 2. The present invention uses the feature extraction module to standardize the key indicators of SMS channels, realizing unified representation and efficient calculation of data. Compared with the traditional method of simply relying on raw data for channel scoring, the present invention reduces the interference of data noise on decision-making, and can ensure the stability and rationality of calculation results in the case of a large number of channel indicator dimensions, improving the robustness of the recommendation system.

[0018] 3. The present invention introduces a feedback learning mechanism to continuously optimize the recommendation strategy, achieving the technical effect of adjusting SMS channel allocation based on actual user behavior. Different from the prior art methods that only rely on static scoring or manual rule adjustment, the present invention can automatically adapt to changes in the market environment. Especially when the quality of SMS channels fluctuates greatly, it can quickly adjust the strategy to avoid the decline of marketing effect caused by channel anomalies.

[0019] 4. The present invention uses a recommendation algorithm module combined with multiple optimization methods to comprehensively evaluate SMS channels, so as to provide the optimal channel combination in different business scenarios; compared with the method of single scoring or fixed threshold screening channels in the prior art, the present invention can not only consider the global optimization of channels, but also dynamically adjust the weights of different indicators, is applicable to more complex SMS business scenarios, and improves the flexibility and adaptability of recommendation decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the system architecture diagram of the present invention; Figure 2 is the method flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] Please refer to Figure 1 , the embodiment of the present invention provides an SMS channel intelligent recommendation system based on an intelligent marketing system, including: A data acquisition module, configured to collect historical data and real-time data of each SMS channel, and transmit the data to a feature extraction module and a multi-objective optimization module; In the SMS channel intelligent recommendation system of the present invention, in order to ensure the accuracy, real-time performance and stability of the recommendation results, the data acquisition module, as the basic support module of the entire system, plays a crucial role; this module is responsible for collecting multi-dimensional data of each SMS channel, providing data support for subsequent feature extraction, optimization calculation and recommendation decision-making, etc.; due to the dynamic changes in the status of SMS channels and the complex and changeable marketing environment, the performance of the data acquisition module will directly affect the overall effect of the recommendation system; generally, in order to ensure the comprehensiveness, timeliness and accuracy of the data, the data acquisition module of the present invention adopts a multi-source data fusion mechanism, combines the synchronous acquisition strategy of historical data and real-time data, and constructs a comprehensive data support system, so as to more accurately capture the dynamic characteristics of SMS channels and improve the effectiveness of the recommendation system.

[0023] In this embodiment, the data acquisition module includes a historical data acquisition unit and a real-time data acquisition unit; among them, the historical data acquisition unit is used to collect multi-dimensional index data of each SMS channel in historical marketing activities, so as to mine the long-term efficiency characteristics of the channel in the feature extraction module; and the real-time data acquisition unit is responsible for obtaining the instant status information of each SMS channel to reflect the working performance of the SMS channel in the current environment.

[0024] Specifically, the historical data collection unit can record key metrics such as the delivery success rate, conversion rate, latency, and cost of each SMS channel; these metrics serve as performance measurement criteria for the SMS channels and play an important role in the multi-objective optimization module of the present invention; for example, the delivery success rate can characterize the reliability of the SMS channel, the conversion rate can measure its marketing effectiveness, the latency is used to evaluate the response speed of the channel, and the cost metric directly reflects the economic benefits of the marketing campaign; generally, these historical data are obtained through the API interface with the SMS channel service provider, or can be recorded and accumulated by the internal system, and the specific choice depends on the actual application scenario.

[0025] In some embodiments, to further improve the accuracy of the data, the historical data collection unit can also introduce a data cleaning mechanism; specifically, the following data cleaning strategies can be adopted: For significantly abnormal data samples (such as a success rate exceeding 100% or a negative cost, etc.), they can be processed by using median replacement or sample elimination; In the case of missing data, methods such as mean filling and interpolation algorithms can be used to fill in the missing data to ensure data integrity; To reduce the volatility and abnormal interference of the data, normalization processing can also be performed on the collected data to convert the data of each dimension to the same numerical interval (such as 0 to 1) to reduce the impact of the dimensional difference between different metrics on the model.

