A marketing management method and system based on big data

Abstract

The application relates to the technical field of marketing, and discloses a marketing management method and system based on big data, which comprises a dynamic situation sensing and demand prediction module: potential consumption demand is predicted based on authorized multi-source data; an active situation declaration and solution generation module: structured life task declarations actively input by users are received, and a solution package is generated; an individual utility preview and verification module: a product is subjected to multi-physical field simulation and multi-dimensional adaptability evaluation; a closed-loop feedback and ethical guarantee module: the system is continuously optimized and data compliance is guaranteed. Through the cooperative operation of the modules, the demand identification mode can be optimized, the marketing precision can be improved in combination with a double path, invalid guidance can be reduced, the user interaction experience can be improved, the decision cost can be reduced, the cold start problem can be alleviated, data compliance can be guaranteed through a multiple privacy protection mechanism, user trust can be enhanced, long-term iterative optimization of the system can be realized, and operation stability can be guaranteed.

Description

Technical Field

[0001] This invention relates to the field of marketing technology, and in particular to a marketing management method and system based on big data. Background Technology

[0002] Marketing management refers to the management process by which enterprises, in order to achieve their business objectives, systematically analyze the market environment, identify customer needs, design value propositions, plan marketing mix, and execute and control related activities to build long-term customer relationships and create sustainable competitive advantages. Its core lies in being customer-centric and integrating product, price, channel, and promotion strategies to ensure that enterprise resources are efficiently allocated to high-value market opportunities.

[0003] Existing technologies in personalized product recommendation and precision marketing suffer from a fundamental flaw: recommendation systems heavily rely on users' historical explicit behaviors such as purchases, clicks, and ratings, failing to effectively identify and respond to users' unexpressed latent needs arising in specific life situations. Mainstream technologies, such as collaborative filtering, content recommendation, or deep learning models, all assume that past behavior predicts future preferences, leading the system to repeatedly recommend products to users that they already know or are similar to. This makes it difficult to break through information silos, and when users face new scenarios such as their first trip abroad or changing jobs, the system's recommendation accuracy drops sharply due to the lack of historical behavioral data. Furthermore, existing solutions generally treat users as static profiles, ignoring the dynamic evolution of their needs with time, environment, and emotions. Some existing technologies attempt to introduce contextual information such as time and location, but mostly use simple rule matching, such as "promote leisure products on weekends," without building cross-domain contextual understanding capabilities; or propose reminder mechanisms based on calendar events, but these are limited to information prompts and are not deeply coupled with the product value system. These limitations cause companies to miss high-intent conversion windows, while users become averse to receiving irrelevant information, ultimately resulting in wasted marketing resources, low conversion efficiency, and a deterioration of customer experience. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a marketing management method and system based on big data. This method accurately predicts users' unexpressed product needs before they enter high-intent life scenarios and pushes highly relevant contextualized value proposals.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] On one hand, the present invention provides a big data-based marketing management system, comprising:

[0007] The system comprises the following modules: Dynamic Context Awareness and Demand Prediction Module: Predicts potential consumption needs based on authorized multi-source data when the user has not actively declared their intentions; Proactive Context Declaration and Solution Generation Module: Receives structured life task declarations actively input by the user and generates a one-stop solution package; Individual Utility Pre-simulation and Verification Module: Performs multi-physics simulation and multi-dimensional adaptability assessment of the product in real-world scenarios; Closed-Loop Feedback and Ethical Assurance Module: Continuously optimizes the system based on user behavior and feedback and ensures data compliance. The Dynamic Context Awareness and Demand Prediction Module and the Proactive Context Declaration and Solution Generation Module are mutually exclusive; when the user completes a proactive declaration, the system generates a solution solely based on the user's declaration.

[0008] As a preferred technical solution of the present invention, the dynamic context perception and demand prediction module includes: a context signal acquisition unit, which acquires locally desensitized structured context features from the user terminal; a context fusion engine unit, which generates a fixed-dimensional context vector by fusing multi-source features through a graph neural network; a cross-domain demand causal reasoning unit, which generates a demand distribution by combining a consumption context knowledge graph and counterfactual reasoning; and a value proposal triggering unit, which pushes a guidance message when the calibration demand intensity exceeds a threshold.

[0009] As a preferred technical solution of the present invention, the cross-domain demand causal reasoning unit in the dynamic context perception and demand prediction module includes: constructing a consumption context knowledge graph containing life scenarios, user roles, and product category nodes; setting edge relationships of causal types such as scenario-demand and role-adaptation, and attaching conditional constraints; simultaneously running observation path reasoning and counterfactual path reasoning, wherein the observation path reasoning is based on the user's historical behavior, and the counterfactual path reasoning is based on the current context; and weightedly fusing the demand probabilities of the two reasoning paths to generate the final demand distribution.

