Marketing decision-making system driven by multi-source data

Abstract

The invention relates to a marketing decision system driven by multi-source data, and relates to the technical field of artificial intelligence, and the system comprises a state sensing module which is used for determining user state information representing the current state of a user according to the real-time interaction behavior of the user collected on terminal equipment; the intention acquisition module is used for acquiring marketing intention information representing a commercial target; the content generation module is used for dynamically generating marketing content by utilizing a generative model in response to the user state information and the marketing intention information; and the content presentation module is used for presenting the dynamically generated marketing content on an interface of the terminal equipment. According to the method and the device, the dynamic change of the user in a specific situation can be captured more accurately and more timely, so that the matching degree of the marketing content and the instant state of the user is remarkably improved, the accuracy and the real-time performance of a marketing decision are enhanced, and the flexibility and the intelligent level of the marketing decision are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a multi-source data-driven marketing decision-making system. Background Technology

[0002] With the rapid development of internet technology and e-commerce, automated marketing and personalized content recommendation systems have become key technologies for improving user interaction efficiency and business conversion rates. Existing technical solutions typically focus on pushing relevant marketing content to users based on their characteristics.

[0003] However, in the implementation of existing technologies, these systems primarily rely on relatively static data accumulated by users over a long period, such as registration information, historical purchase records, or explicit click behaviors, when making decisions. This data is used to construct a static user profile, which, while reflecting long-term user preferences to some extent, struggles to capture the user's immediate state and intent at a specific time and in a specific context. A user's psychological state, such as hesitation, curiosity, or excitement, is dynamic, and these immediate state changes contain crucial information influencing their decisions. Therefore, existing technologies, lacking the ability to effectively perceive the user's immediate state, often result in marketing content that mismatches the user's current needs, thus affecting the effective delivery of marketing information.

[0004] Furthermore, existing marketing systems often face a choice between two potentially conflicting goals when generating or selecting content: satisfying the platform's business objectives and ensuring a seamless user experience. Current technical solutions typically handle this relationship with preset, fixed strategies or weights—for example, always prioritizing business objectives over user experience, or vice versa. This rigid, one-size-fits-all strategy lacks flexibility and cannot be adjusted to different users and contexts. In some scenarios, overly aggressive commercial pushes may alienate users, while in others, an overly conservative strategy may miss business opportunities. Existing technologies lack an intelligent decision-making mechanism that can dynamically balance these two objectives based on specific circumstances.

[0005] Furthermore, most existing automated marketing systems essentially operate as open-loop systems. After a system executes a content delivery decision and action based on its internal model, it typically lacks an effective mechanism to evaluate the direct effects of that specific action and use those effects to automatically and instantly optimize its own decision-making model. While the system may perform periodic offline model updates, this approach fails to utilize the valuable feedback information generated from each user interaction. Consequently, the system model cannot continuously learn and evolve from its own practical experience, and its decision-making capabilities gradually become rigid as user behavior patterns evolve or the market environment changes, resulting in insufficient long-term stability and adaptability of the system's performance. Summary of the Invention

[0006] The purpose of this application is to provide a multi-source data-driven marketing decision-making system that solves the problem that existing automated marketing systems have difficulty in effectively integrating users' real-time dynamic status and changing business objectives when generating content, resulting in insufficient accuracy in content matching and adaptability in decision-making.

[0007] Firstly, a multi-source data-driven marketing decision-making system is provided, which includes: A state awareness module, which is connected to the data acquisition interface of the terminal device, is used to determine user state information representing the current state of the user based on the real-time user interaction behavior collected from the terminal device. The intent acquisition module is used to obtain marketing intent information that represents business objectives from the business rules database or user input interface; The content generation module has its input terminals connected to the output terminals of the state awareness module and the intent acquisition module, respectively. It is configured to dynamically generate marketing content using a generative model in response to the user state information and the marketing intent information. The content presentation module has its input end connected to the output end of the content generation module, and its output end connected to the display interface of the terminal device, for presenting the dynamically generated marketing content on the interface of the terminal device.

[0008] In one embodiment, the state-aware module is specifically used to perform the following operations: collect the user's micro-interaction features to form a feature sequence. And through a pre-defined time series analysis model Process the feature sequence To generate the user state information, wherein the user state information is an emotion vector. Its generation process is represented by the formula: ; in, It is an emotion vector, and its vector dimension corresponds to a preset emotion type; For time series analysis models, such as recurrent neural networks or Transformer networks; It is a feature sequence that includes one or more features extracted from the user's real-time interaction behavior, such as swiping speed, pressure intensity, or trajectory pattern.