[0026] In another possible implementation, the real-time data collection unit adopts a strategy that combines a polling mechanism and an event-triggered mechanism to ensure that the dynamic data of the SMS channels can be captured in a timely manner; specifically, the polling mechanism can regularly collect the instant data of each SMS channel at fixed time intervals (such as every 10 seconds, every 1 minute, etc.); and the event-triggered mechanism immediately starts the data update process when a key event occurs (such as SMS channel anomaly, large-scale sending failure, etc.) to shorten the data lag time and improve the response speed of the system; as an option, the metrics that the real-time data collection unit can record include: the instant delivery status of the SMS channel, the failure rate, user feedback information, network congestion status, etc.

[0027] In this embodiment, to further improve the comprehensiveness and accuracy of the SMS channel status monitoring, the real-time data collection unit can also combine multi-dimensional data features; for example, in some embodiments, the real-time data can be combined with time features (such as peak sending periods, holidays, etc.), geographical features (such as the delivery rate differences in different cities and different network environments), and user features (such as user activity, receiving preferences, etc.) and other information; these data dimensions can further enrich the model input feature space and improve the adaptability and accuracy of the recommendation system.

[0028] The intelligent SMS channel recommendation system of the present invention introduces a data feature extension mechanism in the data collection module to provide a richer information source for subsequent optimization decisions. During the data feature extension process, time series analysis techniques (such as ARIMA, LSTM, etc.) are used to extract the temporal features of the data. In addition, clustering analysis methods (such as K-means, DBSCAN, etc.) can also be used to group the SMS channels according to their characteristics, so that the optimization model is more targeted. In addition, during the data collection process, in combination with the multi-objective optimization model of the present invention, the data of specific indicators also need to be correspondingly transformed and processed. For example, in the multi-objective optimization function, the performance of the SMS channel is measured by the following formula: where Z is the comprehensive scoring result, w1, w2, w3, w4 are the weight coefficients of the delivery rate, conversion rate, cost, and delay, R i , A i , C i , T i are the delivery rate, conversion rate, cost, and delay of the i-th SMS channel respectively, x i is the selection decision variable, and n is the total number of SMS channels.

[0029] In the specific implementation process, the data collection module is not only responsible for directly collecting the data of indicators such as R i , A i , C i , T i , but also needs to perform further processing on some indicators according to the system requirements. For example: To smooth out the abnormal fluctuations in the delivery rate data, the moving average algorithm can be used to smooth R i . The conversion rate A i can be weighted and adjusted in combination with factors such as the marketing cycle and user behavior to improve the timeliness of the indicator. The cost C i can be measured and weighted and combined according to the short-term cost and long-term cost respectively. For the delay T i , a delay tolerance threshold can be introduced to filter the influence of minor fluctuations on the model.

[0030] In some embodiments, to improve the performance of the multi-objective optimization module, the data collection module can also adaptively adjust the weight coefficients of each indicator for different scenarios. For example, in the scenario of emergency notification SMS, the weights of the delivery rate and delay can be increased; while in the scenario of promotional marketing SMS, the weights of the conversion rate and cost can be increased.

[0031] A feature extraction module, configured to extract key features of each SMS channel from the data, generate feature vectors, and transmit the feature vectors to a multi-objective optimization module for comprehensively evaluating the effectiveness of each channel; In the SMS channel intelligent recommendation system of the present invention, the feature extraction module is an important part of the system, responsible for analyzing and processing the historical data and real-time data obtained by the data acquisition module to extract the key features of each SMS channel and generate feature vectors; this module plays a connecting role in the overall system architecture, and its processing effect directly affects the score calculation of the subsequent multi-objective optimization module and the recommendation effect of the recommendation algorithm module; generally, due to the large number of data dimensions of the SMS channel and the complex correlation and non-linear characteristics among the indicators, the feature extraction module needs to have strong data processing and feature screening capabilities to extract the key features that have a significant impact on the marketing effect, avoid data redundancy, and improve the effectiveness and stability of the model.

[0032] In this embodiment, the feature extraction module includes a feature analysis unit and a feature vector generation unit; among them, the main function of the feature analysis unit is to deeply analyze the original data obtained by the data acquisition module and extract the key features that can significantly affect the marketing effect of the SMS channel; the feature vector generation unit then combines the selected features into feature vectors on the basis of feature analysis for subsequent processing by the multi-objective optimization module and the recommendation algorithm module.

[0033] Specifically, the feature analysis unit processes the historical data and real-time data of the SMS channel through various analysis methods; in one possible implementation, the Information Gain method can be used to evaluate the contribution degree of each data indicator to the target variable (such as conversion rate, delivery success rate, etc.), so as to screen out the most influential features; in another possible implementation, the ANOVA method can be used to calculate the variance of the multi-dimensional indicators of each SMS channel, and select the features with larger variance and more obvious fluctuations as key features to enhance the sensitivity of the model to data fluctuations; as an option, the Chi-square test method can also be used to screen out the discrete features with a higher degree of association with the target variable.