[0010] As a preferred technical solution of the present invention, the proactive context declaration and solution generation module includes: a context declaration guidance unit, which guides users to input the main scenario, structured parameters, and free description through a step-by-step form; a multi-granularity intent fusion and parsing unit, which merges structured and unstructured inputs into a standardized intent vector; a dynamic solution package generation unit, which organizes the solution according to the stages of the product's actual usage process, including pre-trip preparation, in-trip use, emergency handling, and effect maintenance; for each stage, it matches core products, auxiliary accessories, and non-commodity service suggestions to form a complete task-oriented solution package; it supports users to add the entire solution to their cart with one click and provides price optimization based on the packaging strategy; and a dual-mode collaborative scheduling unit, which coordinates the mutually exclusive execution of the user-declared path and the predicted path.

[0011] As a preferred technical solution of the present invention, the individual utility pre-demonstration and verification module includes: a spatial semantic parsing unit, which generates a semantically labeled 3D scene model from a monocular image; a product multi-physics digital twin unit, which constructs a product digital model integrating optical, mechanical, thermal, and surface properties; a user lifestyle coupling unit, which injects personalized evaluation criteria based on authorized tags; and a multi-dimensional adaptability evaluation unit, which generates a four-dimensional pre-demonstration report on vision, function, safety, and maintenance.

[0012] As a preferred technical solution of the present invention, the product multiphysics digital twin unit in the individual utility pre-simulation and verification module includes: the optical properties include material reflectivity and transmittance; the mechanical properties include structural stiffness and load-bearing deformation coefficient; the thermal properties include thermal conductivity and temperature rise response; the product digital model supports product dynamic behavior simulation, including changes in space occupancy, airflow diffusion, or evolution of protective effect.

[0013] As a preferred technical solution of the present invention, the closed-loop feedback and ethical protection module includes: an implicit behavior tracking unit that aggregates and statistically analyzes subsequent user behaviors without associating them with personal identities; an explicit feedback analysis unit that analyzes user evaluations and extracts structured improvement signals; a model incremental update unit that fine-tunes each sub-model online; a data minimization unit that enforces field filtering and dynamic consent management; and a fairness monitoring unit that audits output deviations between groups and triggers manual review.

[0014] On the other hand, the present invention also provides a marketing management method based on big data, comprising the following steps:

[0015] When the user does not actively declare, predict the demand based on authorized data and generate a guiding message;

[0016] Detect whether the user has completed the proactive context declaration;

[0017] If completed, generate a solution package based solely on the user's statement and disable prediction results;

[0018] If the prediction is not completed and the predicted intensity exceeds the threshold, a basic scheme will be generated based on the prediction results.

[0019] Perform individual utility simulations on selected products and generate multi-dimensional suitability reports;

[0020] We continuously optimize the model based on user feedback and behavior, while ensuring data ethics.

[0021] As a preferred technical solution of the present invention, the individual utility pre-simulation step includes: parsing scene semantics and spatial rules through SLAM and visual language model; constructing a multi-physics digital twin model of the product and simulating dynamic behavior; coupling user life tags or default standards for personalized evaluation; and outputting a scientifically based four-dimensional pre-simulation report and optimization suggestions.

[0022] As a preferred embodiment of the present invention, the continuous optimization model includes: aggregating implicit behavioral data; providing explicit feedback through natural language parsing; fine-tuning context awareness, demand reasoning, intent parsing, and product simulation models online; and periodically auditing the output fairness of different user groups.

[0023] The present invention has the following beneficial effects:

[0024] 1. Optimize demand identification methods and improve marketing accuracy: Abandon the traditional passive recommendation logic that relies on historical behavior backtracking, and adopt an active adaptation mechanism. Combining dynamic context awareness and active context declaration as dual paths, it can more accurately capture users' potential consumption intentions, reduce ineffective marketing guidance, and help improve the matching degree between marketing content and user needs.

[0025] 2. Improve user interaction experience and reduce decision-making costs: Through step-by-step guided scenario declaration design, personalized solution package generation, and interactive individual utility pre-demonstration verification, users can express their needs more efficiently, understand the actual utility of the product, reduce uncertainty in the user decision-making process, and improve user experience and decision-making efficiency.

[0026] 3. Alleviating the challenges of cold start and new scenario recommendations: The cross-domain demand causal reasoning unit combines consumption context knowledge graph and counterfactual demand generation technology. It does not need to rely excessively on users' historical behavior data. It can still achieve relatively accurate demand prediction in cold start scenarios and new consumption scenarios, thus expanding the applicability of the system.

[0027] 4. Ensure data processing compliance and enhance user trust: Through local data anonymization, retention of raw data on the device, data minimization control, and dynamic consent management mechanisms, user privacy can be effectively protected and privacy protection regulations can be met. At the same time, the traceability of utility simulation evaluation results helps to enhance users' trust in the system.

[0028] 5. Achieve long-term system optimization and improve operational stability: The closed-loop feedback and ethical protection module can achieve continuous iterative optimization of each functional module through implicit behavior tracking, explicit feedback analysis and incremental model update mechanism. At the same time, it can reduce service deviation through fairness monitoring and ensure long-term stable and compliant operation of the system.

