AI-based new media intelligent marketing method

Abstract

The present application relates to the technical field of artificial intelligence driven digital marketing, especially to an AI-based new media intelligent marketing method, comprising: step 1: cross-platform user behavior fusion modeling; step 2: multi-modal content gene library construction: analyzing the visual focus area and text emotional tendency of historical content, generating content gene vectors; step 3: dynamic content generation and real-time optimization: retrieving the gene library according to the space-time behavior matrix to generate initial marketing content; real-time monitoring of user interaction behavior, dynamically replacing the elements of the visual focus area and adjusting the emotional intensity of the text; step 4: closed-loop strategy optimization: calculating the dynamic effect index by fusing multi-platform indicators, triggering the gene mutation mechanism when the index continuously decreases, and generating cross-platform compensation content based on the failed user behavior. By analyzing the cross-platform behavior and content preferences of users, the user demand can be accurately predicted, and the content more in line with the individualized preferences of users can be provided.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence-driven digital marketing technology, and in particular to AI-based intelligent marketing methods for new media. Background Technology

[0002] In the field of AI-driven digital marketing, especially in cross-platform new media intelligent marketing scenarios, existing technologies suffer from the following structural defects:

[0003] 1. Fragmented user behavior across platforms leads to distorted user profiles;

[0004] Mainstream marketing systems rely on single-platform behavioral data (such as Douyin viewing history and WeChat Moments interactions) to build user interest models. When modeling, they only use weighted behaviors within the platform (e.g., video completion rate weight > like weight), failing to capture users' related behaviors across multiple platforms such as Weibo and Xiaohongshu.

[0005] Due to the closed data interfaces and privacy policies of various platforms, the user's cross-platform behavior chain is forcibly broken. For example: After watching a video about new energy vehicles on Douyin, User A searches for similar products on Xiaohongshu the next day. The existing system cannot establish this connection, resulting in:

[0006] One-sided user profile: Douyin categorizes users as having "light interests" while ignoring the strong purchase intent on Xiaohongshu;

[0007] Resource mismatch: Pushing low-level traffic-driving content to users instead of conversion-oriented content.

[0008] Platform data silos are an inherent part of the industry ecosystem, and no single service provider can force access to raw cross-platform data. Current technology lacks a quantitative mechanism for the temporal correlation of cross-platform behavior, and this problem continues to worsen in the multi-platform coexistence environment of new media.

[0009] 2. Content generation is disconnected from user feedback;

[0010] AIGC-based content generation tools (such as Jasper and Copy.ai) generate initial marketing content based on user profiles, which is then statically deployed after manual review. Optimization relies on periodic performance reports (such as 24-hour conversion rates).

[0011] Technical flaws: Content generation and real-time user interaction are completely disconnected, manifested in the following ways:

[0012] Visual focus failure: When the user's gaze leaves a key area (such as a product display image), the system is unable to dynamically replace elements;

[0013] Lack of copywriting adaptability: Unable to adjust the density of emotional words in real time according to the user's reading speed (e.g., strong attractive copy is not triggered when scrolling quickly).

[0014] Existing AI content engines are based on a one-way generation architecture, lacking real-time behavior perception and a closed loop for content self-optimization. This deficiency is particularly prominent in highly interactive scenarios such as short videos, resulting in a decrease in average dwell time of over 30% (industry benchmark data).

[0015] Therefore, there is an urgent need for AI-based new media intelligent marketing methods to solve the above problems. Summary of the Invention

[0016] To achieve the above objectives, this invention provides an AI-based intelligent marketing method for new media, including:

[0017] Step 1: Cross-platform user behavior fusion modeling:

[0018] We collect users' content interaction behavior, implicit attention behavior, and cross-platform association behavior through the interfaces of various platforms;

[0019] Platform weighting factors are determined based on user cross-platform activity, and a spatiotemporal behavior matrix is ​​constructed using a time decay algorithm.

[0020] Step 2: Construction of a multimodal content gene library:

[0021] Analyze the visual focal areas and textual sentiment of historical content to generate content gene vectors;

[0022] Step 3: Dynamic Content Generation and Real-Time Optimization

[0023] The gene pool is retrieved based on the spatiotemporal behavior matrix to generate initial marketing content;

[0024] Real-time monitoring of user interaction behavior, dynamic replacement of elements in the visual focus area and adjustment of the emotional intensity of the copy;

[0025] Step 4: Closed-loop strategy optimization:

[0026] The system integrates metrics from multiple platforms to calculate a dynamic performance index. When the index declines continuously, a gene mutation mechanism is triggered to generate cross-platform compensation content based on the behavior of failed users.

[0027] Preferably, the specific steps of the time decay algorithm include:

[0028] a: Identify the timestamps of user actions on the same target across different platforms and calculate the duration of the action interval;

[0029] b: Set the baseline value of the attenuation coefficient based on the frequency of users' historical cross-platform behavior. The higher the frequency of behavior, the smaller the baseline value of the attenuation coefficient.

[0030] c: Calculate the association strength using the exponential decay model: for every unit increase in the behavior interval duration, the association strength is multiplied by the decay coefficient;

[0031] d: When the interval between actions exceeds the user's historical maximum interval, initiate action chain break detection:

[0032] Extract the mean and standard deviation of the user's cross-platform behavior intervals over the past 30 days;

[0033] If the current interval exceeds the mean plus three times the standard deviation, it is determined that the behavioral chain is broken, and the association strength is reset to the preset ratio of the initial value;

[0034] e: The correlation strength of each platform is weighted and fused with the platform weight factor, and the result is output to the spatiotemporal behavior matrix.