[0034] In some embodiments, to further improve the generalization ability and stability of the model, the feature analysis unit can also introduce a dimensionality reduction processing mechanism and use the PCA method to perform dimensionality reduction on the feature data; specifically, the PCA method can extract the principal components that best represent the data distribution based on the covariance matrix, reduce the feature dimension while retaining the main information of the data; generally, the PCA algorithm calculates the principal components according to the following formula: Y = XW; Among them, Y is the feature matrix after dimensionality reduction; X is the original feature matrix; W is the weight matrix composed of eigenvectors, and the eigenvectors correspond to the principal components of the original features. The calculation of the weight matrix W can be obtained by decomposing the covariance matrix, and the covariance matrix is defined as: Σ = m 1 ∑ i m ＝1 (X i - μ)(X i - μ) T ; Among them, Σ is the covariance matrix; m is the number of samples; X i represents the feature vector of the i-th sample; μ is the sample mean; (X i - μ) is the sample data after removing the mean; (X i - μ) T is the transpose of the sample data after removing the mean; As an option, in some scenarios with a large amount of data, the singular value decomposition (SVD) method can be used to replace the PCA method to further improve the dimensionality reduction efficiency; The singular value decomposition method can represent the original feature matrix X as: X = UΣV T ; Among them, X can be any matrix, not necessarily a square matrix; U is the left singular vector matrix; Σ is a diagonal matrix, and the diagonal elements are singular values; V is the right singular vector matrix; T (i.e., V T ) represents the transpose of the matrix V.

[0035] In terms of eigenvector generation, the eigenvector generation unit is responsible for combining and converting the effective feature combinations selected by the feature analysis unit into eigenvectors for further processing by the multi-objective optimization module and the recommendation algorithm module; Generally, the construction of eigenvectors can include various types of features to comprehensively characterize the performance features of the SMS channel; For example: Numerical features such as delivery success rate, conversion rate, cost, and delay; Categorical features such as SMS sending time, marketing campaign type, and user group characteristics; Environmental features such as geographical features and time period features (such as weekdays and weekends, day and night).

[0036] In a possible implementation manner, the eigenvector generation unit can use the normalization method to process the numerical features to ensure that the numerical ranges between the feature dimensions are similar and avoid the problem of unstable gradients during model training; Specifically, the normalization method can use the minimum-maximum normalization (Min-Max Normalization) method, and the normalization formula is as follows: Among them, X norm represents the normalized eigenvalue; X represents the original eigenvalue; X min and X max are respectively the minimum and maximum values in the feature data.

[0037] In another possible implementation, to further improve the robustness of the model to abnormal data, the feature vector generation unit can use the Z-score normalization method to normalize the data; its formula is: Among them, X std represents the normalized eigenvalue; X represents the original eigenvalue; μ is the mean of the feature data; σ is the standard deviation of the feature data.

[0038] In some embodiments, to improve the adaptability and scene matching degree of the model, the feature vector generation unit can also perform weighted adjustment on the feature data according to different marketing scenarios; for example, in the scenario emphasizing delivery timeliness, the weight of the delay feature can be increased in the feature vector; in the scenario of controlling marketing costs, a higher weight can be given to the cost feature; this weighting mechanism helps to optimize the performance of the model and make it more in line with the actual needs in different scenarios.

[0039] The multi-objective optimization module is used to calculate the comprehensive score of each channel according to the feature vector, in combination with the multi-objective optimization algorithm, and the multi-objective optimization module generates the optimal SMS channel combination through interaction with the recommendation algorithm module; In the SMS channel intelligent recommendation system of the present invention, the multi-objective optimization module undertakes the core decision-making function. Its purpose is to calculate the comprehensive score of each SMS channel and determine the optimal SMS channel combination on the basis of comprehensively considering various performance indicators of the SMS channel (such as delivery rate, conversion rate, cost, delay, etc.); due to the existence of uncertain factors such as performance fluctuations and environmental interference in the actual application of the SMS channel, and there is often a trade-off relationship between various indicators, so this module uses the multi-objective optimization algorithm to balance the conflict relationship between different indicators and ensure that the recommendation result has a better marketing effect while meeting the business requirements; generally, this module realizes the dynamic optimization of the SMS channel combination plan by introducing a weighted scoring mechanism and the Lagrange multiplier method to meet the personalized needs in different scenarios.

