Marketing information management method and system based on multi-channel collaborative pushing

Abstract

The embodiment of the invention discloses a marketing information management method and system based on multi-channel collaborative pushing, and relates to the technical field of digital marketing. According to one specific embodiment, the method comprises the steps that user data from multiple channels are collected and integrated to construct a unified user portrait, and the multiple channels comprise a public domain channel and / or a private domain channel; generating personalized marketing information content based on the user portrait; according to the user portrait and the characteristics of each channel, selecting from a plurality of alternative channels and cooperatively forming a push channel combination; content pushing is executed through the pushing channel combination, and pushing limitation is applied; monitoring key indexes of the pushing effect in real time; and optimizing at least one of the user portrait, the content strategy, the channel strategy and the push strategy based on the effect data. According to the embodiment, unified and intelligent management and collaboration of multiple marketing channels are realized, and the accurate reach rate and the overall conversion efficiency of the marketing information are remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of information technology and digital marketing, and more specifically, to a method and system for integrating and managing multiple marketing channels to achieve precise and automated collaborative information delivery. Background Technology

[0002] In the era of digital marketing, touchpoints between businesses and consumers are highly dispersed across various public and private channels, including social media, search engines, email, mobile applications, instant messaging tools, and short video platforms. While this multi-channel coexistence expands market reach, it also presents significant challenges to marketing management: user data across channels is isolated, forming "data silos" that hinder businesses from developing a unified and comprehensive understanding of their customers; marketing campaigns are often planned and executed independently by different teams on each channel, leading to inconsistencies in information, internal resource consumption, and even the repeated sending of conflicting information to the same user, damaging the brand experience; and manual cross-channel coordination, content adaptation, and performance analysis are extremely labor-intensive, slow to respond, and difficult to achieve large-scale precision marketing.

[0003] Existing technological solutions mostly focus on automating single channels (such as email marketing automation tools) or simply publishing content simultaneously across multiple channels. These solutions lack the ability to deeply integrate and analyze multi-channel data, making it impossible to make decisions based on unified user insights; channel selection relies heavily on fixed rules or manual experience, failing to dynamically optimize channel combinations and push strategies based on real-time feedback; and they have not established a systematic cross-channel collaborative management and performance evaluation system. Therefore, marketing efficiency and ROI are difficult to improve effectively.

[0004] In view of this, existing technologies lack a marketing information management solution that can intelligently solve the optimal cross-channel push strategy based on a unified, dynamically updated user profile, under global budget and user experience constraints, and form a closed-loop optimization based on real-time feedback data. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a marketing information management method, system, electronic device, and computer-readable medium based on multi-channel collaborative push. This solution aims to construct a comprehensive user profile by unifying and integrating multi-channel data, and based on this, intelligently perform multi-channel collaborative decision-making, personalized content generation, automated push execution, and end-to-end effect optimization. In particular, it achieves efficient, accurate, and integrated management of marketing information by constructing and solving a cross-channel collaborative optimization decision model and establishing a data-driven end-to-end strategy iteration mechanism.

[0006] In a first aspect, embodiments of the present invention provide a marketing information management method based on multi-channel collaborative push, comprising: Collect and integrate user data from multiple channels to build a unified user profile, including public and / or private channels; Based on the user profile, personalized marketing information content matching the user characteristics is generated; Based on the user profile and the channel characteristics of multiple alternative push channels, a push channel combination is selected and coordinated from the multiple alternative push channels; The personalized marketing information is pushed to target users through the aforementioned push channel combination, and push restriction rules are applied during the push process; the push restriction rules include control over push frequency, user fatigue, push time period or content type. Real-time monitoring of the effectiveness data after push notifications through different channels, and judgment of the overall and channel-specific push effectiveness based on preset key performance indicators; Based on the performance data, at least one of the following is optimized: the user profile, the strategy for generating personalized marketing information content, the strategy for forming the push channel combination, and the push restriction rules.

[0007] Secondly, embodiments of the present invention provide a marketing information management system based on multi-channel collaborative push, comprising: The multi-channel data integration module is configured to collect and integrate user data from multiple channels to build a unified user profile, including public channels and / or private channels. The personalized content generation module is configured to generate personalized marketing information content that matches the user's characteristics based on the user profile. The multi-channel collaborative decision-making module is configured to select and coordinate a push channel combination from the multiple candidate push channels based on the user profile and the channel characteristics of the multiple candidate push channels. The intelligent push execution module is configured to push the personalized marketing information content to the target user through the push channel combination, and to apply push restriction rules during the push process; The omnichannel performance monitoring module is configured to monitor the performance data after push notifications are sent through different channels in real time, and to judge the overall and channel-specific performance based on preset key performance indicators. The strategy iteration and optimization module is configured to optimize at least one of the following based on the performance data: the user profile, the generation strategy of the personalized marketing information content, the formation strategy of the push channel combination, and the push restriction rules.