[0035] Preferably, the identification of the visual focus region includes:

[0036] a: A gaze heatmap of content keyframes is generated using a convolutional neural network, wherein the gaze probability is represented by pixel values;

[0037] b: Perform region clustering on the heatmap and merge adjacent high-probability pixels to form candidate regions;

[0038] c: Calculate the probability density integral of each candidate region and select the region with the largest integral value as the visual focus region;

[0039] d: Record the center coordinates and coverage area of ​​the visual focus area. When the user's gaze deviates from this area in real time:

[0040] Obtain the closest distance between the line of sight and the boundary of the focal area;

[0041] If the distance continues to increase and exceeds the dynamic threshold, it is determined as a line-of-sight event.

[0042] Preferably, the process for adjusting the emotional intensity of the copywriting is as follows:

[0043] a: Capture the user's screen scrolling speed in real time and calculate its deviation rate from the user's historical average speed;

[0044] b: When the offset rate falls below the first threshold, the emotion enhancement mode is activated.

[0045] Identify the sentiment polarity of the currently displayed paragraph;

[0046] Insert sentiment words of the same polarity and higher intensity into subsequent paragraphs; the insertion density is negatively correlated with the offset rate.

[0047] c: When the offset rate is higher than the second threshold, the interactive wake-up mode is activated.

[0048] Insert a dynamic multiple-choice component at the top of the next screen of content, with options strongly related to the marketing theme;

[0049] Generate branch text in real time based on user selections;

[0050] d: The first and second threshold values ​​are dynamically set through a machine learning model.

[0051] Input user historical dwell time distribution and conversion rate data;

[0052] Output the speed threshold boundary that maximizes content engagement.

[0053] Preferably, the gene mutation mechanism includes:

[0054] a: Calculate the correlation coefficients between each dimension of the content gene vector and the dynamic effect index;

[0055] b: Filter weakly associated dimensions whose correlation coefficients are below the adaptive threshold;

[0056] c: Apply directional perturbation to the weakly correlated dimensions:

[0057] If it is a continuous dimension, Gaussian random perturbation is applied within the range of values;

[0058] If it is a discrete dimension, select an alternative template with a similarity greater than a preset value from the mutation library;

[0059] d: After generating the variant content, conduct A / B testing on a small user base and select the variant that improves the performance index for full deployment.

[0060] Preferably, the specific steps for generating cross-platform compensation content are as follows:

[0061] a: Extract a subset of the spatiotemporal behavior matrix of users who failed in marketing on platform P, and label the reasons for failure;

[0062] b: Input a subset of the behavior matrix into the content generation model of platform Q, and perform differential deconstruction:

[0063] Compare the differences in users' historical behavioral patterns on the P and Q platforms;

[0064] Identify unmet potential interest dimensions of users in platform Q;

[0065] c: Generate compensatory content gene vectors:

[0066] Enhance visual elements that are unique to the Q platform;

[0067] Weaken copywriting elements related to the reasons for failure;

[0068] d: Generate compensating content through an adversarial network and push it to the user the next time they log in to the platform Q.

[0069] Preferably, the process of constructing the dynamic effect index includes:

[0070] a: Define the core metrics set: conversion rate, sharing rate, cross-platform dissemination depth, and user dwell time;

[0071] b: Train the weight allocation model using historical data:

[0072] Input: Product type, user profile category, characteristics of current trending events;

[0073] Output: Initial weighting coefficients for each indicator;

[0074] c: Adjust weights in real time during content delivery:

[0075] For every additional preset number of user interaction samples, the covariance matrix between indicators is recalculated.

[0076] The weighting coefficients are scaled proportionally to the magnitude of the covariance change.

[0077] d: Perform a logarithmic transformation on the normalized weighted summation result to eliminate the influence of the long-tail distribution.

[0078] Preferably, determining the dynamic threshold includes:

[0079] a: Establish a user gaze stability model:

[0080] Extract the fluctuation characteristics of the duration of user gaze in their browsing history;

[0081] The standard deviation of the line-of-sight trajectory is calculated as the baseline fluctuation.

[0082] b: Real-time monitoring of gaze movement acceleration; if the acceleration exceeds a multiple of the baseline fluctuation, threshold adaptation is triggered.

[0083] The larger the multiplier, the smaller the threshold, in order to respond quickly to anomalies;

[0084] The smaller the multiplier, the larger the threshold should be to avoid false triggering.

[0085] Preferably, the logic for generating the branch text is as follows:

[0086] a: Construct a decision tree-structured content template library, where each branch node corresponds to a user selection path, i.e., a branch path;

[0087] b: After the user selects an option:

[0088] Match a pre-defined template based on the semantics of the options;

[0089] If a match fails, the GPT model is invoked to generate coherent copy based on the option content in real time.

[0090] c: Embed tracking tags in the branch copy, and subsequent conversion behaviors will be attributed to that branch path.

[0091] Preferably, the platform preference identification in the differential deconstruction includes:

[0092] a: Extract behavioral feature dimensions unique to platform Q;

[0093] b: Calculate the deviation of the user from the platform average level in this dimension;

[0094] c: If the deviation exceeds the significance threshold, add this dimension to the compensation content enhancement list.

[0095] The beneficial effects of this invention are:

[0096] 1. This invention collects user behavior data from various platforms, determines platform weight factors based on platform activity, and uses a time decay algorithm to establish a spatiotemporal behavior matrix, thereby achieving temporal correlation of cross-platform behaviors.

[0097] 2. This invention generates a content gene vector by analyzing the visual focus areas and textual sentiment of historical content, and then generates initial marketing content based on a spatiotemporal behavior matrix. After real-time monitoring of user behavior, the system can dynamically adjust the content, replace elements in the visual focus areas, and automatically adjust the emotional intensity of the copy based on the user's scrolling speed.