[0029] 6. Enrich marketing logic and enhance long-term operational value: Compared to existing marketing systems that only focus on traffic distribution or content reach, this solution reconstructs the core marketing logic. Through one-stop solution delivery and long-term closed-loop optimization, it helps improve user conversion efficiency and helps to explore long-term customer lifetime value, providing more support for marketing operations. Attached Figure Description

[0030] Figure 1 The flowchart illustrates a big data-based marketing management system proposed in this invention.

[0031] Figure 2 A flowchart for the dynamic context awareness and demand forecasting module;

[0032] Figure 3 Flowchart for the proactive context declaration and solution generation module;

[0033] Figure 4 Flowchart for the individual utility simulation and verification module;

[0034] Figure 5 This is a flowchart for the closed-loop feedback and ethical protection module. Detailed Implementation

[0035] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0036] Reference Figure 1 This is a big data-based marketing management system, comprising a dynamic context awareness and demand prediction module, a proactive context declaration and solution generation module, an individual utility pre-simulation and verification module, and a closed-loop feedback and ethical assurance module. These modules collaborate efficiently through a unified data middleware and event bus, forming a complete technical closed loop of dual-path triggering, multimodal response, verifiable delivery, and adaptive optimization. It abandons the traditional passive recommendation logic that relies on historical behavior backtracking, instead adopting a proactive adaptation mechanism that combines multi-source authorized data fusion, cross-domain causal reasoning, and physical field simulation. This improves the demand identification method from passive response to proactive prediction. Compared to existing marketing systems that only focus on traffic distribution or content reach, this system reconstructs the essential logic of marketing, improving marketing accuracy, conversion efficiency, and long-term customer lifetime value.

[0037] Please refer to the appendix. Figure 2The dynamic context awareness and demand prediction module is an automated demand identification module for users who do not actively interact. It infers the user's high-probability future consumption intentions in real time based on authorized multi-source heterogeneous data, without any explicit user action, thus achieving seamless and proactive marketing guidance. This module includes a context signal acquisition unit, a context fusion engine unit, a cross-domain demand causal reasoning unit, and a value proposal triggering unit.

[0038] The context signal acquisition unit obtains context features from user terminal devices after local anonymization processing, ensuring that the data collection process complies with the principle of minimum necessity and privacy protection standards. Data sources include trip events in the operating system calendar, the place of residence and movement trajectory returned by the geolocation service, regional meteorological indices provided by the weather forecast application interface, physiological state summaries uploaded by wearable devices, and semantic keywords extracted from publicly available text on social platforms through local natural language processing. All raw data is retained only on the user device, and only structured feature vectors are uploaded to the server, such as the destination city code, expected length of stay, local UV intensity level, and tags of recently mentioned activities, effectively avoiding the leakage of raw privacy data. Specifically, the user is node u, the set of context elements is E={e1,e2,...,en}, the confidence level of each element ei is ρi∈(0,1], the timestamp is τi, and the current system time is tnow. Then the fused context vector su∈R d It can be represented as:

[0039] ;

[0040] Where ωi=ρi*exp(-λ(tnow-τi)), ϕ(.) is a learnable function that maps contextual elements to embedding vectors, ⨁ represents aggregation operations (such as weighted summation or attention pooling), λ>0 is the time decay coefficient, which controls the decay rate of the influence of historical events; GNNΘ is a graph neural network with parameter Θ, used to generate the final contextual representation.

[0041] The context fusion engine unit performs deep fusion on the collected structured feature vectors to generate a high-dimensional user context representation with fixed dimensions. This unit adopts a graph neural network architecture, with the user as the central node and various context elements as weighted edge connections. The edge weights are dynamically adjusted according to the confidence level of the elements. For example, the weight of trip events manually labeled by the user is set to a higher value, while the weight of events inferred only from location trajectories is set to a lower value. The fusion process introduces a time decay function, so that the influence weight of recent events on the current context is higher than that of distant events. The output is a context vector of fixed length, the dimension of which is determined according to the model training effect. In typical implementations, it can be 128-dimensional, 256-dimensional, or 512-dimensional, including static user attributes such as place of residence and age range, as well as dynamic attributes such as the itinerary for the next seven days and current health status.

[0042] The cross-domain demand causal reasoning unit infers users' potential product needs based on context vectors, effectively solving the problems of cold start and recommendation failure in new scenarios. This unit includes a consumption context knowledge graph subunit and a counterfactual demand generation subunit: the consumption context knowledge graph subunit pre-constructs a large-scale semantic network, with nodes covering life scenarios, user roles, product categories and specific models. The edge relationships are defined as causal types such as scenario-demand and role-fit, and are accompanied by conditional constraints. For example, the association weight between island vacation scenarios and the need for high-level waterproof and sun protection is set to a higher value, while the weight is reduced accordingly if the user frequently resides in tropical regions. The counterfactual demand generation subunit adopts a structural causal model, and simultaneously calculates the demand probability based on the user's historical behavior and the counterfactual path assuming no historical behavior and only based on the current situation. After weighted fusion, the final demand distribution is generated. Specifically, let D be a discrete product demand variable (taking values ​​from the product category set), x be a user situation vector, and h be a user historical behavior vector (which may be empty). Then, the final demand probability distribution P(D=d|x,h) is a weighted fusion of the observation path Pobs(d|x,h) and the counterfactual path Pcf(d|x).