[0009] In one embodiment, the intent acquisition module is specifically used to structure a preset business objective, including target products, marketing strategies, and brand tone, into a marketing intent vector as the marketing intent information.

[0010] In one embodiment, the content generation module determines optimal marketing content by optimizing a comprehensive utility function. The comprehensive utility function includes a user experience objective related to the user state information and a business objective related to the marketing intent information.

[0011] In one embodiment, the comprehensive utility function is expressed through a dynamic fusion coefficient. For the user experience target item and the aforementioned business objectives Perform a weighted combination. Its functional form is expressed by the formula: ; in, For the comprehensive utility function, For marketing content to be evaluated, For dynamic fusion coefficients, This is a user experience objective, and its numerical value represents the degree of matching between the marketing content to be evaluated and the user's state information. For business objectives, the numerical value represents the degree of match between the marketing content to be evaluated and the marketing intent information.

[0012] In one embodiment, the dynamic fusion coefficient By a policy function Based on user's historical profile information and current user status information It is determined dynamically. Its determination method is expressed by the formula: ; in, For dynamic fusion coefficients, For the policy function, The historical profile information extracted from the user database that characterizes the user's long-term behavioral features. The user state information determined by the state awareness module.

[0013] In one embodiment, the system further includes a feedback optimization module. The input of this module is connected to the data acquisition interface of the terminal device, and its output is connected to the state awareness module or the content generation module. The feedback optimization module is used to collect subsequent user interactions with the marketing content presented by the content presentation module, and to generate optimization parameters based on these interactions to update the time-series analysis model in the state awareness module or the strategy function in the content generation module.

[0014] Secondly, this application provides a multi-source data-driven marketing decision-making method, which includes the following steps: S1. Determine user status information representing the user's current state based on the real-time user interaction behavior collected on the terminal device; S2. Obtain marketing intent information that represents business objectives; S3. In response to the user status information and the marketing intent information, dynamically generate marketing content using the generative model; S4. The dynamically generated marketing content is displayed on the interface of the terminal device.

[0015] In summary, this application includes at least one of the following beneficial technical effects: 1. This application, by setting up a state awareness module, can collect and analyze users' micro-interaction behaviors on terminal devices in real time, and convert them into user state information representing the user's immediate emotional and intentional state. Compared with traditional technologies that rely solely on static user profiles or discrete behavioral events, this invention can more accurately and timely capture the dynamic changes of users in specific situations, thereby significantly improving the matching degree between marketing content and users' immediate state, and enhancing the accuracy and real-time nature of marketing decisions; 2. This application achieves an intelligent and dynamic balance between user experience goals and commercial marketing goals by introducing a comprehensive utility function controlled by a dynamic fusion coefficient into the content generation module. The strategy function can dynamically adjust the emphasis on both based on the user's long-term profile and current status, avoiding the rigidity of traditional fixed strategies. This allows the system to flexibly adopt generation strategies that emphasize either empathy or objectives in different contexts, improving the flexibility and intelligence of marketing decisions. 3. By setting up a feedback optimization module, this application can use the user's subsequent interaction with the generated content as a feedback signal to continuously iterate and optimize the time-series analysis model in the state awareness module or the strategy function in the content generation module. This enables the application to learn and evolve from its own decision-making practice, and to automatically adapt to changes in user behavior patterns and market environment, thereby improving the long-term effectiveness and automation level of the system. Attached Figure Description

[0016] Figure 1 This is the system architecture diagram of this application; Figure 2 This is the flowchart of the method used in this application. Detailed Implementation

[0017] Combined with appendix Figure 1 This application will be described in further detail below.

[0018] Example: A multi-source data-driven marketing decision-making system, comprising: The state awareness module is used to determine user state information that represents the user's current state based on the real-time user interaction behavior collected on the terminal device. The input end of this module establishes a communication connection with the data acquisition interface of the terminal device. Its substantive function is to convert the continuous, physical-level real-time interactive behaviors performed by the user on the terminal device interface into a machine-understandable structured information that represents the user's current psychological state, namely user state information.