[0098] 3. In this invention, when the dynamic effectiveness index of content marketing continuously declines, the system automatically triggers a gene mutation mechanism based on the changes in the effectiveness index to optimize the content through perturbation. By calculating the correlation between each dimension of the content gene vector and the effectiveness index, weakly correlated dimensions are selected and targeted perturbations are applied, thereby achieving fine-grained optimization of the content. After A / B testing verification, the mutation scheme with the greatest improvement in the effectiveness index is selected and fully deployed. This mechanism ensures that the system can automatically adjust the content when the content effectiveness declines, maintaining continuous optimization of marketing effectiveness.

[0099] 4. In this invention, after a user's marketing attempt fails on a certain platform, the system can extract a subset of the behavior matrix based on the user's cross-platform behavior data, perform differential decomposition, identify the user's potential interest dimensions on different platforms, and generate compensatory content gene vectors. Through generative adversarial networks, the system can prioritize pushing these compensatory contents when the user logs in again, thereby improving the accuracy of cross-platform marketing, avoiding the waste of marketing resources, and increasing user engagement and conversion rates. Attached Figure Description

[0100] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0101] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0102] Figure 2 This is a flowchart illustrating the steps involved in constructing the dynamic effect index according to the method of the present invention.

[0103] Figure 3 This is a flowchart of the platform preference identification steps in the differential deconstruction method of the present invention. Detailed Implementation

[0104] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0105] Please see Figures 1-3 This invention provides an AI-based intelligent marketing method for new media. In step 1, in traditional digital marketing, data barriers between different platforms lead to fragmented user behavior and a one-sided interest model. This invention collects user content interaction behavior (such as likes, comments, and shares), implicit attention behavior (such as browsing history and search records), and cross-platform related behavior (e.g., following a product on one platform and then searching for related information on another) through various platform interfaces (such as Douyin, WeChat, and Xiaohongshu). Next, the system determines platform weight factors based on user activity on each platform and constructs a spatiotemporal behavior matrix using a time decay algorithm. The application of the time decay algorithm adjusts the importance of user behavior based on temporal differences, giving higher weight to behaviors more recent to the current time, thus more accurately reflecting the user's latest interests and needs. Through this method, cross-platform data is effectively integrated, forming a more accurate user profile and solving the problems of data fragmentation and resource mismatch.

[0106] In step 2, content generation requires not only understanding the sentiment of the text but also considering the impact of visual elements on user behavior. In this step, the system analyzes the visual focal areas of historical content (such as product images, brand logos, CTA button locations, etc.) and their corresponding textual sentiments (such as positive, negative, or inquiring), then generates a content gene vector. This gene vector integrates the visual and emotional elements of the content. Through analysis of historical data, the system can identify which visual focal points and textual sentiments effectively enhance user engagement and store them in the content gene library. In this way, the system can optimize visual and emotional elements in subsequently generated marketing content, thereby improving user interaction experience and conversion rates.

[0107] In step 3, when generating initial marketing content, the system retrieves suitable content elements from the content gene library based on the spatiotemporal behavior matrix from step 1. Next, the system monitors user interaction behavior in real time, such as dwell time, scrolling speed, and click locations. When a user's gaze leaves a key area, the system automatically replaces the visual elements in that area (e.g., changing the display image or adjusting the product angle) to maintain user attention. Simultaneously, the system adjusts the emotional intensity of the copy based on user reading behavior. For example, when a user scrolls quickly, the system adjusts the emotional tone of the copy to make it more concise and powerful, enhancing its appeal. This mechanism ensures that content can be optimized in real time based on actual user behavior, improving user engagement and content attractiveness.

[0108] Step 4 integrates marketing performance metrics from multiple platforms, such as user dwell time, interaction rate, and conversion rate, to monitor the marketing effectiveness of the content by calculating a dynamic performance index. When this index declines continuously, the system triggers a gene mutation mechanism. This mechanism analyzes user behavior on other platforms to generate cross-platform compensatory content. This compensatory content provides content different from previous marketing campaigns based on user interests and behavioral characteristics across different platforms, compensating for failed marketing strategies. Through this closed-loop strategy optimization, the system can automatically adjust content, ensuring that marketing campaigns remain effective as user behavior changes and improving conversion rates.

[0109] Through innovative technologies such as cross-platform data fusion, dynamic content generation and optimization, and closed-loop strategy optimization, the accuracy and effectiveness of new media intelligent marketing have been effectively improved, demonstrating significant commercial value.

[0110] In one possible implementation, the system first obtains the timestamps of user interactions with the same target (such as a product, advertisement, or brand) on different platforms via a cross-platform interface. Each time a user performs a target-related action (such as clicking, browsing, or liking) on ​​different platforms, a timestamp is recorded. Next, the system calculates the time interval between different actions to obtain the behavior interval duration. The behavior interval duration serves as a crucial basis for subsequent association strength calculations, reflecting the user's level of attention and sustained interest in the target.

[0111] The system sets a baseline value for the decay coefficient based on the frequency of users' historical behaviors. Generally, frequent cross-platform behavior indicates a higher interest in a particular target; therefore, a smaller baseline decay coefficient means the user's interest lasts longer and the association weakens more slowly. Conversely, lower behavior frequency indicates more intermittent user attention to the target; a larger baseline decay coefficient means the association weakens rapidly. This setting can be flexibly adjusted based on each user's personalized behavioral history to improve the model's accuracy.