[0043] ;

[0044] α = σ(||h||0 / κ), where ||h||0 is the number of non-zero historical behaviors, i.e., behavior sparsity; σ(.) is the Sigmoid activation function; κ>0 is the behavior salience threshold; when h is empty or extremely sparse, α→0, and the system mainly relies on counterfactual paths for prediction. This part can achieve accurate demand prediction without relying on users' historical behavior data, effectively overcoming the bottleneck of cold start technology.

[0045] The value proposal triggering unit decides whether to push guidance information based on the demand intensity calibration results. Calibration factors include situational urgency (e.g., trip departure countdown), user sensitivity (e.g., frequency of historical complaints), and external environmental variables (e.g., high-temperature warnings). The system only triggers a lightweight reminder when the demand intensity exceeds a preset threshold. For example, if the system detects that a user is about to travel to Sanya and needs a travel preparation guide, the user can click to enter the solution generation process. This avoids unsolicited commercial information pushes, protects user experience, and complies with privacy compliance requirements. Demand intensity I is represented as:

[0046] ;

[0047] Where D is the set of candidate product categories; Δt=tevent-tnow is the remaining time of the event; β(Δt)=1-exp(-μΔt) is the countdown activation function, μ>0 controls the rate of urgency growth; ςu∈[0,1] is the user sensitivity index (calculated based on historical complaints, unsubscriptions, etc.); γ(ςu)=1-ςu is the sensitivity inhibition factor; the system triggers a guidance message only when I>Ith (preset threshold).

[0048] Please refer to the appendix for details. Figure 3 The Proactive Context Declaration and Solution Generation Module is the structured intent expression and solution output unit for proactively interacting users. It guides users to efficiently declare their complex life tasks and generates highly relevant one-stop solution packages. This module runs in parallel with the Dynamic Context Awareness and Demand Prediction Module, complementing each other and jointly covering different user interaction preferences. This module includes a context declaration guidance unit, a multi-granularity intent fusion and parsing unit, a dynamic solution package generation unit, and a dual-mode collaborative scheduling unit.

[0049] The contextual declaration guidance unit adopts a step-by-step guided form design to reduce the cognitive load on users. This unit first guides users to select from preset main scenario types, which cover common life task categories such as travel, home renovation, gift-giving, health care, education and learning, and sports and fitness. Subsequently, the system dynamically loads the corresponding structured parameter fields according to the selected main scenario. For example, in the travel scenario, these include destination, departure date, number of people traveling with, mode of transportation, accommodation preferences, and special needs; in the home renovation scenario, they include house area, house type, decoration style preference, budget range, and functional priority. Finally, users can optionally add free descriptive text to express personalized needs that cannot be fully conveyed through structured fields.

[0050] All form fields provide intelligent default values ​​based on user profiles or context, and support real-time validity checks. If there are obvious conflicts or the input parameters are outside the reasonable range, the system will prompt the user to make corrections to ensure the completeness and accuracy of the context declaration.

[0051] The multi-granularity intent fusion parsing unit deeply integrates users' structured declaration data with free-form descriptive text to generate a high-precision user intent vector. This unit first maps structured fields to predefined standard context labels, ensuring the standardization and computability of intent expression. Second, it uses a domain-fine-tuned large language model to perform deep semantic parsing on the free-form descriptive text, extracting implicit needs not explicitly stated by the user. For example, if a user inputs "First time going to Sanya, afraid of sunburn," it can parse out fine-grained needs such as "high protection level," "waterproof and sweatproof," and "suitable for sensitive skin." Simultaneously, the system resolves intent conflicts discovered during parsing. For instance, when a user declares a high budget but prefers low-priced products, the core needs are determined based on historical behavior or contextual priority, and an interpretable intent summary is generated. This summary clarifies the weight and dependencies of each need item, avoiding intent ambiguity that could lead to solution adaptation deviations. Specifically, let the structured intent vector be vstruct∈R. d The free text, after being encoded by a large language model, yields a semantic vector vtext∈R. d Then, the intent vector `vintent` is fused using a gating mechanism:

[0052] ;

[0053] Where g = σ(W[vstruct; vtext] + b), W ∈ R d×2d b∈R d σ is a learnable parameter; σ(.) is the element-wise Sigmoid function; ⊙ is the Hadamard product (element-wise multiplication); g∈(0,1) d It is an adaptive gating weight vector that dynamically balances the contributions of structured and textual information.