[0019] First, during the data acquisition phase, this module continuously captures raw user interaction data streams from the front-end interface of the terminal device. It's important to note that this interaction data stream is not merely discrete click or purchase events, but rather a continuous, high-frequency sampled sequence of micro-behaviors over a period of time. In one optional implementation, this raw interaction data stream can consist of a series of time-ordered tuples of behavioral events. Each tuple records various physical parameters at the moment of the interaction, such as a timestamp, screen coordinates of the touch point, displacement vector during swiping, and screen pressure value on pressure-sensitive devices.

[0020] After obtaining the raw interaction behavior data stream, in order to extract deep features that can effectively reflect the user's psychological state, the state awareness module will perform feature engineering on the data stream. This process is carried out within a preset time window, converting the raw data within the window into one or more feature vectors.

[0021] Microscopic interaction features may include, but are not limited to, one or more of the following: Sliding speed and acceleration: These represent the rate of change in a user's browsing speed and can be used to distinguish between aimless browsing and the behavior of quickly searching for a target.

[0022] Trajectory shape: such as the curvature or smoothness of the sliding trajectory, can be used to reflect the decisiveness of the user's operation.

[0023] Dwell time: The duration of a user's visual or operational presence on a specific area or element of the interface.

[0024] Changes in pressure applied: The dynamic changes in the pressure applied by the user when pressing the screen during interaction can reflect the intensity of their interest or willingness to confirm.

[0025] Zoom gesture parameters: such as the frequency and magnitude of zooming in or out of an image or area, can reflect the user's attention to detail.

[0026] Through the feature engineering process described above, the raw data stream collected within a time window is transformed into a feature sequence with a time-series relationship. .

[0027] Subsequently, the core step of the state-aware module lies in utilizing a pre-defined time-series analysis model. For feature sequences The model aims to learn and identify the specific emotional or intentional states implied behind different sequences of interactive behavior patterns. Preferably, a time-series analysis model is used. It can be a pre-trained deep neural network model, such as the encoder structure in a gated recurrent unit network, a long short-term memory network, or a Transformer network. These model structures are good at capturing and processing time dependencies in sequential data.

[0028] The model receives feature sequences. As input, it outputs a standardized, multi-dimensional user state information. In this embodiment, the user state information is specifically an emotion vector. This process can be formally expressed by a formula: ; In this formula: This represents the final output sentiment vector. This vector is... A multidimensional real vector, where each dimension corresponds to a predefined basic sentiment or intention category, and the value on that dimension quantifies the strength of the association between the user's current state and that sentiment category.

[0029] This represents a pre-defined time series analysis model.

[0030] This represents a feature sequence that has a temporal relationship as input to the model.

[0031] In this way, the state awareness module realizes the transformation from low-level, physical interaction signals to high-level, abstract emotional state information, and its output emotion vector This provides crucial information about the user's real-time status for subsequent content generation modules, enabling the system to surpass traditional recommendation methods based on static user profiles and achieve more dynamic and accurate content matching.

[0032] The intent acquisition module is used to acquire marketing intent information that represents business objectives; The function of this module is to provide a clear and structured business objective constraint for the subsequent content generation module, ensuring that the final generated marketing content not only adapts to the user's state but also serves the preset business purpose.

[0033] Specifically, the intent acquisition module obtains raw marketing tasks through one or more preset interfaces. In one optional implementation, these interfaces may include an interface for connecting to a business rules database, which can automatically trigger marketing tasks based on conditions such as inventory levels and product launch dates. In another optional implementation, these interfaces may include a human-computer interaction interface, such as a back-end management system for marketing operations personnel, through which they can manually configure and launch marketing campaigns.

[0034] The initial marketing task here is typically unstructured or semi-structured information, the content of which aims to describe the core elements of a marketing campaign. As an example, these elements may include, but are not limited to, one or more of the following: The target audience specifies one or more specific products or categories of products that this marketing campaign is targeting.

[0035] The core strategy defines the main purpose of this marketing campaign, such as brand image promotion, new product marketing, or clearance sales for specific products.

[0036] Brand tone defines the brand image or style that needs to be conveyed in marketing content, such as luxury, approachability, trendiness, or professionalism.

[0037] Urgency defines the urgency or time constraint of a marketing campaign, which influences the tone or call to action strength of subsequent content.

[0038] After acquiring the aforementioned high-level marketing tasks, the core function of the intent acquisition module is to perform a structured transformation process. The purpose of this process is to convert these diverse and descriptive business objectives into a standardized, machine-processable data format. In this embodiment, this standardized data format specifically refers to marketing intent information, more specifically, an m-dimensional marketing intent vector. .