[0112] Step c calculates the association strength of the behavior using an exponential decay model. This model dynamically adjusts the weights of user behaviors based on the relationship between the time decay coefficient and the interval between behaviors. As the interval between behaviors increases, the association strength decreases exponentially; that is, the longer the time elapsed since the behavior occurred, the lower the relevance of the user's interest. This model can more accurately reflect the declining trend of user attention over time, thus better predicting the probability of user response in real-time marketing.

[0113] In step d, when a user's behavior interval exceeds its historical maximum interval, the system activates a behavior chain break detection mechanism. By extracting the mean and standard deviation of the user's cross-platform behavior intervals over the past 30 days, the system can assess whether the current behavior interval is abnormal. When the current interval exceeds the mean plus three times the standard deviation, the system determines it as a "behavior chain break," at which point the user's association strength is reset to a preset percentage of the initial value (e.g., 20%). This mechanism can detect situations where a user's interests suddenly shift or behavior is interrupted, preventing incorrect marketing decisions based on outdated behavior.

[0114] In step e, the system combines the behavioral correlation strength of each platform with the platform's weighting factor for weighted fusion. The platform weighting factor reflects the importance of each platform in the user's overall interests. For example, a platform with higher activity or greater influence on users is assigned a higher weight. After fusing these weighted correlation strengths, a spatiotemporal behavioral matrix is ​​formed, which is used for subsequent marketing content optimization and personalized recommendations.

[0115] By meticulously analyzing the changing trends of user behavior across platforms and combining technologies such as exponential decay and behavior chain break detection, a more accurate and dynamic user behavior analysis method is provided, offering strong support for intelligent marketing.

[0116] In one possible implementation, a convolutional neural network (CNN) is first used to analyze the video or image content to generate a gaze heatmap. The heatmap uses different pixel values ​​to represent the probability of a user's gaze, where highlighted areas represent regions the user is likely to gaze at, while under-bright areas indicate areas with a lower probability of gaze. Through automatic learning and feature extraction, the CNN can accurately identify areas of visual attention within the content, especially in complex content or long videos, effectively identifying potential points of interest for the user.

[0117] Next, the system performs region clustering on the generated heatmap, merging adjacent high-probability pixels to form multiple candidate regions. This process is accomplished using clustering algorithms (such as K-means or DBSCAN), with the goal of effectively segmenting the areas the user is likely to focus on, avoiding interference from single pixels. The merged candidate regions help to further identify areas that may be the visual focus, providing clear region definitions for subsequent fine-grained calculations.

[0118] At this point, the system calculates the probability density integral for each candidate region. The magnitude of the integral represents the total probability of gaze in that region. Generally, regions with higher probability density indicate greater user interest; therefore, the system selects the region with the largest integral value as the visual focus area. This selection process ensures that the system captures the most important gaze areas, providing the most accurate user attention information for intelligent marketing.

[0119] The system records the center coordinates and coverage area of ​​the selected visual focus area and tracks the user's gaze trajectory in real time. By continuously monitoring the relationship between the gaze and the focus area, the system can identify whether the user has deviated from the focus area. When the closest distance between the gaze trajectory and the boundary of the focus area continues to increase and exceeds a preset dynamic threshold, the system determines it as a "gaze departure event." At this point, the marketing system can immediately adjust its strategy, such as reducing ad displays in that area or pushing content that better matches the user's interests.

[0120] Visual focus area identification, through precise technical means, can not only effectively capture users' interest hotspots, but also adjust marketing strategies based on users' dynamic behavior, improve the effectiveness of new media marketing, and thus optimize user experience.

[0121] In one possible implementation, the system first captures the user's scrolling speed in real time during reading. This process relies on the user's touch behavior or mouse scrolling data. By sensing the user's swiping or scrolling speed, the system can calculate the deviation rate from the user's historical average speed based on historical data. A lower deviation rate usually indicates that the user's reading pace is slow or that they are lingering on a particular paragraph, while a higher deviation rate may indicate that the user is quickly browsing or is interested in a certain section of content. This information provides the basis for subsequent sentiment adjustment.

[0122] When the calculated offset rate falls below a preset first threshold, the system activates an emotion enhancement mode. In this mode, the system first identifies the emotional polarity (e.g., positive or negative) of the currently displayed paragraph. Next, the system inserts words with the same emotional polarity but higher emotional intensity into subsequent paragraphs to enhance the user's emotional experience. The insertion density of these emotional words is negatively correlated with the user's offset rate; that is, if the user stays on the page for a longer period, the system will appropriately increase the insertion density of emotional words to maintain the user's interest and emotional engagement. This process, through precise emotion analysis and intelligent adjustment, helps improve the user's emotional response to the content.

[0123] When a user's offset rate exceeds a second threshold, the system activates an interactive wake-up mode, prompting the user to participate. In this mode, a dynamic multiple-choice question component is inserted at the top of the next screen's content. These questions are highly relevant to the current marketing theme, designed to stimulate user thought and encourage interaction. Based on the user's choices, the system generates different branching text in real time, adjusting subsequent content presentation according to user interests and responses. This allows the system to form a more personalized and interactive connection with the user, enhancing user engagement.

[0124] The first and second thresholds are not fixed, but dynamically set through a machine learning model. The system inputs users' historical dwell time distribution and conversion rate data, and uses machine learning algorithms (such as regression analysis and decision trees) to analyze this data to determine the most effective offset thresholds. These thresholds are optimized to maximize content engagement and ensure that emotional reinforcement and interactive wake-up modes are activated at the most appropriate times.

[0125] By using intelligent sensing of user behavior, emotional enhancement, and interactive engagement mechanisms, we can not only improve users' emotional engagement but also achieve more personalized and precise marketing strategies, thereby enhancing the attractiveness of content and marketing effectiveness.