[0008] Thirdly, embodiments of the present invention provide an electronic device, including: one or more processors; a storage device having one or more programs stored thereon; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the method described in the first aspect above.

[0009] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in the first aspect above.

[0010] The technical solution provided by this invention breaks down channel data barriers through unified data integration, constructing a panoramic user profile; based on the profile and channel characteristics, it enables intelligent multi-channel collaborative decision-making, achieving optimal allocation of marketing resources; through automated execution and refined management, it ensures consistency and user experience in cross-channel push notifications; and relying on end-to-end monitoring and data-driven strategy optimization, it forms a continuously improving marketing closed loop. Compared with existing technologies, this invention significantly improves the efficiency, accuracy, and final conversion rate of marketing management in a multi-channel environment. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0012] Figure 1 This is a flowchart of a marketing information management method based on multi-channel collaborative push provided by an embodiment of the present invention.

[0013] Figure 2 This is a structural block diagram of a marketing information management system based on multi-channel collaborative push provided in an embodiment of the present invention.

[0014] Figure 3 This is a schematic diagram of multi-channel data integration and user profile construction in one embodiment of the present invention.

[0015] Figure 4 This is a schematic diagram of multi-channel collaborative decision-making and push execution in one embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be described in detail 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.

[0017] It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0018] The algorithms and models involved in this invention include, but are not limited to: User profile construction model: The XGBoost classification model is used to predict user value level, and the Word2Vec model is used to generate behavioral sequence embedding representation.

[0019] Multi-channel decision-making model: The Deep Q-Network reinforcement learning model is adopted, the state is the user profile and the channel state, the action is the channel selection, and the reward is the conversion rate.

[0020] In one specific implementation, the State of the Deep Q-Network reinforcement learning model is defined as the concatenation of the user profile feature vector and the real-time state vectors of each channel; the Action space is defined as the selection of m alternative channels (e.g., represented using m-dimensional one-hot vectors); and the Reward is defined as whether the user converts within a predetermined time window after the push notification (e.g., +1 for conversion, 0 for no conversion, or a reward value set based on the conversion amount). The model uses historical push logs as experience replay data for offline training and online updates.

[0021] Push frequency model: Based on a Poisson process and user response history, this model dynamically estimates the optimal interval for users to receive push notifications. As part of the push frequency control rules, this model is used to dynamically adjust the frequency of user-level push notifications.

[0022] Content generation model: Based on the GPT-based generation model, it takes user tags and product information as input and outputs personalized recommendation copy. Example 1

[0023] Please see Figure 1 This illustrates a flowchart 100 of a marketing information management method based on multi-channel collaborative push according to an embodiment of the present invention. The method can be executed by a server, server cluster, or dedicated computing device, and includes the following steps: Step S110: Multi-channel data integration and user profile construction.

[0024] This step aims to break down data silos and create a unified center for user perception.

[0025] In practice, the following sub-steps may be included: S111, Multi-channel data collection: Data Sources: User data is collected from one or more channels, depending on the actual business needs. For example, data may be collected only from WeChat Official Accounts (public domain) and brand apps (private domain), or it may cover multiple sources such as Weibo, news feed ads, WeChat Work, mini programs, and CRM systems. Data collection methods include API calls, SDK tracking, and log file parsing.

[0026] Data types: The collected data may include, but is not limited to, user identification information (after anonymization), browsing / clicking / searching behavior data, transaction order data, content interaction data (likes, comments, shares), customer service records, etc.

[0027] S112, Data Cleaning and Standardization: The collected raw data undergoes preprocessing, including deduplication, invalid data filtering, format standardization, and outlier handling. For example, it unifies the user ID mapping system across different channels and converts timestamps to a unified timezone format.

[0028] This step resolved the issues of inconsistent raw data quality and varying formats, laying a reliable foundation for subsequent data fusion and analysis. This is a concrete method for achieving "cleaning and standardization," ensuring the data quality for building user profiles.

[0029] S113, Data Association and Fusion: User identification and association are performed using rule-based or machine learning methods. For identifiable users, precise association is achieved through unique identifiers such as mobile phone number, email address, and UnionID; for anonymous users, temporary association across channels and sessions can be achieved using technologies such as device fingerprinting and cookies.

[0030] The associated data is integrated into a unified data model, such as a user event table and a wide table of user attributes.