[0039] Marketing Intent Vector Each dimension is designed to correspond to a specific, quantifiable marketing parameter. To achieve the transformation from the original task to a vector, the intent acquisition module has corresponding coding logic configured internally.

[0040] Preferably, different encoding methods can be used for different types of marketing task elements. For ordered or continuous data such as urgency, numerical mapping or normalization can be performed directly and then filled into the corresponding dimensions. For identifier information such as target objects, their IDs or codes can be directly used as a dimension value of the vector.

[0041] Through the above structuring and vectorization processes, the originally abstract business objective is transformed into a precise mathematical expression. This marketing intent vector... It fully encapsulates all the key constraints of this marketing mission.

[0042] Ultimately, the intent acquisition module will generate a marketing intent vector. Output to the content generation module. This vector is compared with the sentiment vector from the state-aware module. Together, these factors form the basis for the content generation module's decision-making and creation. This allows the system of this invention to generate content within a constraint space jointly defined by the user's current state and business objectives, thereby ensuring the dual relevance of the output results.

[0043] The content generation module dynamically generates marketing content using a generative model in response to user status information and marketing intent information. The input of this module is connected to the output of the state awareness module and the intent acquisition module, respectively, so that after receiving user state information representing the user's current state and marketing intent information representing the business goal, a dynamic content generation process is executed.

[0044] This content generation module is a decision-making and generation system based on optimization theory. Its core working principle is to model the content generation process as a process of solving an optimization problem, that is, to find the optimal marketing content that maximizes a preset comprehensive utility function within a huge content space defined by a generative model.

[0045] Specifically, the module receives an emotion vector from the state-aware module. And the marketing intent vector from the intent acquisition module This serves as the basis for their decision-making. Their goal is to generate marketing content comprised of visual and textual elements. This marketing content It is the comprehensive utility function defined by the following formula. The optimal solution: ; In the definition of this comprehensive utility function, each component has a clear technical connotation.

[0046] Among them, user experience target items Intended for candidate marketing content The degree of matching with the user's current emotional state is quantitatively evaluated. In a specific implementation, in order to calculate... The module contains a content sentiment assessment model. This model can analyze candidate content. The visual style, color scheme, composition, and tone and wording of the copy are mapped to emotional vectors. Within the same n-dimensional sentiment vector space. Subsequently, the sentiment mapping vector of the content is calculated and compared with the user's sentiment vector. The cosine similarity between them can be used to obtain a standardized value. Numerical value.

[0047] At the same time, business objectives Intended for candidate marketing content Satisfy the preset marketing intent vector The extent to which marketing intent is quantified is crucial. Since marketing intent itself is multi-dimensional, The calculation is typically implemented as a weighted sum of the achievement rates of multiple sub-objectives. Preferably, this module can integrate multiple dedicated evaluators, for example, using an image classification model to determine whether the style of the generated image conforms to... The text specifies the brand tone; it utilizes keyword matching or intent recognition capabilities within a natural language processing model to determine whether the generated copy contains the brand tone. The necessary promotional information or calls to action specified in the document. The evaluation results of each sub-objective are weighted and summed to obtain the final result. Numerical value.

[0048] Of particular importance is the dynamic fusion coefficient in the comprehensive utility function. This is key to achieving a dynamic balance between user experience and business objectives. This coefficient is not a preset, fixed constant, but rather determined by an independent strategy function. The policy function is dynamically calculated based on the specific context of each interaction. The calculation process can be formally expressed by the following formula: ; In this formula: The policy function can be a lightweight neural network or a parameterized logistic regression model.

[0049] This represents historical profile information extracted from a user database that characterizes the user's long-term value or behavioral preferences. The purpose of introducing this is to make the system's decisions more long-term and strategic.

[0050] This represents user state information, provided in real-time by the state awareness module, that characterizes the user's current state. The purpose of introducing this is to enable the system's decisions to adapt to real-time contexts.

[0051] This strategy function Output value Constrained between 0 and 1, its function is equivalent to a dynamic regulator. When When the value of approaches 1, the overall utility function will be mainly composed of user experience objective items. The dominant approach will prioritize generating content that maximizes user emotional resonance; conversely, when... When the value of approaches 0, the business objective item This will take a dominant role, with the system focusing more on generating content that effectively achieves marketing intent. This allows the system to intelligently make trade-offs in different interaction scenarios.