[0126] In one possible implementation, the system first transforms the content into a gene vector, where each dimension represents a different feature of the content, such as the emotional tone of the copy, keywords, image style, and layout. Next, the system calculates the correlation coefficient between each dimension and the performance index based on dynamic performance metrics (such as click-through rate, engagement rate, and conversion rate). The correlation coefficient measures the strength of the relationship between each content feature and marketing effectiveness, helping the system identify which content features have a significant impact on marketing results and which features may not contribute sufficiently.

[0127] Next, the system filters out weakly correlated dimensions whose relevance coefficients are below a set adaptive threshold. This adaptive threshold is dynamically adjusted; based on historical data and real-time feedback, the system continuously optimizes it to ensure accurate selection of weakly correlated dimensions. Through this process, the system can identify features that have a minimal impact on content effectiveness, thus avoiding unnecessary adjustments to these features in subsequent steps and saving computational resources.

[0128] For weakly correlated dimensions, the system applies targeted perturbations to explore potential optimization space. The specific perturbation methods are as follows:

[0129] For numerical features, such as article length and image brightness, the system applies Gaussian random perturbations within the feature's value range. Gaussian random perturbations can simulate varying degrees of variation, exploring the feature's impact on the performance index. The magnitude and range of the perturbation can be dynamically adjusted using historical data to ensure that the perturbed feature covers different possible scenarios.

[0130] For discrete features, such as color selection and title style, the system selects an alternative template from a pre-built mutation library whose similarity to the current feature is greater than a preset value. The mutation library contains a variety of validated content templates that may improve performance. The system selects the most suitable alternative based on feature similarity, thereby ensuring the feasibility and effectiveness of the mutated content.

[0131] Once variant content is generated, the system selects a small user group for A / B testing. In A / B testing, the system compares the variant content with the original content, evaluating their impact on user behavior and dynamic performance metrics. By collecting and analyzing user responses in real time, the system can determine which version of the content performs better in terms of performance metrics. Typically, performance metrics include user behavior data such as click-through rate, time on dwell, and conversion rate. The system uses this data as a basis to select the best-performing variant content.

[0132] Ultimately, based on the A / B test results, the system will select the variant with the greatest improvement in performance and deploy it to the entire system. At this point, the optimized content will be displayed on a larger scale, ensuring maximum marketing effectiveness.

[0133] By dynamically adjusting content features, implementing targeted perturbations, and conducting data-driven optimization, the marketing effectiveness of the content has been effectively improved. It can maintain high flexibility and accuracy in a constantly changing market environment, ultimately increasing user engagement and conversion rates.

[0134] In one possible implementation, the system first extracts the spatiotemporal behavior matrix of users who failed in their marketing efforts from platform P. The spatiotemporal behavior matrix refers to user behavior data on platform P, including data across multiple dimensions such as timestamps, geographic location, browsing duration, and click records. Based on the pre-set goals of the marketing campaign (such as purchases or registrations), the system marks user behaviors that did not achieve the expected goals. These failed behaviors are assigned failure reason tags, such as "failed purchase," "low interaction rate," and "not following." Through these tags, the system can distinguish different failure reasons, providing data support for subsequent analysis and optimization.

[0135] In this step, the system inputs a subset of the spatiotemporal behavior matrix of failed users from platform P into the content generation model of platform Q. In this way, the system compares the differences in users' historical behavioral patterns on platforms P and Q. Platforms P and Q may have different user interfaces, interaction methods, or content presentations; therefore, user behavior patterns will also differ. The system uses differential deconstruction analysis to identify the differences between platforms P and Q, particularly potential interest dimensions that may not be fully satisfied on platform Q. For example, users may not be paying attention to certain types of content or performing certain behaviors on platform Q. Through this analysis, the system can identify and fill in the unmet interests of users on platform Q, helping to improve personalized content recommendations.

[0136] Next, based on the analysis results from the first two steps, the system will generate a compensatory content gene vector. The content gene vector represents various feature dimensions of the content, such as visual elements, copywriting style, and color scheme. When generating compensatory content, the system will:

[0137] Based on the user preferences of Platform Q, enhance the visual elements that users frequently interact with or tend to pay more attention to on Platform Q (such as specific image styles, video types, or interface designs).

[0138] Based on the failure reason tags, the system will avoid using copywriting elements that may lead to user churn (such as overly direct promotional information or overly commercial language), thereby reducing potential negative impacts.

[0139] These adjustments will make the generated compensatory content more aligned with the preferences of platform Q users and, to some extent, compensate for marketing failures on platform P.

[0140] Finally, the system utilizes Generative Adversarial Networks (GANs) to synthesize compensating content. Through adversarial training, GANs can generate more natural and highly personalized content. In this process, the generator produces compensating content, while the discriminator evaluates the authenticity and suitability of this content. After repeated optimization, the generated compensating content is considered the most suitable for the user's needs. The system will prioritize pushing this compensating content to the user the next time they log in to the platform, thereby improving user engagement and conversion rates.

[0141] This AI-based new media intelligent marketing approach utilizes cross-platform behavioral analysis and generative adversarial networks to optimize content, effectively improving personalized recommendations and the accuracy of content matching, thereby enhancing the effectiveness of marketing campaigns and increasing long-term user engagement.

[0142] In one possible implementation, the first step is to define key metrics for measuring marketing effectiveness. These metrics include: conversion rate, sharing rate, cross-platform reach, and user dwell time. The specific meaning of each metric is as follows:

[0143] Conversion rate: refers to the percentage of users who complete a target action (such as purchase or registration), reflecting the impact of marketing activities on user behavior.

[0144] Share rate: refers to the frequency with which users share marketing content with others, reflecting the content's reach and virality.