[0031] The association analysis and fusion algorithm can employ a user identity association method based on graph neural networks. By constructing a heterogeneous graph of user-device-behavior, it learns the embedded representations of user nodes to achieve accurate mapping of user identities across channels. Interest fusion can use a label-weighted fusion model based on collaborative filtering to calculate user interest preference vectors by integrating behavioral data from various channels. For example, the graph neural network can construct a heterogeneous graph with users, devices, and behaviors as nodes, and use graph convolutional networks (GCNs) or graph attention networks (GATs) to learn node embeddings. By calculating the similarity of the embedding vectors, it achieves accurate mapping of user identities across channels.

[0032] S114, User Profile Calculation and Dynamic Update: Tag system construction and calculation: Based on the fused data, multi-level user tags are calculated using statistical rules and machine learning models (such as clustering, classification, and collaborative filtering). For example: Demographic attribute tags: inferred from registration information or models.

[0033] Behavioral preference tags: such as "frequent browsing of digital products" and "prefers weekend shopping".

[0034] Value stage labels: User value levels are divided based on the RFM model (high value, potential, average, retention).

[0035] Predictive labels: such as "probability of purchase in the next 30 days" and "churn risk score".

[0036] Dynamic update mechanism: Establish an update process that combines real-time and batch processing. Key user behaviors (e.g., when a user completes an order with a transaction amount exceeding a preset threshold) trigger real-time tag updates (e.g., immediately updating their "value stage tag" to "high-value user"); the system runs batch calculation tasks periodically (e.g., daily) to update the profiles of all users.

[0037] This step specifically implements association analysis and fusion algorithms to transform fragmented multi-channel data into structured, computable, unified user profiles. Its technical effect lies in providing enterprises with a 360-degree view of users, which is the cornerstone of all subsequent personalized and collaborative decision-making, directly overcoming the problem of "data silos." The association analysis and fusion algorithms include at least one of the following: a user identity association algorithm based on graph neural networks, and an interest fusion model based on collaborative filtering.

[0038] Step S120: Personalized content generation.

[0039] This step is based on a unified user profile to achieve personalized content preparation for each user.

[0040] S121, User Insight Analysis: Analyze the profile tags of the target user group to identify common needs and individual differences. For example, identify a user group with tags such as "fitness enthusiast," "health-conscious," and "white-collar worker."

[0041] S122, Intelligent Content Matching and Generation: Content library matching: The system has a built-in tagged content library. By calculating the match between user profile tags and content tags, it automatically filters the most relevant content. For example, it matches the above user group with an article on "Office Bodyweight Fitness Guide" and a video on "Low-Calorie Lunch Recipes".

[0042] Dynamic content synthesis: For scenarios with extremely high personalization requirements, a template engine is used. For example, it can generate SMS or email content that includes the user's name, last purchased item, and exclusive discount coupons.

[0043] The dynamic content synthesis is based on a natural language generation model, such as a Transformer-based text generation model. It takes user profile tags and marketing objectives as input and outputs personalized marketing copy. It also supports a template engine, which generates content containing user names, historical behaviors, and personalized offers through variable substitution.

[0044] S123, Content Format Adaptation: Based on the channels to be used in step S130, automatically convert the content into an adapted format, such as generating WeChat Official Account articles, WeChat Moments posters, Push notification text, etc.

[0045] This step is the specific implementation of "matching or dynamically generating content materials" and "personalized rendering." Its technical effect lies in transforming a unified user profile into specific, differentiated marketing messages, solving the problems of homogenized traditional marketing content and its mismatch with user needs, thus providing content assurance for improving click-through rates and conversion rates.

[0046] Step S130: Multi-channel collaborative decision-making.

[0047] This step is the intelligent hub that determines "when, through which channel, and to whom to push what content".

[0048] S131, Channel Feature Library Management: Maintain a dynamically updated knowledge base that records the features of each available channel (such as channel A, B, C), including static attributes (content format, cost, typical user profile, maximum reach) and dynamic status (current time period activity, historical conversion performance, platform rule restrictions).

[0049] S132, Collaborative Decision-Making Model Calculation: Inputs: A unified profile of the target users, marketing campaign objectives (such as new user acquisition, user activation, and conversion), budget and resource constraints.

[0050] Model: Construct an optimization problem. The objective function can be to maximize total expected revenue (such as GMV) or the number of conversions, while constraints include total budget, user harassment limit (each user can receive a maximum of N marketing messages per day), capacity limits of each channel, etc.

[0051] Solution and Output: Solve using linear programming, heuristic algorithms, or reinforcement learning models. The output is a "multi-channel push plan" tailored for each user. For example: User A's plan is "Send an exclusive coupon via WeChat at 10:00 AM today; if not used today, send a reminder via APP push at 2:00 PM tomorrow"; User Group B's plan is "Publish an event article via WeChat Official Account and simultaneously promote the topic on Weibo".