[0054] The dynamic solution package generation unit, based on the integrated user intent, invokes the cross-product collaborative value network to generate a complete solution package containing core products, auxiliary accessories, and service suggestions. This solution package is organized according to the logical flow of the product in actual use. For example, travel-related tasks are divided into four stages: pre-trip preparation, in-trip use, emergency treatment, and effect maintenance; health care tasks are divided into symptom identification, intervention implementation, effect monitoring, and long-term management. For each stage, the system matches corresponding product combinations and non-commodity service suggestions. For example, under the intent of "Sanya travel + fear of sunburn," the pre-trip preparation stage includes high-SPF waterproof sunscreen and sunscreen lip balm; the in-trip use stage includes portable reapplication samples and sun hat recommendations; the emergency treatment stage includes after-sun repair masks and cold compress suggestions; and the effect maintenance stage includes a skin barrier repair guide. Users can add the entire solution package to their cart with one click, and the system provides price optimization based on the packaging strategy, improving decision-making efficiency and the purchasing experience.

[0055] The dual-mode collaborative scheduling unit manages the collaborative logic between the proactively declared path and the passively predicted path. When a user enters the system process through proactive declaration, the system prioritizes the explicitly declared context and needs as the core basis for solution generation, ignoring low-confidence prediction results from the dynamic context awareness module. When the user does not take proactive action, but the system detects a high-confidence life situation through the passive path, the prediction guidance mechanism is activated. The two types of paths operate independently without interfering with each other. The system dynamically selects the optimal service mode based on the user's current interaction state, ensuring both user control and basic service capabilities in scenarios without proactive declaration, achieving smooth collaboration and efficient connection between the two paths.

[0056] Please refer to the appendix. Figure 4 The Individual Utility Pre-simulation and Verification module is the value confirmation unit of this system. Before a user makes a product selection, it uses an interactive and quantifiable simulation and evaluation of the actual utility of the target product in its specific usage environment, based on the user's real-life scenarios and individual characteristics. This reduces decision-making uncertainty and improves product suitability and user trust. This module includes a spatial semantic parsing unit, a product multi-physics digital twin unit, a user lifestyle coupling unit, and a multi-dimensional suitability evaluation unit.

[0057] The spatial semantic parsing unit is used to perform structured analysis on real-life scene images or videos provided by users, identifying functional areas, key objects, and spatial constraints. This unit first generates a 3D point cloud model of the scene using monocular vision and simultaneous localization and mapping (SLAM) algorithms. Based on this, it calls a visual language model trained on scene understanding to label semantic objects in the point cloud, such as window orientation, power outlet location, high-frequency walking paths, children's activity areas, and furniture layout. Simultaneously, the system automatically extracts spatial rules related to product deployment, including but not limited to circulation width, comfortable viewing distance, lighting distribution characteristics, and safety isolation distance. These semantic information and rules together constitute a structured environmental context, providing fundamental support for subsequent product utility simulation.

[0058] Multiphysics digital twin units are used to construct high-fidelity digital models of products. These models not only include geometric shape and material properties, but also integrate physical attribute parameters closely related to product functionality. Specifically, the product's physical state evolves over time, satisfying the following:

[0059] ;

[0060] Where Ψ(r,t) represents the physical field at spatial location r and time t; πprod is the product's inherent physical parameter vector, ϵenv∈R qFor environmental context parameters, Fphys is an operator driven by physical laws or data, which can be solved numerically using the finite element method or neural operators. Among them, physical properties include optical properties such as material reflectivity and transmittance, mechanical properties such as structural stiffness and load-bearing deformation coefficient, thermal properties such as thermal conductivity and temperature rise response, and surface properties such as friction coefficient and stain resistance. Based on these parameters, the model supports dynamic behavior simulation, such as the space occupation changes of unfoldable furniture during operation, airflow or sound field diffusion during the operation of home appliances, and the evolution of the protective effect of personal care products under different light or humidity conditions. The simulation process is executed on user terminal devices or edge servers to ensure real-time performance and privacy, and avoid the risk of privacy leakage caused by uploading simulation data to servers.

[0061] The user lifestyle coupling unit is used to inject users' individual lifestyle characteristics into the product utility simulation process, realizing a leap from general adaptation to personalized verification. Users can selectively authorize the provision of lifestyle tags related to their lifestyle, such as family composition, daily routines, health status, occupational characteristics, or interests. The system activates corresponding special evaluation standards based on the authorized tags. For example, when the "children in the family" tag is detected, the child safety assessment rules are automatically activated, including the smoothness of corners, the risk of small parts falling off, and the non-toxic certification of materials. When the "working from home" tag is detected, the ergonomic verification process is initiated to assess whether the desktop height, monitor viewing distance, and sitting posture support meet occupational health standards. If the user does not authorize the provision of lifestyle tags, the system adopts the industry-standard safety and adaptation benchmarks as the default evaluation standards to ensure that basic verification capabilities are not affected.