[0145] Cross-platform dissemination depth: measures the effectiveness of content dissemination across multiple platforms, indicating the scope of content spread.

[0146] User dwell time: refers to the length of time users stay on the platform, reflecting the attractiveness of the content and user engagement.

[0147] These metrics are defined to comprehensively capture different dimensions of marketing campaigns and provide data support for subsequent model training.

[0148] Furthermore, the system will train a weight allocation model using historical data. Specific inputs include:

[0149] Product type: Different types of products may have different appeal to different user groups, so product type needs to be used as an input to affect the model's predictions.

[0150] User profile classification includes basic user information (such as age, gender, region, etc.) and behavioral habits. These features help the model identify user needs and thus determine the weight of each indicator.

[0151] Characteristics of current trending events: such as holidays, social hot topics and other external factors, which often affect user behavior, so they need to be considered in the model.

[0152] The model outputs the initial weight coefficients for each indicator, which represent the contribution of each indicator to the dynamic performance index under specific conditions.

[0153] During actual content delivery, the system monitors user interactions in real time. For every additional preset number of user interaction samples, the covariance matrix between metrics is recalculated. The covariance matrix reflects the interrelationships and influence levels between various metrics. By analyzing changes in covariance, the system can determine the relative importance of each metric.

[0154] Based on changes in covariance, the system will proportionally scale the weight coefficients of each indicator. For example, if the covariance between a certain indicator and other indicators changes significantly, the system will adjust the weight of that indicator to ensure the accuracy and timeliness of the dynamic performance index.

[0155] In the final stage of data processing, the system normalizes the weighted sum to eliminate the influence between data of different magnitudes, resulting in a more balanced contribution of each indicator. Then, the system performs a logarithmic transformation on the normalized weighted result to eliminate the influence of long-tailed distributions. Long-tailed distributions often lead to extreme values ​​having an excessive impact on the overall result; through logarithmic transformation, the system can smooth the influence of these extreme values, thereby obtaining a more stable and reliable dynamic performance index.

[0156] Through multi-dimensional data analysis and real-time feedback mechanisms, we can not only improve the accuracy of marketing effectiveness evaluation, but also adjust strategies in real time according to different scenarios, ensuring the efficiency and adaptability of marketing activities.

[0157] In one possible implementation, the system first needs to establish a user gaze stability model, the purpose of which is to analyze the stability of the user's gaze while browsing content, especially in terms of content attractiveness and interactivity.

[0158] By analyzing users' historical browsing data, the system extracts information on the duration each user spends on a page and calculates the fluctuations in gaze duration. These fluctuations help the system understand users' attention patterns; for example, spending more time on certain content or pausing at specific locations reflects the user's level of interest in the content.

[0159] Based on historical user data, the standard deviation of their gaze trajectory is calculated as a baseline fluctuation. This standard deviation represents the typical fluctuation range of user gaze movement and reflects the user's usual behavioral patterns. A larger standard deviation indicates that the user's gaze movement on the page is more unstable, and vice versa.

[0160] The system will monitor the user's eye movement speed and acceleration in real time through sensors or front-end data capture technology on the user's device. When the user's gaze jumps significantly or moves frequently, the acceleration will increase accordingly, which may indicate that the user has developed a strong interest in the content or is exhibiting abnormal behavior (such as suddenly shifting attention).

[0161] When the acceleration of the line of sight exceeds a certain multiple of the baseline fluctuation, the system triggers a dynamic threshold adaptation mechanism. This mechanism dynamically adjusts the threshold based on the magnitude of the acceleration.

[0162] When the acceleration is significantly greater than the baseline fluctuation, it indicates an abnormal change in the user's gaze, which may suggest that the user's attention has shifted elsewhere or that there is a system problem. Therefore, the threshold is lowered to allow the system to respond more quickly to such anomalies and make timely adjustments.

[0163] When the acceleration change is small, meaning the line of sight changes relatively smoothly, it indicates that the user is still browsing normally. The system then increases the threshold to avoid overly sensitive reactions and reduce the risk of accidental triggering.

[0164] By introducing a line-of-sight stability model and a dynamic threshold adaptive mechanism, the system can respond to changes in user behavior in real time, optimize marketing strategies, and significantly improve the accuracy and user experience of the system, thus possessing high practical value.

[0165] In one possible implementation, the system first needs to build a decision tree-based content template library. A decision tree is a model that generates different content based on user-selected paths. Each branch node represents a user's selection path, and different user choices lead to different path developments. This content template library contains a variety of preset text templates, and the system determines the path and selects the appropriate text template based on user interactions (such as clicks and selections). This approach ensures the personalization and precision of marketing content, enabling the generation of customized marketing information in real time based on user needs and behaviors.

[0166] After a user makes a selection within the marketing content, the system needs to match a suitable pre-set template based on the semantics of that option. The system performs semantic matching based on the user's selected keywords or category, finding the template that best matches that selection path from the template library for display. For example, if a user selects "summer clothing" as an option, the system will select a summer clothing promotional copy template related to that selection from the decision tree.

[0167] If the system cannot find a perfectly matching template, it will invoke the GPT model to generate a coherent piece of copy in real time based on the user's selections. The GPT model can generate copy that matches the user's interests through natural language processing technology, based on the user's selected context and intent, thereby filling the gaps in the template library and ensuring the coherence and appeal of the copy content.

[0168] Within each generated branch copy, the system embeds a tracking tag. This tracking tag is a unique identifier used to track and record the user's behavioral path. When the user performs subsequent actions or conversions (such as purchases or registrations), the system can associate these actions with that branch path, clearly indicating that these conversions were driven by a specific copy branch. In this way, the system can accurately evaluate the effectiveness of different branch copy, thereby optimizing subsequent marketing strategies.