[0052] This step specifically implements the "preset multi-channel decision-making model" and "collaborative optimization calculation." Its core effect lies in systematically solving the challenges of resource allocation and collaboration in a multi-channel environment, avoiding the arbitrariness of manual decision-making and conflicts between channels, and achieving intelligent and optimal combination of marketing resources (channels, content, timing), which is key to improving overall marketing efficiency. The multi-channel decision-making model is a reinforcement learning-based collaborative push strategy model. Its inputs include user profile vectors, channel feature matrices, and marketing objectives, and its output is the push channel combination and its execution sequence.

[0053] The collaborative optimization calculation is constructed using the following mathematical model: Objective function:

[0054] Constraints:

[0055]

[0056] in: n: represents the total number of target users.

[0057] m: indicates the total number of available alternative push channels.

[0058] This represents the predicted conversion rate of user i in channel j, which is predicted by a machine learning model trained on historical data.

[0059] The decision variable (whether to push, for example, 0 means no push, 1 means push) indicates whether to push marketing content to user i through channel j.

[0060] For push notification costs.

[0061] B represents the total budget for the activity.

[0062] This is the maximum number of messages a user can receive per day.

[0063] The solution can be obtained using genetic algorithms, linear programming, or reinforcement learning strategy gradient methods.

[0064] In the actual solution implementation, the system transforms the above optimization problem into a computable form. For example, when using linear programming, the objective function and constraints are arranged into a standard form, and a commercial mathematical optimization solver (such as CPLEX or Gurobi) is called for computation. When using a genetic algorithm, chromosome encoding is defined as all... The binary strings are arranged in sequence, and the population is iteratively evolved through selection, crossover, and mutation operations to approximate the optimal solution, using a fitness function (i.e., the objective function value) as the evaluation criterion. During the iteration process, a penalty function method is used to handle constraints, significantly reducing the fitness of individuals that violate budget or user fatigue constraints.

[0065] In one specific embodiment of the collaborative optimization calculation, the predicted conversion rate With decision variables ∈{0,1} (indicating whether to push to user i through channel j), push cost Total budget B and user fatigue constraints Together, they constitute the basic optimization model. It is predicted by a machine learning model (such as XGBoost) that uses user profile features, channel context features, and historical interaction data as input to determine whether a user is converted into a target. This represents the estimated cost (such as CPC or CPM) of sending a push to user i through channel j.

[0066] In this embodiment, the predicted conversion rate It is obtained through an offline-trained, online-service machine learning model. This model takes user profile features, channel context features, and historical interaction data as input, and is trained based on whether the user has converted into a target (e.g., using XGBoost or a deep neural network). The push cost... This cost originates from the fixed cost unit price pre-configured in the channel feature library, or is a cost estimate dynamically obtained based on the real-time bidding environment. It represents the estimated cost of pushing content to user i through channel j. This cost can be the channel's fixed cost per click (CPC), cost per impression (CPM), or a weighted average cost estimated based on historical data.

[0067] In a preferred embodiment, the collaborative decision-making model in step S132 further considers the interaction effects between channel combinations to simulate the complex impact of multi-channel reach on user decisions in the real world. The system achieves this advanced decision-making capability through the following steps: 1. Construction and learning of the interaction effect matrix: The system is maintained and dynamically updated. The inter-channel interaction effect matrix E, where m is the total number of channels. Matrix elements This represents the incremental impact value on the final conversion probability when channel j and channel k work together on the same user within a preset time window (e.g., 24 hours), i.e., the inter-channel interaction effect value.

[0068] > 0: indicates a synergistic effect, meaning that the effect of the channel combination (j, k) is better than the sum of the effects of the two channels individually.

[0069] < 0: indicates interference suppression effect, that is, the channel combination produces redundancy or conflict, resulting in the reduction of effect.

[0070] = 0: indicates no interaction.

[0071] The specific calculation method is as follows: a. Statistical estimation method based on historical A / B testing: In historical marketing campaigns, the system intentionally designs an experimental group (receiving push notifications from both channels j and k) and a control group (receiving only channel j, only channel k, or no push notifications). By calculating the conversion rate difference between the experimental and control groups, and after eliminating the independent effects of each channel, the following results are obtained. The unbiased estimate. The formula can be simplified to:

[0072] in, The conversion rate of the experimental group. The conversion rate is used as the control group. The system periodically performs sparse testing on all possible channel combinations and uses Bayesian smoothing to handle data sparsity issues.