[0040] In this embodiment, the multi-objective optimization module includes an objective function unit and an optimization algorithm unit; among them, the objective function unit is responsible for constructing and calculating the optimization objective function according to the feature vector generated by the feature extraction module; the optimization algorithm unit uses the Lagrange multiplier method and the multi-objective optimization algorithm to solve the objective function and output the optimal SMS channel combination.

[0041] Specifically, the optimized objective function constructed by the objective function unit is as follows: Among them, w1, w2, w3, and w4 are the weight coefficients of the delivery rate, conversion rate, cost, and delay. R i , A i , C i , T i are respectively the delivery rate, conversion rate, cost, and delay of the i-th SMS channel, x i is the selection decision variable, and n is the total number of SMS channels.

[0042] In this embodiment, in order to further improve the flexibility of the objective function, the weight coefficients w1, w2, w3, and w4 can be dynamically adjusted according to the requirements of different scenarios; as an option, in the scenario where the delivery timeliness is emphasized, the weight of w4 can be appropriately increased; in the scenario of controlling the marketing cost, the weight of w3 can be preferentially increased; in the scenario where user feedback and marketing effects are emphasized, the proportion of w2 can be increased; through this mechanism, the self-adaptability of the model can be effectively improved to meet the requirements of different scenarios.

[0043] In some embodiments, to ensure the feasibility of the SMS channel combination, the optimization algorithm unit uses the Lagrange multiplier method to integrate the constraint conditions in the objective function; Specifically, the constraint condition can be defined as the limit on the number of selected channels, and the constraint condition is as follows: Among them, x i is the selection decision variable, n is the total number of SMS channels, and k is the maximum number of selected SMS channels.

[0044] Based on the above constraint conditions, the optimization algorithm unit constructs a Lagrangian function by introducing the Lagrange multiplier λ, and its form is: Among them, L is the Lagrangian function; x i is the decision variable; n is the total number of SMS channels; R i , A i , C i , T i are respectively the delivery rate, conversion rate, cost, and delay of the i-th SMS channel; w1, w2, w3, and w4 are the weight coefficients of the delivery rate, conversion rate, cost, and delay; λ is the Lagrange multiplier; k is the target value in the constraint condition.

[0045] In the specific implementation process of the optimization algorithm, the optimization algorithm unit can adopt optimization algorithms such as multi-objective genetic algorithm (such as NSGA-II), particle swarm optimization algorithm (PSO), or simulated annealing algorithm (SA) to improve the search efficiency and avoid falling into local optimal solutions; in a possible implementation manner, the optimization algorithm unit completes the solution of the optimal channel combination through the following steps: First, initialize the population or solution set to ensure that each solution in the solution set satisfies the constraint conditions; Secondly, calculate the objective function value of each solution, and use the non-dominated sorting method to screen out non-inferior solutions; Then, use operations such as crossover and mutation to iterate the population and gradually search for better solutions; Finally, when the maximum number of iterations or the convergence condition is satisfied, output the optimal SMS channel combination.

[0046] In some embodiments, to improve the stability and convergence speed of the algorithm, the optimization algorithm unit can also introduce a random perturbation mechanism to avoid falling into local optimal solutions; specifically, by randomly selecting some solutions for small perturbations in each round of iteration, the solution set has stronger exploration ability in the search space.

[0047] In another possible implementation manner, to improve the robustness of the system to abnormal data, the optimization algorithm unit can further introduce a fault tolerance mechanism; for example, when a sudden failure occurs in the SMS channel, the system can automatically remove the abnormal channel from the recommendation list and recalculate the recommendation result using the multi-objective optimization model; in some marketing activities, to improve the flexibility of the model, the weight parameters can also be dynamically adjusted according to user feedback data to further optimize the recommendation strategy.

[0048] The recommendation algorithm module is used to generate a recommendation list of SMS channels based on the output of the multi-objective optimization module and real-time feedback data, and feedback the result to the feedback learning module to optimize the recommendation strategy; In the SMS channel intelligent recommendation system of the present invention, the recommendation algorithm module is the core calculation and decision-making module of the system, responsible for generating a recommendation list of SMS channels according to the channel scores and related feature data output by the multi-objective optimization module, combined with real-time feedback information; the recommendation algorithm module plays a key role in the system, and its performance directly affects the accuracy, stability, and marketing effect of SMS channel selection; generally, the recommendation algorithm module needs to have strong adaptability to cope with the dynamic changes of the SMS channel environment; for this reason, this module adopts the deep Q network (DQN) as the core model of the recommendation algorithm and introduces a strategy adjustment mechanism to further improve the accuracy and stability of the recommendation result.