[0062] The multi-dimensional adaptability assessment unit generates a comprehensive utility simulation report covering multiple dimensions based on the fusion results of the aforementioned spatial semantics, product digital twins, and user lifestyle patterns. The assessment dimensions include visual adaptability, functional adaptability, safety compliance, and long-term maintainability. Visual adaptability assesses the product's color coordination, style consistency, and degree of visual interference under current lighting and background environments; functional adaptability verifies the matching degree between the product's size, performance parameters, and user task requirements and spatial conditions; safety compliance identifies potential risk points, such as insufficient tipping stability, violations of electrical safety distances, and the presence of material allergens; long-term maintainability predicts the product's wear rate, cleaning difficulty, and performance degradation trend under typical usage intensity; the comprehensive adaptability score (Stotal) can be obtained through weighted summation.

[0063] ;

[0064] Where: M = {visual adaptability, functional adaptability, security compliance, long-term maintainability} is the set of evaluation dimensions; wm ∈ (0,1) is the weight of the m-th dimension, which can be adjusted by user preferences; Sm(.) ∈ [0,1] is the standardized scoring function of the m-th dimension; senv is the environmental semantic vector, and ℓuser is the user's life tag.

[0065] Users can view optimization suggestions for any evaluation item and preview the effects of the adjusted product configuration in real time. All evaluation conclusions are accompanied by traceable technical evidence, such as references to relevant national standards, material test data, or human factors engineering models, to enhance the credibility and persuasiveness of the evaluation results.

[0066] Please refer to the appendix. Figure 5 The closed-loop feedback and ethical assurance module is the long-term operation and compliance governance unit of this system. It is used to continuously optimize the performance of each functional module and ensure that the system meets the ethical requirements of privacy protection, fairness and transparency throughout the entire process of data processing, user interaction and decision output. This module includes an implicit behavior tracking unit, an explicit feedback analysis unit, a model incremental update unit, a data minimization unit and a fairness monitoring unit.

[0067] The implicit behavior tracking unit monitors users' subsequent behavioral patterns after interacting with system output content, providing objective evidence for model optimization. The behavioral signals collected by this unit include click responses to introductory messages, browsing depth of solution packages, interactive operations during individual utility simulations, and whether related products are ultimately added to the cart or purchased. All behavioral data is processed at the aggregation and statistical level, without being associated with individual user identifiers, making it impossible to infer a specific user's identity from behavioral sequences, ensuring that the tracking process meets anonymization and de-identification requirements.

[0068] The explicit feedback analysis unit processes user-submitted evaluations, ratings, or written opinions, extracting structured improvement signals. This unit performs natural language understanding and intent classification on user feedback, identifying positive or negative comments regarding solution suitability, product simulation authenticity, service suggestion usability, and interface usability, and transforms these into structured labels or weight adjustment instructions that can be recognized by the model training process. For example, when multiple users report that "sunscreen solutions do not consider sensitive skin," the system will generate an optimization task to "enhance skin type compatibility verification."

[0069] The model incremental update unit continuously iterates on each sub-model of the system based on implicit and explicit feedback data. This unit adopts an online learning or periodic fine-tuning mechanism to regularly update the dynamic context-aware model, demand causal reasoning graph, intent parsing language model, product physical simulation parameters, and multi-dimensional evaluation rule base. The update process supports canary releases and A / B testing. Let the model parameters be θ, the loss function be L(θ), and the new feedback data batch be Dnew. Then, the parameter update follows the following rules:

[0070] ;

[0071] Where η>0 is the learning rate; R(θ) is the regularization term (such as L2 regularization or knowledge distillation constraint) to prevent catastrophic forgetting; λreg≥0 is the regularization strength hyperparameter.

[0072] This ensures that while improving model performance, the stability of online services is not affected. Model version change records and effect evaluation reports are automatically archived to meet auditability requirements.

[0073] The data minimization unit is implemented throughout the system's data lifecycle, enforcing field-level filtering and usage restrictions. This unit ensures that data uploaded to the server contains only the structured features necessary to achieve specific functions, eliminating all raw information or redundant fields not directly related to marketing management. Simultaneously, the system incorporates a dynamic consent management mechanism, providing users with a visual data authorization panel that allows them to view their current authorization scope at any time and enable, suspend, or revoke permissions for data categories such as calendar, location, and health. Once a user revokes authorization for a certain type of data, the system immediately stops the collection, transmission, and model invocation of the relevant data, ensuring that data processing adheres to the principle of minimum necessity throughout.

[0074] The fairness monitoring unit is used to periodically evaluate the consistency of the system's output and the fairness of its services across different user groups. This unit performs distribution statistics and bias detection on the recommendation results, solution composition, and utility evaluation conclusions according to preset dimensions such as age range, gender, geographical region, device type, or spending power level. Let the group index be g∈G, and G be the set of all protected groups. Then, the average fit score bias of group g is set as follows:

[0075] ;

[0076] Where E[.] represents the mathematical expectation; if Δg>ϵ (ϵ>0 is the fairness tolerance threshold), then the correction mechanism is triggered.