[0073] b. Machine Learning-Based Prediction Method: This method models the channel interaction effect as a supervised learning problem. The feature vector includes: attribute features of channels j and k (e.g., type, content format), user profile features, and temporal context. Training data comes from historical push logs, with labels representing the residual between the observed actual conversion rate and the sum of the model's predicted independent effect conversion rates. Training is performed using gradient boosting trees or shallow neural networks, and the model's predicted output is... The predicted values. This method can be generalized to channel combinations that have not been directly tested.

[0074] 2. Extended multi-objective optimization model: Based on the original optimization model, an interaction effect term is introduced. The objective function is then modified as follows:

[0075] The objective function described above aims to maximize the weighted sum of the expected conversion rate from a single channel and the additional expected revenue from pairwise interactions between channels. The model automatically weighs the following factors through a unified optimization framework: adding push channels for users to obtain positive interaction revenue (…). > 0), whether consuming the user's limited push quota will crowd out opportunities from other high-value independent channels; and the decision to split delivery channels to avoid negative interaction effects. All constraints (budget, user fatigue) work together in this global optimization process.

[0076] α is the weighting coefficient for the interaction effect. The remaining constraints (budget, user fatigue) remain unchanged.

[0077] Add new decision variables and constraints: It is a 0-1 auxiliary variable. ∈{0,1} indicates whether user i has been assigned to both channels j and k. Logical constraints need to be added to align it with the main decision variable. , Association, for example: >= + - 1 and <= min( , ).

[0078] The method for determining the weighting coefficient α: α is a parameter between 0 and 1, used to balance the relative importance of independent effects and interaction effects. The core assumption of this extended model is that the combined influence of multiple channels on users can be approximated by the linear weighted sum of their pairwise interaction effects. Its determination methods include: Grid search and cross-validation: On the historical dataset, split the dataset into training and validation sets, try different α values ​​(e.g., 0, 0.1, 0.2, ..., 1.0), and select the α value that maximizes the total marketing revenue simulated on the validation set as the system parameter.

[0079] Dynamic adjustment based on business rules: In the early stages of an activity or when data is insufficient, set a small α (e.g., 0.1) to focus on independent effects; as the system accumulates enough interaction effect data, gradually increase α (e.g., to 0.5) to enhance the exploration and utilization of combination strategies.

[0080] 3. Model Solving and Application: The extended optimization problem remains a (mixed) integer programming problem, but its scale is larger. It can be solved efficiently using decomposition algorithms (such as Lagrange relaxation) or heuristic algorithms (such as genetic algorithms with local search). The solution output... and They jointly defined a globally optimal push solution that not only considers the value of a single channel, but also proactively seeks positive collaboration and avoids negative interference.

[0081] The above steps, by modeling the non-independence between channels, upgrade the decision-making model from "isolated selection" to "combinatorial optimization." It solves the technical problem of traditional methods neglecting channel synergy, enabling proactive planning of channel combination sequences with positive synergistic effects (such as "public domain advertising traffic generation → private domain community building → one-to-one precise outreach"), while avoiding negative interference caused by content conflicts between channels or distracted user attention. This achieves globally optimal marketing resource allocation at the system level, significantly improving overall marketing efficiency in complex multi-channel environments.

[0082] Step S140: Intelligent push execution and control.

[0083] This step ensures that collaborative decisions are executed accurately and reliably, and protects the user experience.

[0084] S141, Task Scheduling and Distribution: Based on the scheme output in S132, the push execution engine generates a specific push task queue and sends the content out at the scheduled time through the official API of each channel or a reliable proxy.

[0085] S142, Push control rules include at least one of push frequency control, user fatigue management, time period restrictions, and content type restrictions.

[0086] Global rules: Set enterprise-level security limits, such as a daily limit on the total number of marketing messages received by all users.

[0087] Personalized fatigue model: This model trains a push frequency model based on users' historical interaction data and uses logistic regression or time series prediction models to dynamically output the push frequency that the user can currently accept. The model calculates a dynamic "user acceptance" score based on the user's recent responses to push notifications (such as clicking, ignoring, or disabling notification permissions). When the score falls below a threshold, the push frequency for that user is automatically reduced or the push of certain types of content is paused.

[0088] This step, while pursuing marketing effectiveness, introduces an intelligent management and control mechanism to effectively prevent excessive disturbance to users, reduce user churn and complaint risks, and ensure the long-term sustainability of marketing activities.

[0089] Step S150: Monitoring the effectiveness of all channels.

[0090] This step establishes a transparent and quantifiable feedback system for results.