[0049] In this embodiment, the recommendation algorithm module includes a deep learning unit and a policy adjustment unit. Among them, the deep learning unit performs policy learning on the short message channel features based on the Deep Q-Network (DQN) to generate a recommendation list. The policy adjustment unit dynamically adjusts the model parameters of the deep learning unit according to the system feedback data to optimize the recommendation effect.

[0050] Specifically, the deep learning unit adopts the Deep Q-Network (DQN) model to optimize the recommendation policy of the short message channel through the reinforcement learning mechanism. In the DQN model, the short message channel recommendation problem of the system is modeled as a Markov decision process (MDP), and its state space, action space, and reward mechanism are as follows: In terms of the definition of the state space, the state of the system includes the feature vector of the short message channel and the historical marketing data. As an option, the state vector can be expressed as: S t =[f1,f2,…,f m ,h1,h2,…,h p ; where S t represents the system state at time t; f1,f2,…,f m are the feature data of the short message channel, including multi-dimensional information such as delivery rate, conversion rate, delay, cost, geographical features, and time features; h1,h2,…,h p are the historical marketing data, usually including historical click-through rate, failure rate, user feedback and other indicators.

[0051] In terms of the definition of the action space, the recommendation decision of the system can be expressed as the selection combination of the short message channel. As an option, the action can be defined as a vector: A t =[x1,x2,…,x n ; where A t represents the recommendation decision at time t; x i is the decision variable, whose value is 0 or 1. When the value is 1, it means selecting the i-th short message channel, and when the value is 0, it means not selecting the short message channel; n represents the total number of short message channels.

[0052] In terms of the definition of the reward mechanism, in order to improve the guidance of the recommendation policy to the marketing effect, in this embodiment, the reward value is dynamically calculated according to the actual marketing effect of the short message channel. The reward value calculation formula is as follows: R t =α1R delivery +α2R conversion -α3R cost -α4R delay ; where R tRepresents the immediate reward value at time t; R delivery Represents the actual delivery success rate of the SMS channel; R conversion Represents the actual conversion rate of the SMS channel; R cost Represents the actual sending cost of the SMS channel; R delay Represents the actual delay of the SMS channel; α1, α2, α3, α4 are the weight parameters of each index respectively.

[0053] In this embodiment, based on the above state, action, and reward mechanisms, the deep learning unit uses the following Q-value update formula to iteratively optimize the recommendation strategy: Among them, Q(s t ,a t ) is the Q-value of performing action a t in state s t ; α is the learning rate; R t is the immediate reward obtained after performing action a t in state s t ; γ is the discount factor; s t+1 is the new state transferred to after performing action a t ; a ′ is the action with the largest Q-value among all the actions that can be selected in the new state s t +1; is the possible optimal Q-value in the new state s t+1 , that is, the maximum future return under the optimal strategy.

[0054] In some embodiments, to further improve the stability of the DQN model, the deep learning unit can adopt an experience replay mechanism (Experience Replay); specifically, the experience replay mechanism stores the state, action, reward, and next state in each interaction process in a replay buffer, and randomly samples from the replay buffer during model training to prevent the model from falling into a local optimal solution due to the similarity of consecutive data samples.

[0055] In another possible implementation, to improve the convergence efficiency of the model in a sparse reward environment, the deep learning unit can introduce a target network (Target Network) mechanism; specifically, the target network, as a delayed copy of the original Q network, has its parameters updated relatively slowly to ensure the stability of Q-value estimation during model training.

[0056] In this embodiment, the policy adjustment unit is responsible for dynamically adjusting the model parameters of the Deep Q-Network (DQN) according to the real-time marketing effect and user feedback data. As an option, the policy adjustment unit can use the Stochastic Gradient Descent (SGD) method or the Adaptive Gradient Optimization Algorithm (such as Adam) to update the model parameters. In addition, to further improve the adaptability of the model, the policy adjustment unit can also dynamically adjust the weight parameters of the reward function in combination with the feedback data to optimize the learning objective of the model.