[0052] Guided by the aforementioned optimization framework, this module utilizes one or more internally configured generative models to actually produce marketing content. Preferably, this module may include an image generation model for generating visual elements, such as a conditional generative adversarial network or a latent diffusion model, and a language generation model for generating text elements. Both generative models use sentiment vectors... Marketing Intent Vector The basic information of the target product is used as input conditions to ensure that the generation process is controlled and the target is clear.

[0053] In summary, the content generation module, through a well-defined optimization process, merges the potentially conflicting goals of user state and commercial intent, and drives the generative model to create the optimal solution in this context—the optimal marketing content. It is then output to the content presentation module.

[0054] The content presentation module is used to display dynamically generated marketing content on the interface of the terminal device.

[0055] The input end of this module establishes a data connection with the output end of the content generation module. Its fundamental purpose is to effectively render and display the optimal marketing content produced by the content generation module after complex optimization decisions on the user interface of the terminal device.

[0056] Specifically, the content presentation module receives the optimal marketing content output by the content generation module as input. It should be noted that this marketing content... It is a composite data structure that encapsulates dynamically generated visual elements and their matching text elements.

[0057] Upon receiving marketing content Subsequently, the core function of the content presentation module lies in seamlessly integrating it into the front-end user interface of the terminal device. This process is not a simple static page loading, but a dynamic process of replacing interface elements. Preferably, in order to complete the content update without interrupting the user's current browsing flow or causing a complete interface refresh, this module can utilize asynchronous data communication and front-end rendering technology.

[0058] As an example, this module communicates with the system backend to receive content. Furthermore, by utilizing document object model manipulation techniques, the system precisely locates and replaces pre-defined placeholder elements in the user interface used to display marketing content. In this way, the system can achieve real-time, dynamic, and partial updates of content, ensuring that the marketing information seen by the user is highly relevant to its real-time state.

[0059] More importantly, the content presentation module plays a crucial, bridging role in the overall closed-loop workflow of the system. Its content presentation action not only marks the end of the system's decision-making process but also serves as the starting point for the next round of perception and optimization. This module generates content... The behavior presented to the user is essentially a way of conveying new information to the user.

[0060] Subsequent user interactions after viewing the presented content, such as careful viewing, rapid scrolling, clicking, or new browsing patterns arising from these interactions, will be captured again by the terminal device's data collection interface. This data on subsequent interactions constitutes the most direct and authentic implicit feedback on the content generation and presentation effect.

[0061] In a preferred embodiment, this feedback data is transmitted to the feedback optimization module included in the system of the present invention, serving as the core basis for the module to perform model optimization. This provides indispensable practical verification and data input for the continuous evolution and self-optimization of the entire system.

[0062] The feedback optimization module is used to collect users' subsequent interaction behaviors with the marketing content presented by the content presentation module, and to optimize the state awareness module or content generation module based on the subsequent interaction behaviors.

[0063] The starting point for this feedback optimization module is to dynamically generate the best marketing content in the content presentation module. After being displayed to the user, this module collects data on the user's reaction upon encountering the new content. The subsequent interactions that follow. These subsequent interactions differ in nature from those used for initial state awareness; they reflect a specific decision made by the user in response to the system (i.e., the generation and presentation of content). (instant response)

[0064] As an example, these subsequent interactive behaviors may include: Users performed clear, positive actions such as clicking and converting on the presented content.

[0065] Users spent significantly more time or repeatedly viewed the presented content areas than average.

[0066] Users exhibited clear negative behaviors such as quickly swiping over or ignoring the presented content.

[0067] The overall interaction pattern of users after encountering content undergoes a measurable change, such as from hesitant to decisive.

[0068] After collecting these subsequent interaction behaviors, the core task of the feedback optimization module is to convert these raw behavioral data into one or more feedback signals that can be used for model training. This feedback signal is a quantitative evaluation of the effectiveness of the content generation and presentation behavior. Preferably, this module can map positive behaviors to positive reward signals and negative behaviors to negative reward or penalty signals.

[0069] After receiving valid feedback signals, the module uses the practical experience gained from this interaction to fine-tune and iterate the core model of the system. In this embodiment, the optimization process mainly targets the following two objectives: Firstly, it involves the timing analysis model in the state-aware module. This module optimizes the model because the initially trained model may be biased in its association with certain interaction patterns and emotional states. By collecting a large number of data pairs showing specific content and triggering specific subsequent interactions, this module can help correct this bias.