[0169] By employing technologies such as intelligent decision trees, semantic matching, GPT-generated copy, and tracking tags, highly personalized and automated content generation can be achieved, which not only improves marketing efficiency but also provides precise data support for subsequent performance analysis and strategy optimization.

[0170] In one possible implementation, user behavior typically differs across different social or media platforms (such as Platform Q). Platform Q has its own unique user behavior patterns, such as browsing frequency, interaction methods, and click habits. Extracting platform-specific behavioral feature dimensions refers to identifying specific behavioral dimensions on Platform Q through in-depth analysis of user behavior. For example, a platform might have users who frequently exhibit preferences for certain content types (such as videos, text / images, and comments). The system will extract these behavioral features to form dimensions available for subsequent analysis.

[0171] Next, the system needs to evaluate each user's performance across these behavioral dimensions and compare it to the platform's average level. Specifically, the system calculates the difference between a user's performance on that specific dimension and the average level of all users on platform Q—that is, the deviation. Deviation can be measured using statistical methods, such as standard deviation and mean difference, to measure the degree to which a user's behavior differs from that of other users on the platform. For example, if the average daily video viewing time for users on platform Q is 30 minutes, and a certain user's viewing time is 60 minutes, then that user's deviation in the video viewing time dimension is 30 minutes, indicating that the user deviates from the platform's average behavioral pattern.

[0172] When a user's deviation in a certain dimension exceeds the system's preset significance threshold, it means that the user's behavior differs significantly from that of a typical user on Platform Q. Based on this difference, the system will add that specific dimension to the compensation content enhancement list. The compensation content enhancement list is an adjustment mechanism for user behavior differences. In subsequent content pushes, the system will enhance or adjust content related to that dimension to compensate for or optimize the user's experience in that dimension. For example, if a user deviates significantly in video viewing time, the system may push more short video content to that user or recommend videos they are more interested in, to bring them closer to the average user behavior pattern on Platform Q, thereby increasing user engagement and platform activity.

[0173] Through precise platform preference identification and content enhancement mechanisms, not only are personalized and efficient content recommendations achieved, but also crucial data is provided for subsequent marketing decisions. These technological features significantly improve the accuracy and effectiveness of marketing campaigns.

[0174] The following examples will illustrate this in detail:

[0175] This invention is applied to an e-commerce platform whose user behavior analysis system can provide personalized recommendations based on users' historical behavior data, browsing history, and purchasing preferences, thereby boosting sales and improving user experience. The system uses multiple algorithms combined with user deviation analysis to optimize recommended content.

[0176] Specifically, the platform records user behavior data in real time, including user actions such as browsing products, clicking, searching, and purchasing. Each action is recorded using data fields such as timestamp, product ID, and action type (browsing, clicking, purchasing, etc.).

[0177] Example of collected data:

[0178] User A: Browses product A (Time: 10:05), clicks on product B (Time: 10:15), purchases product C (Time: 10:30)

[0179] User B: Browses product D (Time: 10:10), clicks on product E (Time: 10:20), purchases product F (Time: 10:40)

[0180] User interest vectors are generated using behavioral data. For each user, their interest values ​​across different product categories are calculated as a result of feature extraction.

[0181] In this embodiment, the e-commerce platform has 5 product categories (Category A, Category B, Category C, Category D, and Category E). Each user's interest vector is calculated by weighting the user's behavior weights across each category.

[0182] For user A, in this embodiment, there were 2 views in category A, 1 click in category B, 1 purchase in category C, and no activity in categories D and E. The interest vector can be calculated as follows:

[0183] I A =(0.4,0.2,0.2,0.1,0.1)

[0184] Each element represents the interest level of that category (i.e., behavior frequency / total number of behaviors).

[0185] This invention employs deviation analysis to assess the difference between user behavior and the overall user behavior of the platform. Deviation measures the difference between a user's interests and the platform's average user interests, and the formula is:

[0186] ;

[0187] in, It represents user A's interest in category i. It represents the average interest level of all users on the platform in category i.

[0188] For example, if the average interest level of all users on the platform in category A is 0.25, then for user A, the deviation from category A is:

[0189] ;

[0190] The greater the deviation, the more the user deviates from the behavior of most users on the platform, and the more the system needs to recommend content that matches their unique interests.

[0191] The system adjusts its recommendation strategy based on the degree of deviation. Users with a large deviation will receive more personalized product recommendations, while users with a small deviation will be recommended products that are generally popular on the platform.

[0192] The product recommendation formula is: ;

[0193] in, This is user A's final recommendation list. It is the deviation of category i. It is the weight of category i (set according to the importance of the product category). It is a category Rate the product.

[0194] Based on the final calculated recommendation list, the platform displays personalized product recommendations to users. The recommendations may include multiple product categories and are sorted based on the user's historical purchases and browsing preferences.

[0195] Behavioral data is collected once per second to ensure real-time updates of user behavior.

[0196] When calculating the deviation, a threshold (such as 0.3) is set. When a user's deviation exceeds this threshold, the system will prioritize pushing highly personalized product recommendations.

[0197] Each user's interest vector is normalized between 0 and 1 to avoid bias caused by class imbalance.

[0198] Weighting Adjustment: The weight of a category is dynamically adjusted based on the sales performance of the product. For example, best-selling products have a higher weight.

[0199] To verify the superiority of the recommendation system of this invention, a comparative experiment was conducted with a traditional recommendation system based on product click-through rate. The experimental group used the deviation algorithm of this invention, while the control group used a traditional recommendation method based on user behavior frequency.