[0077] If a particular group consistently receives significantly lower fit scores, limited product options, or missing key service recommendations, a manual review process is triggered. The algorithm ethics committee intervenes to investigate whether there is data bias, feature engineering defects, or model discrimination, and adjusts the training strategy or introduces corrective constraints accordingly. This mechanism ensures that the system pursues personalization without compromising the fairness of basic services.

[0078] Based on the aforementioned marketing management system based on big data, this invention also proposes a marketing management method based on big data, comprising the following steps:

[0079] Step S1: Dynamic Context Awareness and Demand Forecasting

[0080] S11: The context signal acquisition unit obtains context features that have been locally desensitized from the user terminal device. The original data is only retained on the device side, and only the structured feature vector is uploaded to the server.

[0081] S12: The context fusion engine unit adopts a graph neural network architecture, which uses the user as the central node and various context elements as weighted edge elements for fusion. It introduces a time decay function and outputs a fixed-dimensional context vector.

[0082] S13: The cross-domain demand causal reasoning unit is based on the context vector. Through the consumption context knowledge graph sub-unit and the counterfactual demand generation sub-unit, the demand probabilities of the observation path and the counterfactual path are weighted and integrated to generate the final demand distribution.

[0083] S14: The value proposal triggering unit calibrates the intensity of the demand based on the urgency of the situation, user sensitivity, and external environmental variables. If the intensity exceeds the preset threshold, a lightweight guiding message is generated, waiting for user interaction to trigger it.

[0084] Step S2: Determining the Source of Intent and Generating a Solution

[0085] S21: Detect whether the user has completed an active context declaration in the current session;

[0086] S22: If the user has completed the proactive context declaration, the following sub-steps are executed: S221: Obtain the user's main scenario type, structured parameters, and optional free description text through the context declaration guidance unit; S222: The multi-granularity intent fusion parsing unit parses the input into a standardized user intent vector; S223: Using the user intent vector as the sole basis, the dynamic solution package generation unit is invoked to generate a personalized solution package; S224: The demand distribution from the dynamic context awareness and demand prediction module is disabled as input for solution generation to ensure user control; S23: If the user has not completed the proactive context declaration, the following sub-steps are executed: S231: Obtain the calibrated demand distribution and intensity value from the dynamic context awareness and demand prediction module; S232: Verify whether the demand intensity exceeds a preset trigger threshold; S233: If it exceeds the threshold, the demand distribution is used as the intent basis to invoke the dynamic solution package generation unit to generate a basic solution package; S234: If it does not exceed the threshold, the current marketing guidance process is terminated, and no solution is generated to avoid unnecessary disruption.

[0087] Step S3: Individual Utility Pre-play and Verification

[0088] S31: The spatial semantic parsing unit generates a 3D point cloud model of the user's real scene using monocular RGB images and SLAM algorithm, and annotates semantic objects and extracts spatial constraint rules;

[0089] S32: The product multi-physics digital twin unit constructs a high-fidelity digital model of the target product, integrates physical property parameters such as optics, mechanics, and thermal, and supports dynamic behavior simulation;

[0090] S33: The user lifestyle coupling unit injects personalized simulation context based on user-authorized lifestyle tags or industry default standards; S34: The multi-dimensional adaptability evaluation unit generates a comprehensive pre-simulation report covering visual adaptability, functional adaptability, security compliance, and long-term maintainability based on the fused environment, product, and user data, and provides interactive optimization suggestions and real-time effect previews.

[0091] Step S4: Closed-loop feedback and ethical safeguards

[0092] S41: The implicit behavior tracking unit monitors users' subsequent behavior in response to guidance messages, solutions, and preview content based on aggregated statistics. All data is de-identified and not associated with personal identity.

[0093] S42: The explicit feedback analysis unit performs natural language parsing on user-submitted evaluations, ratings, or opinions to extract structured improvement signals;

[0094] S43: The model incremental update unit adopts an online learning mechanism to periodically fine-tune and optimize the context-aware model, demand reasoning graph, intent parsing model, and product simulation parameters.

[0095] S44: The data minimization unit enforces field filtering at each data processing stage, retaining only necessary features, and provides a visual authorization control interface through the dynamic consent management unit, allowing users to adjust or revoke data usage permissions at any time;

[0096] S45: The fairness monitoring unit periodically audits the distribution balance of the system output results according to dimensions such as age, region, and equipment type. If a significant deviation is detected, a manual review and model correction process is triggered.

[0097] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A marketing management system based on big data, characterized in that, include: Dynamic context awareness and demand prediction module: predicts potential consumption needs based on authorized multi-source data when users do not actively declare their intentions; Proactive Context Declaration and Solution Generation Module: Receives structured life task declarations proactively input by users and generates a one-stop solution package; Individual utility simulation and verification module: Performs multi-physics simulation and multi-dimensional adaptability evaluation of the product in real-world scenarios; Closed-loop feedback and ethical assurance module: Continuously optimizes the system based on user behavior and feedback and ensures data compliance; The dynamic context perception and demand prediction module and the proactive context declaration and solution generation module are mutually exclusive paths. When the user completes the proactive declaration, the system generates a solution based solely on the user's declaration.