[0091] S151, Full-link data tracking and attribution: Deploy tracking points at key nodes such as push messages, landing pages, and conversion pages to ensure that the complete path from exposure to conversion can be tracked and that the conversion credit can be reasonably allocated to different channels reached (such as using a time decay attribution model).

[0092] S152, Multi-dimensional Effect Analysis: Overall dashboard: Displays the core metrics of the event in real time (CTR, CVR, ROI).

[0093] Channel Dimension Analysis: Comparative analysis of the performance of different channels in terms of exposure, clicks, conversion, and cost.

[0094] User segmentation analysis: Analyzing the differences in response among different user profile groups.

[0095] S153, A / B Testing: The system supports A / B testing of different push strategies (such as different channel combinations and different copywriting) to scientifically evaluate the advantages and disadvantages of the strategies.

[0096] This step provides data-driven decision-making support, making marketing effectiveness measurable, analyzable, and comparable. It fundamentally changes the traditional "black box" nature of marketing and provides a clear direction for optimization.

[0097] Step S160: Iterative optimization of the strategy.

[0098] This step drives the marketing system to learn and evolve on its own.

[0099] S161, Data Feedback Closed Loop: The effect data monitored in S150 (especially conversion labels and negative feedback labels) is used as new training data and fed back to the user profile model (S114) to update user interests and preferences and the prediction model.

[0100] The performance data is updated in real time with user interest vectors through an online learning mechanism, specifically using incremental gradient descent to update the parameters of the collaborative filtering model; at the same time, the system triggers batch retraining of the profile model daily, integrating recent behavioral data and conversion tags to optimize user segmentation and prediction accuracy.

[0101] S162, Strategy Model Optimization: Content and channel strategy optimization: Based on A / B testing results and channel performance analysis, automatically adjust the parameters or weights of the content recommendation algorithm and channel selection model. For example, if a certain type of content is found to have a higher conversion rate on short video channels, then similar content will be given priority consideration to that channel in the future.

[0102] Push strategy optimization: Dynamically adjust model parameters based on the intervention effect of the user fatigue model (whether it reduces complaints and maintains conversion rate).

[0103] This step establishes a complete data loop from "execution" to "monitoring" and then to "optimization," enabling the entire marketing system to be adaptive, continuously learn from historical experience, and constantly improve the accuracy and efficiency of future marketing activities, thus realizing the intelligent evolution of marketing.

[0104] In a further embodiment, between step S150 and step S160, the following step is also included: Step S155: Evaluation of strategy effects and real-time correction based on incremental causal inference.

[0105] This step aims to overcome the relevance bias of traditional attribution models and accurately estimate the net causal effect of each specific push action on user conversion.

[0106] 1. Estimation framework for incremental causality (ICE): For a marketing campaign, the system estimates the ICE for each "processed" user i (i.e., one who received a specific strategy T, such as "channel A + copy B + time C"). ICE is defined as: The incremental causal effect of user i. It refers to the observed result (such as whether or not it is transformed). It is a counterfactual outcome—that is, the potential outcome if the user does not accept policy T.

[0107] 2. The core method of counterfactual estimation: because Since direct observation is not possible, the system uses one of the following two mainstream causal inference methods for estimation: Method A: Propensity Score Matching (PSM) Step 1: Calculate propensity scores. Use logistic regression or gradient boosting machine models to calculate user profile features. Using the unified user profile, pre-activity behavioral characteristics, and time context as independent variables, predict the probability that user i will accept strategy T, i.e., the propensity score. .

[0108] Step 2: Matching Control Groups. In the pool of users who did not receive policy T (they may have received other policies or received no push notifications), for each user i in the processing group, find one or more control group users j whose propensity scores are closest (typically using caliper nearest neighbor matching; the caliper is usually set to 0.2 times the standard deviation of the propensity score, e.g., ...). ).

[0109] Step 3: Effect Estimation. The mean ICE for the treatment group users is estimated as follows: ,in It refers to the number of people in the processing group. Let be the set of matching control group users for i, that is, the set of all matching users found from the control group within the neighborhood of the tendency score of user i in the treatment group. M is the number of matches, the number of control group users matched by each user in the treatment group (e.g., 1:1 or 1:K matching).

[0110] Method B: Dual Machine Learning (DML) Step 1: Data Preparation. Divide the dataset into two parts (or use cross-fitting) and define the features X, the processing variable W (whether to accept strategy T), and the outcome variable Y.

[0111] Step 2: Train two prediction models: Model g(X): Predicts E[W | X], which is the propensity score (i.e., the expectation that W=1 given feature X).

[0112] Model m(X): Predicts E[Y | X], which is a feature-based baseline outcome prediction (predicting outcome Y based solely on feature X under the assumption of no policy intervention).