[0070] Secondly, and more importantly, is the strategy function in the content generation module. Optimize the system. This optimization process aims to teach the system how to better balance user experience and business objectives.

[0071] This module adopts an online learning paradigm, treating each interaction and its feedback result as a micro-batch of training samples to perform real-time, small-amplitude gradient updates on the parameters of the above model.

[0072] In summary, by establishing a data feedback path from action results to the decision model, the feedback optimization module enables this invention to learn and evolve from its own practice, ensuring the long-term effectiveness of its internal model and its adaptability to changes in the user and business environment.

[0073] Combined with appendix Figure 2Another embodiment of this application provides a multi-source data-driven marketing decision-making method, including the following steps: S1. Determine user status information representing the user's current state based on real-time user interaction behavior collected on the terminal device; S2. Obtain marketing intent information that represents business objectives; S3: Respond to user status information and marketing intent information, and dynamically generate marketing content using generative models; S4. Present dynamically generated marketing content on the interface of the terminal device.

[0074] The method in this embodiment can be used to execute the above system embodiment, and its principle and technical effect are similar, so it will not be described again here.

[0075] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Identical components are represented by the same reference numerals. Therefore, all equivalent changes made to the structure, shape, and principle of this application should be covered within the scope of protection of this application.

Claims

1. A multi-source data-driven marketing decision-making system, characterized in that, include: The state awareness module is used to determine user state information that represents the user's current state based on the real-time user interaction behavior collected on the terminal device. The intent acquisition module is used to acquire marketing intent information that represents business objectives; The content generation module dynamically generates marketing content using a generative model in response to the user status information and the marketing intent information. The content presentation module is used to present the dynamically generated marketing content on the interface of the terminal device.

2. The multi-source data-driven marketing decision-making system according to claim 1, characterized in that, The state awareness module is specifically used to collect the user's micro-interaction features to form a feature sequence, and to process the feature sequence through a preset time series analysis model to generate the user state information, wherein the user state information is an emotion vector, and its generation process is represented by the following formula: ； in, For sentiment vectors, For time series analysis models, It is a characteristic sequence.

3. The multi-source data-driven marketing decision-making system according to claim 1, characterized in that, The intent acquisition module is specifically used for: The pre-defined business objectives, including target products, marketing strategies, and brand tone, are structured into a marketing intent vector to serve as the marketing intent information.

4. The multi-source data-driven marketing decision-making system according to claim 1, characterized in that, The content generation module generates the marketing content by optimizing a comprehensive utility function; The comprehensive utility function includes user experience objectives related to the user state information and business objectives related to the marketing intent information.

5. A multi-source data-driven marketing decision-making system according to claim 4, characterized in that, The comprehensive utility function uses dynamic fusion coefficients to weight and combine the user experience objective and the business objective, and its functional form is as follows: ； in, For the comprehensive utility function, For marketing content to be evaluated, For dynamic fusion coefficients, For user experience goals, For business objectives.

6. The multi-source data-driven marketing decision-making system according to claim 5, characterized in that, The dynamic fusion coefficient is dynamically determined by the strategy function based on the user's historical profile information and current user status information, and the determination method is as follows: ； in, For dynamic fusion coefficients, For the policy function, For users' historical profile information, This refers to user status information.

7. The multi-source data-driven marketing decision-making system according to claim 1, characterized in that, The generative model in the content generation module includes: Image generation models used to generate visual content; A language generation model used to generate text content.

8. A multi-source data-driven marketing decision-making system according to claim 1, characterized in that, The marketing content includes: Visual elements related to the target product; Marketing copy that matches the visual elements.

9. A multi-source data-driven marketing decision-making system according to claim 1, characterized in that, Also includes: The feedback optimization module is used to collect users' subsequent interaction behaviors with the marketing content presented by the content presentation module, and optimize the state perception module or the content generation module based on the subsequent interaction behaviors.

10. A multi-source data-driven marketing decision-making method, comprising a multi-source data-driven marketing decision-making system according to any one of claims 1-9, characterized in that, Includes the following steps: S1. Determine user status information representing the user's current state based on real-time user interaction behavior collected on the terminal device; S2. Obtain marketing intent information that represents business objectives; S3. Responding to the user status information and the marketing intent information, dynamically generate marketing content using a generative model; S4. The dynamically generated marketing content is displayed on the interface of the terminal device.

Images