[0057] In some embodiments, to avoid overfitting during model training, the policy adjustment unit can also introduce an Early Stopping mechanism. Specifically, when the performance of the model on the validation set no longer improves, the model training is terminated in advance to reduce unnecessary training overhead and improve the generalization ability of the model.

[0058] In another possible implementation, to cope with the rapid changes in the SMS channel environment, the policy adjustment unit can also adopt a Dynamic Exploration mechanism. Specifically, the policy adjustment unit introduces a higher exploration probability in the early training stage of the model to improve the model's exploration ability for unknown policies. As the model training progresses, the exploration probability is gradually reduced to make the model approach the converged optimal policy.

[0059] The feedback learning module is used to dynamically adjust the policy of the recommendation algorithm module according to the real-time marketing effect and user feedback data. In the SMS channel intelligent recommendation system of the present invention, the feedback learning module, as an important adjustment mechanism of the system, is responsible for dynamically adjusting the recommendation policy according to the recommendation results generated by the recommendation algorithm module and the actual marketing effect of the SMS channel. By introducing a real-time data feedback mechanism and a model parameter optimization mechanism, this module ensures that the recommendation algorithm module can continuously adapt to the changes in the SMS channel environment and improve the accuracy and stability of the recommendation results. Generally, due to the large uncertainty in the state of the SMS channel and the strong dynamic fluctuations in the user response situation during marketing activities, the feedback learning module needs to have strong self-adjustment and fast response capabilities to ensure that the recommendation system always maintains a better recommendation effect.

[0060] In this embodiment, the feedback learning module includes a real-time data feedback unit and a model adjustment unit. Among them, the real-time data feedback unit is responsible for collecting the marketing effect of the SMS channel and user feedback data and transmitting this data to the model adjustment unit. The model adjustment unit then dynamically adjusts the parameters of the recommendation algorithm module according to the feedback data to optimize the recommendation policy.

[0061] Specifically, the core of the real-time data feedback unit lies in the timeliness of data collection and the comprehensiveness of feedback content; in a possible implementation, the real-time data feedback unit can periodically obtain the marketing effect data of the SMS channel through the status monitoring interface of the SMS channel; these data can include indicators such as the delivery success rate, conversion rate, failure rate, and user unsubscription situation of the SMS channel.

[0062] In some embodiments, to further enrich the source of feedback data, the real-time data feedback unit can also introduce a user behavior analysis mechanism; specifically, this mechanism can monitor the behavior data of users after receiving SMS, including indicators such as SMS opening time, number of link clicks, and interaction behaviors; these data help to deeply analyze the correlation between the recommendation strategy and the marketing effect, so as to further optimize the decision-making of the recommendation algorithm module.

[0063] In another possible implementation, the real-time data feedback unit can also combine external environmental data, such as holidays, meteorological conditions, peak traffic hours, etc., to enrich the data input of the recommendation system and improve the adaptability of the recommendation model.

[0064] In this embodiment, the model adjustment unit is responsible for dynamically optimizing the parameters of the recommendation algorithm module according to the feedback data provided by the real-time data feedback unit; in a possible implementation, the model adjustment unit can adjust the parameters of the recommendation algorithm module based on gradient descent algorithms (such as Stochastic Gradient Descent SGD, Adam, etc.) to minimize the prediction error of the recommendation strategy; specifically, the model parameter update formula can be expressed as: where θ t+1 is the model parameter at time t + 1; θ t is the model parameter at time t; η is the learning rate, which controls the amplitude of parameter update; is the gradient of the loss function L with respect to the model parameter; The loss function L can usually be dynamically defined according to the performance results of the recommendation strategy.

[0065] In some embodiments, to improve the convergence speed of the recommendation algorithm module, the model adjustment unit can further combine the momentum optimization mechanism (Momentum) to accelerate the parameter update process by introducing the momentum term of the gradient; its parameter update formula is: θ t+1 = θ t - v t+1 ; where v t+1 is the momentum term at time t + 1; μ is the momentum coefficient, which controls the update amplitude of the momentum term; v t is the momentum term at time t.

[0066] As an option, the model adjustment unit can also introduce a Target Network mechanism to improve the stability of model parameter updates. In this mechanism, the parameters of the recommendation algorithm module are divided into two independent network structures: an Online Network and a Target Network. Specifically, the Online Network is responsible for updating the parameters based on the current state and feedback data, while the Target Network, as a delayed copy, uses a slower parameter update strategy to provide a more stable Q-value target.