[0113] Step 3: Effect Estimation. After eliminating confounding bias through orthogonalization (residualization), the average ICE of policy T can be obtained by solving a simple linear regression: The estimated coefficient θ represents the average causal effect of policy T (in the final estimation step of DML, the coefficient obtained through regression represents the average causal effect (ATE) of policy T on outcome Y). ε is the random error term. This method can more flexibly handle high-dimensional features and complex nonlinear relationships.

[0114] 3. Strategy Dimension Deconstruction and Root Cause Analysis: The system not only evaluates the overall effect of strategy T, but also deconstructs the strategy into multiple atomic dimensions (such as channel, content theme, push time, and discount strength). By fitting a causal tree or a DML-based heterogeneity treatment effect model, the system can estimate the heterogeneous causal effects of different user segments (defined by profile features) across various strategy dimensions. For example, the model might output: "For high-value users, the ICE (Income, Interest, and Cost) of pushing exclusive coupons via WeChat at lunchtime is significantly positive; however, for new users, the ICE of the same strategy is not significant." 4. Real-time error correction and strategy injection: High-leverage strategy accumulation: The strategy dimensions with significantly positive ICE and their applicable user segments are automatically packaged into "strategy templates" and stored in the strategy knowledge base for direct use in subsequent similar marketing campaigns or as initial strategies.

[0115] Invalid Policy Circuit Breaker: For policies with a significantly negative or zero ICE value, the system immediately sends a "circuit breaker" command to the push execution module (S140), automatically excluding the policy combination from subsequent push plans. A warning report is also generated.

[0116] Dynamic weight update: The average ICE estimate of each strategy dimension (such as a specific channel) is normalized and used as its "causal contribution weight". This is updated in real time to the channel feature library of the multi-channel collaborative decision-making module (S132) to replace or weight the original historical statistical weights (such as click rate). This allows channel selection to be directly based on causal contribution, making the optimization direction more accurate.

[0117] Step S155 addresses the fundamental problem common in existing effectiveness evaluation techniques (such as attribution models) that correlation does not equal causation. By introducing causal inference, the system can more accurately and quickly answer the core question, "What would happen if I didn't make this push?" thereby separating the true contribution of marketing actions from mixed market factors. This transforms strategy optimization from a "blind adjustment" based on potentially contaminated correlation data into a "fine-tuning" based on causal relationships, significantly accelerating strategy iteration, reducing trial-and-error costs, and ensuring that optimization always points to truly effective strategies, thus significantly improving the system's adaptive accuracy and the certainty of marketing ROI.

[0118] Example 2 Please see Figure 2 This diagram illustrates a structural block diagram of a marketing information management system based on multi-channel collaborative push, according to an embodiment of the present invention. The system 200 can be used to implement the method described in Embodiment 1, and is typically deployed in the cloud, including the following modules: Multi-channel data integration module 210: corresponds to method step S110. For example... Figure 3As shown, it is responsible for collecting data from various channels, cleaning, associating and integrating it, and calculating user profiles, which are then stored in the profile database 214.

[0119] Personalized content generation module 220: corresponds to method step S120. It calls up profile data and matches it from the content library 221 or generates personalized content through the engine 222.

[0120] Multi-channel collaborative decision-making module 230: corresponds to method step S130. For example... Figure 4 As shown, as the "brain" of the system, it uses the channel feature library 231 and the intelligent engine 232 to combine user profiles and activity goals to formulate the optimal multi-channel push plan, which is then processed by the conflict resolver 233.

[0121] The conflict resolver 233 is used to arbitrate and resolve potential resource conflicts in the push plan output by the optimization model. For example, when multiple high-priority marketing tasks compete for the same channel's limited sending quota during the same time period, the resolver dynamically schedules tasks based on their global strategic weight, real-time ROI prediction, and user fatigue status to ensure that critical tasks are executed first and maintain overall system efficiency. Its decision-making is based on a comprehensive score of multiple factors, including task priority, real-time ROI prediction, and user fatigue.

[0122] Intelligent push execution module 240: Corresponds to method step S140. It receives decision instructions, accurately executes pushes through scheduler 241 and adapter 242, and implements push restrictions by controller 243.

[0123] Omnichannel Performance Monitoring Module 250: Corresponds to method step S150. It collects performance data from each channel, analyzes it through the calculation engine 252, and visualizes it on the dashboard 253.

[0124] Strategy Iteration and Optimization Module 260: Corresponds to method step S160. Analyze the performance data, optimize the model through training platform 262, and schedule the new strategy to the relevant modules.

[0125] Each module works together through service interfaces and message queues to form an organic whole, achieving closed-loop management of multi-channel marketing activities.