2. The marketing management system based on big data according to claim 1, characterized in that, The dynamic context awareness and demand prediction module includes: The context signal acquisition unit acquires locally desensitized structured context features from the user terminal; The context fusion engine unit generates a fixed-dimensional context vector by fusing multi-source features through a graph neural network. Cross-domain demand causal reasoning unit, which combines consumption context knowledge graph and counterfactual reasoning to generate demand distribution; The value proposal trigger unit pushes a guiding message when the intensity of calibration demand exceeds a threshold.

3. The marketing management system based on big data according to claim 2, characterized in that, The cross-domain demand causal reasoning unit in the dynamic context awareness and demand prediction module includes: Construct a consumption context knowledge graph that includes life scenarios, user roles, and product category nodes, set causal edge relationships between scenarios and needs, roles and adaptations, and attach conditional constraints. Simultaneously run observation path reasoning and counterfactual path reasoning. Observation path reasoning is executed based on the user's historical behavior, while counterfactual path reasoning is executed based on the current context. The results of observation path reasoning and counterfactual path reasoning are weighted and fused to generate the final demand distribution.

4. The marketing management system based on big data according to claim 1, characterized in that, The proactive context declaration and solution generation module includes: The context declaration guidance unit guides users to input the main scenario, structured parameters, and free description through a step-by-step form; The multi-granularity intent fusion parsing unit fuses structured and unstructured inputs into a standardized intent vector; The dynamic solution package generation unit organizes solutions according to the stages of the product's actual usage process; for each stage, it matches core products, auxiliary accessories, and non-commodity service suggestions to form a complete task-oriented solution package; it supports users to add the entire solution to their cart with one click and provides price optimization based on the packaging strategy; The dual-mode collaborative scheduling unit coordinates the mutually exclusive execution of user-declared paths and predicted paths.

5. A marketing management system based on big data according to claim 1, characterized in that, The individual utility simulation and verification module includes: The spatial semantic parsing unit generates a semantically annotated 3D scene model from a monocular image; Multi-physics digital twin units for products are used to construct digital models of products that integrate optical, mechanical, thermal, and surface properties. User lifestyle coupling unit, injecting personalized evaluation criteria based on authorized tags; The multi-dimensional adaptability assessment unit generates a pre-performance report covering four dimensions: visual, functional, safety, and maintenance.

6. A marketing management system based on big data according to claim 5, characterized in that, The product multiphysics digital twin unit in the individual utility pre-simulation and verification module includes: The optical properties include material reflectivity and transmittance; the mechanical properties include structural stiffness and load-bearing deformation coefficient; and the thermal properties include thermal conductivity and temperature rise response. The product digital model supports simulation of dynamic product behavior, including changes in space occupancy, airflow diffusion, or evolution of protective effects.

7. A marketing management system based on big data according to claim 1, characterized in that, The closed-loop feedback and ethical protection module includes: Implicit behavior tracking unit, which aggregates and analyzes subsequent user behavior without being associated with personal identity; The explicit feedback analysis unit analyzes user reviews and extracts structured improvement signals. The model incremental update unit allows for online fine-tuning of each sub-model. Minimize data units, enforce field filtering and dynamic consent management; The fairness monitoring unit audits discrepancies among groups and triggers manual review.

8. A marketing management method based on big data, based on the marketing management system based on big data according to any one of claims 1-7, characterized in that, Includes the following steps: When the user does not actively declare, the dynamic context awareness and demand prediction module predicts the demand based on the authorized data and generates a guiding message. Detect whether the user has completed the proactive context declaration; If completed, the solution package is generated solely based on the user's statement through the proactive context declaration and solution generation module, and prediction results are disabled. If the prediction is not completed and the predicted intensity exceeds the threshold, a basic scheme will be generated based on the prediction results. The individual utility simulation and verification module is used to perform individual utility simulations on selected products and generate a multi-dimensional adaptability report. The closed-loop feedback and ethical protection module continuously optimizes the model and ensures data ethics based on user feedback and behavior.

9. A marketing management method based on big data according to claim 8, characterized in that, The individual utility pre-simulation steps include: Analyze scene semantics and spatial rules using SLAM and visual language models; Construct a multi-physics digital twin model of the product and simulate its dynamic behavior; Personalized assessments can be performed by coupling user lifestyle tags or default standards; Output a scientifically based four-dimensional pre-simulation report and optimization suggestions.

10. A marketing management method based on big data according to claim 8, characterized in that, The continuous optimization model includes: Aggregate and statistically analyze implicit behavioral data; Explicit feedback in natural language parsing; Online fine-tuning of context awareness, demand reasoning, intent parsing, and product simulation models; Regularly audit the fairness of output across different user groups.

Images