[0067] In another possible implementation, to prevent the recommendation algorithm module from falling into a local optimum, the model adjustment unit can introduce an Exploration mechanism. Specifically, during the recommendation process, the model can introduce a certain probability of random recommendation behavior based on the ∈-Greedy strategy to ensure that the model can continue to explore better recommendation strategies even in the case of sparse data or abnormal feedback. The probability function of the recommendation decision can be expressed as: where π(a|s) represents the probability of selecting action a in state s; a * represents the action with the largest current Q-value; |A| is the total number of optional actions; ∈ is the exploration probability, with a value between 0 and 1, and a higher ∈ indicates a stronger exploration tendency.

[0068] In this embodiment, to further improve the flexibility of model parameter adjustment, the model adjustment unit can also dynamically correct the parameters of the reward mechanism according to the historical feedback data of the SMS channel. In one possible implementation, a reward decay mechanism can be introduced to gradually reduce the influence of historical rewards on model parameter updates and improve the model's response speed to recent feedback data. The reward value update formula of the reward decay mechanism is as follows: R t = (1 - λ)R t-1 + λR new ; where R t is the updated reward value at the current moment; R t-1 is the reward value at the previous moment; R new is the newly received feedback reward value; λ is the decay coefficient that controls the influence degree of historical rewards on the current reward.

[0069] Please refer to Figure 2 , the present invention also provides a smart recommendation method for the SMS channel based on the construction of a smart marketing system, including the following steps: S1. Collect historical data and real-time data of each SMS channel through the data collection module. The historical data includes the delivery success rate, conversion rate, latency, and cost of each channel. Among them, the collection of historical data can be based on long-term monitoring results and combined with statistical information of different time periods and different marketing activities to ensure the stability and representativeness of the data. The acquisition of real-time data depends on the status monitoring interface of the SMS channel, which can capture the latest channel status changes in a short time, thus providing accurate input for subsequent optimization calculations. S2. Extract key features of each SMS channel from the data through the feature extraction module and generate feature vectors. Specifically, the feature extraction module can adopt data preprocessing methods such as standardization and normalization to eliminate the differences in the feature value scales between different channels and enhance the consistency of the data. In addition, in some embodiments, the feature extraction module can further apply principal component analysis (PCA) or dimensionality reduction algorithms to reduce the data dimension, improve the calculation efficiency, and at the same time retain the key factors affecting the selection of SMS channels. S3. Through the multi-objective optimization module, according to the feature vectors, calculate the comprehensive score of each channel in combination with the multi-objective optimization algorithm, and determine the optimal SMS channel combination according to the comprehensive score. Generally, the multi-objective optimization module can adopt methods such as weighted summation method, Pareto optimization, or evolutionary algorithm to seek a balance among multiple evaluation indicators. In some embodiments, this module can also dynamically adjust each weight parameter according to different business requirements. For example, in marketing activities, it focuses on the conversion rate, while in system notification SMS, it pays more attention to the delivery success rate to ensure that the recommended results match the actual business needs. S4. Generate a recommended list of SMS channels through the recommendation algorithm module. The recommended list is based on the output of the multi-objective optimization module and real-time feedback data. The recommendation algorithm module not only depends on the static scoring results of multi-objective optimization but also combines recent user feedback and market environment changes for dynamic adjustment. As a possible implementation method, this module can introduce collaborative filtering, rule-based recommendation, or reinforcement learning methods to provide the optimal SMS channel recommendation scheme in different application scenarios, while ensuring that the recommended list has good adaptability and robustness. S5. Through the feedback learning module, dynamically adjust the strategy of the recommendation algorithm module according to the real-time marketing effect and user feedback data to optimize the recommendation effect. In practical applications, the feedback learning module can adopt supervised learning or reinforcement learning mechanisms to continuously optimize the recommendation strategy of SMS channels during long-term operation. In addition, in some embodiments, this module can also combine anomaly detection technology to identify factors that may cause the decline in the effect of SMS channels, such as a sudden decrease in the service quality of a certain channel or channel restrictions due to policy changes, so as to adjust the strategy in time to ensure the stability and reliability of the recommendation system.