[0126] Example 3 This invention also provides an electronic device, including: one or more processors; a storage device having one or more programs stored thereon; and when the one or more programs are executed by the one or more processors, causing the one or more processors to perform the method described in Embodiment 1.

[0127] Example 4 This invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in Embodiment 1.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A marketing information management method based on multi-channel collaborative push, characterized in that, include: Collect and integrate user data from multiple channels to build a unified user profile, including public and / or private channels; Based on the user profile, personalized marketing information content matching the user characteristics is generated; Based on the user profile and the channel characteristics of multiple alternative push channels, a push channel combination is selected and coordinated from the multiple alternative push channels; Personalized marketing information is pushed to target users through the aforementioned combination of push channels, and push restriction rules are imposed during the push process; Real-time monitoring of the effectiveness data after push notifications through different channels, and judgment of the overall and channel-specific push effectiveness based on preset key performance indicators; Based on the performance data, at least one of the following is optimized: the user profile, the strategy for generating personalized marketing information content, the strategy for forming the push channel combination, and the push restriction rules.

2. The method according to claim 1, characterized in that, The collection and integration of user data from multiple channels to build a unified user profile includes: Collect basic user behavior data from at least one public channel and / or collect in-depth user interaction and transaction data from at least one private channel; Clean and standardize data from different sources; Through correlation analysis and fusion algorithms, the processed data is integrated into a unified user profile that includes multi-dimensional tags of user attributes, behavioral preferences, and spending power.

3. The method according to claim 1, characterized in that, The process of generating personalized marketing information content that matches user characteristics based on the user profile includes: Analyze the user profile to extract the user's interests, preferences, and needs. Based on the aforementioned interests, preferences, and needs, content materials are matched from or dynamically generated from the content material library; The content materials are rendered in a personalized manner to suit different users or user groups, thus forming the personalized marketing information content.

4. The method according to claim 1, characterized in that, The step of selecting and coordinating push channel combinations from the multiple candidate push channels based on the user profile and channel characteristics includes: Maintain a channel feature library containing channel attributes and real-time status information; Using the user profile and marketing campaign objectives as input, collaborative optimization calculations are performed based on a preset multi-channel decision model to output the push channels and their collaborative execution strategies assigned to each target user or user group, thus forming the push channel combination.

5. The method according to claim 1, characterized in that, The process of pushing personalized marketing information to target users through the aforementioned push channel combination, and imposing push restriction rules during the push process, includes: Based on the channels and strategies determined by the aforementioned push channel combination, the interfaces of the corresponding channels are scheduled to execute content distribution tasks. Based on the global push frequency strategy and the individual user's interaction history, the push frequency and timing control for the target user are dynamically adjusted and executed.

6. The method according to claim 1, characterized in that, The real-time monitoring of the effect data after push through different channels, and the judgment of the overall and channel-specific push effect based on preset key performance indicators, includes: Collect user interaction data at each touchpoint in the push notification chain and associate it with the corresponding push notification channel; Calculate key performance indicators (KPIs) for the whole and for each channel, including impressions, click-through rate, conversion rate, and return on investment. A / B testing was used to compare the effectiveness of different combinations of push channels and content strategy versions.

7. The method according to claim 1, characterized in that, The optimization of at least one of the following based on the performance data—the user profile, the personalized marketing content generation strategy, the push channel combination formation strategy, and the push restriction rules—includes: The effect data is used as feedback to update the feature extraction model or label calculation model used to build user profiles; Analyze the characteristics of high-conversion content and channel combinations to optimize content generation strategies and multi-channel collaborative selection strategies; Based on user feedback on push notifications, the push frequency model and the degree of content personalization are dynamically adjusted.

8. A marketing information management system based on multi-channel collaborative push, characterized in that, include: The multi-channel data integration module is configured to collect and integrate user data from multiple channels to build a unified user profile, including public channels and / or private channels. The personalized content generation module is configured to generate personalized marketing information content that matches the user's characteristics based on the user profile. The multi-channel collaborative decision-making module is configured to select and coordinate a push channel combination from the multiple candidate push channels based on the user profile and the channel characteristics of the multiple candidate push channels. The intelligent push execution module is configured to push the personalized marketing information content to the target user through the push channel combination, and to apply push restriction rules during the push process; The omnichannel performance monitoring module is configured to monitor the performance data after push notifications are sent through different channels in real time, and to judge the overall and channel-specific performance based on preset key performance indicators. The strategy iteration and optimization module is configured to optimize at least one of the following based on the performance data: the user profile, the generation strategy of the personalized marketing information content, the formation strategy of the push channel combination, and the push restriction rules.

9. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program thereon, wherein the program, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images