[0070] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The SMS channel intelligent recommendation system built based on the intelligent marketing system is characterized by: include: Data collection module, used to collect historical data and real-time data of each SMS channel, and transmit the data to feature extraction module and multi-objective optimization module; A feature extraction module, used to extract key features of each SMS channel from the data and generate feature vectors, and transmit the feature vectors to the multi-objective optimization module for comprehensive evaluation of the effectiveness of each channel; A multi-objective optimization module, used to calculate the comprehensive score of each channel according to the feature vector in combination with a multi-objective optimization algorithm, and the multi-objective optimization module generates an optimal SMS channel combination by interacting with a recommendation algorithm module; A recommendation algorithm module, used to generate a recommendation list of SMS channels based on the output of the multi-objective optimization module and real-time feedback data, and feed the results back to the feedback learning module to optimize the recommendation strategy; The feedback learning module is used to dynamically adjust the strategy of the recommendation algorithm module according to real-time marketing effects and user feedback data.

2. According to claim 1, the SMS channel intelligent recommendation system based on the intelligent marketing system is characterized in that: The data acquisition module comprises: The historical data collection unit is used to collect the delivery rate, conversion rate, latency, and cost indicators of each channel in historical marketing activities; the real-time data collection unit is used to collect the status information of the SMS channel in real time, including delivery status, failure rate, and user feedback.

3. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The feature extraction module comprises: The feature analysis unit is used to analyze the historical data of each channel, extract the key factors affecting the marketing effect, and generate a channel feature vector.

4. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The multi-objective optimization module includes: An objective function unit is used to construct and calculate an optimization objective function, wherein the objective function considers the comprehensive effects of delivery rate, conversion rate, cost and delay to determine the optimal SMS channel combination; The optimization algorithm unit is used to calculate the optimal SMS channel combination according to the multi-objective optimization algorithm.

5. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The recommendation algorithm module includes: A deep learning unit is used to optimize the SMS channel selection strategy by training a deep neural network based on the feature vector of the SMS channel; The strategy adjustment unit is used to adjust the parameters of the deep learning unit according to real-time feedback data and marketing effects.

6. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The feedback learning module includes: Real-time data feedback unit, used to collect user feedback data and feed it back to the recommendation algorithm module; The model adjustment unit is used to adjust the parameters of the recommendation algorithm module according to real-time data.

7. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The objective functions in the multi-objective optimization module include: Among them, Z is the comprehensive score result, w1, w2, w3, w4 are the weight coefficients of delivery rate, conversion rate, cost and delay, R i ,A i ,C i ,T i are the delivery rate, conversion rate, cost and delay of the ith SMS channel, respectively, i is the selection decision variable, and n is the total number of SMS channels.

8. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The optimization algorithm unit in the multi-objective optimization module uses the Lagrange multiplier method, and the Lagrange function is: Where L is the Lagrangian function; x i is the decision variable; n is the total number of SMS channels; R i ,A i ,C i ,T i are the delivery rate, conversion rate, cost and delay of the i-th SMS channel respectively; w1, w2, w3, w4 are the weight coefficients of delivery rate, conversion rate, cost and delay; λ is the Lagrange multiplier; k is the target value in the constraint condition.

9. The SMS channel intelligent recommendation system based on the intelligent marketing system according to claim 1 is characterized in that: The recommendation algorithm module uses a deep Q network for strategy optimization, and the Q value update formula is: Among them, Q(s t ,a t ) in state s t Next, perform action a t Q value; α is the learning rate; R t In status t Next, perform action a t The immediate reward obtained after γ is the discount factor; s t+1 To perform action a t The new state after transfer; a ′ In the new state t +1: Among all the actions available, the one with the largest Q value; In the new state t+1 The optimal possible Q value, that is, the maximum future return under the optimal strategy.

10. The SMS channel intelligent recommendation method based on the intelligent marketing system is applied to the SMS channel intelligent recommendation system based on the intelligent marketing system as described in any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Collect historical data and real-time data of each SMS channel through a data acquisition module, wherein the historical data includes the delivery success rate, conversion rate, delay, and cost of each channel; S2. extracting key features of each SMS channel from the data through a feature extraction module, and generating a feature vector; S3. Calculate the comprehensive score of each channel according to the feature vector and the multi-objective optimization algorithm through the multi-objective optimization module, and determine the optimal SMS channel combination according to the comprehensive score; S4, generating a recommendation list of SMS channels through a recommendation algorithm module, wherein the recommendation list is based on the output of the multi-objective optimization module and real-time feedback data; S5. Through the feedback learning module, the strategy of the recommendation algorithm module is dynamically adjusted according to the real-time marketing effect and user feedback data to optimize the recommendation effect.

Images