[0200] Experimental Design:

[0201] Experiment subjects: 1000 e-commerce platform users

[0202] Experiment duration: 1 month

[0203] Evaluation metrics: User engagement (click-through rate), purchase conversion rate, recommendation accuracy

[0204] Experimental results:

[0205] Experimental group (deviation recommendation): User engagement increased by 20%, and purchase conversion rate increased by 15%.

[0206] Control group (traditional recommendation): User engagement increased by 5%, and purchase conversion rate increased by 8%.

[0207] Experimental results demonstrate that the personalized recommendation system based on deviation analysis of this invention can significantly improve user engagement and purchase conversion rate compared to traditional methods, and is more in line with users' personalized needs.

[0208] By analyzing deviations in user behavior, this invention achieves more accurate personalized recommendations, effectively improving user experience and conversion rates on e-commerce platforms. The system can provide the most suitable product recommendations based on each user's unique interests, solving the problem that traditional recommendation systems cannot accurately meet user needs. The recommendation algorithm and system design of this invention are highly innovative and practical, significantly improving the commercial benefits of e-commerce platforms.

[0209] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0210] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An AI-based intelligent marketing method for new media, characterized by: include: Step 1: Cross-platform user behavior fusion modeling: We collect users' content interaction behavior, implicit attention behavior, and cross-platform association behavior through the interfaces of various platforms; Platform weighting factors are determined based on user cross-platform activity, and a spatiotemporal behavior matrix is ​​constructed using a time decay algorithm. Step 2: Construction of a multimodal content gene library: Analyze the visual focal areas and textual sentiment of historical content to generate content gene vectors; Step 3: Dynamic Content Generation and Real-Time Optimization The gene pool is retrieved based on the spatiotemporal behavior matrix to generate initial marketing content; Real-time monitoring of user interaction behavior, dynamic replacement of elements in the visual focus area and adjustment of the emotional intensity of the copy; Step 4: Closed-loop strategy optimization: The dynamic performance index is calculated by integrating indicators from multiple platforms. When the index declines continuously, a gene mutation mechanism is triggered to generate cross-platform compensation content based on the behavior of failed users. The gene mutation mechanisms include: a: Calculate the correlation coefficients between each dimension of the content gene vector and the dynamic effect index; b: Filter weakly associated dimensions whose correlation coefficients are below the adaptive threshold; c: Apply directional perturbation to the weakly correlated dimensions: If it is a continuous dimension, Gaussian random perturbation is applied within the range of values; If it is a discrete dimension, select an alternative template with a similarity greater than a preset value from the mutation library; d: After generating the variant content, conduct A / B testing on a small user base and select the variant that improves the performance index for full deployment; The specific steps of the time decay algorithm include: a: Identify the timestamps of user actions on the same target across different platforms and calculate the duration of the action interval; b: Set the baseline value of the attenuation coefficient based on the frequency of users' historical cross-platform behavior. The higher the frequency of behavior, the smaller the baseline value of the attenuation coefficient. c: Calculate the association strength using the exponential decay model: for every unit increase in the behavior interval duration, the association strength is multiplied by the decay coefficient; d: When the interval between actions exceeds the user's historical maximum interval, initiate action chain break detection: Extract the mean and standard deviation of the user's cross-platform behavior intervals over the past 30 days; If the current interval exceeds the mean plus three times the standard deviation, it is determined that the behavioral chain is broken, and the association strength is reset to the preset ratio of the initial value; e: Weight and fuse the correlation strength of each platform with the platform weight factor, and output the spatiotemporal behavior matrix; The process of constructing the dynamic effect index includes: a: Define the core metrics set: conversion rate, sharing rate, cross-platform dissemination depth, and user dwell time; b: Train the weight allocation model using historical data: Input: Product type, user profile category, characteristics of current trending events; Output: Initial weighting coefficients for each indicator; c: Adjust weights in real time during content delivery: For every additional preset number of user interaction samples, the covariance matrix between indicators is recalculated. The weighting coefficients are scaled proportionally to the magnitude of the covariance change. d: Perform a logarithmic transformation on the normalized weighted summation result to eliminate the influence of the long-tail distribution.

2. The AI-based intelligent marketing method for new media according to claim 1, characterized in that: The identification of the visual focus area includes: a: A gaze heatmap of content keyframes is generated using a convolutional neural network, wherein the gaze probability is represented by pixel values; b: Perform region clustering on the heatmap and merge adjacent high-probability pixels to form candidate regions; c: Calculate the probability density integral of each candidate region and select the region with the largest integral value as the visual focus region; d: Record the center coordinates and coverage area of ​​the visual focus area. When the user's gaze deviates from this area in real time: Obtain the closest distance between the line of sight and the boundary of the focal area; If the distance continues to increase and exceeds the dynamic threshold, it is determined as a line-of-sight event.

3. The AI-based intelligent marketing method for new media according to claim 1, characterized in that: The process of adjusting the emotional intensity of the copywriting is as follows: a: Capture the user's screen scrolling speed in real time and calculate its deviation rate from the user's historical average speed; b: When the offset rate falls below the first threshold, the emotion enhancement mode is activated. Identify the sentiment polarity of the currently displayed paragraph; Insert sentiment words of the same polarity and higher intensity into subsequent paragraphs; the insertion density is negatively correlated with the offset rate. c: When the offset rate is higher than the second threshold, the interactive wake-up mode is activated. Insert a dynamic multiple-choice component at the top of the next screen of content, with options strongly related to the marketing theme; Generate branch text in real time based on user selections; d: The first and second threshold values ​​are dynamically set through a machine learning model. Input user historical dwell time distribution and conversion rate data; The output is the speed threshold boundary that maximizes content engagement. The logic for generating the branch text is as follows: a: Construct a decision tree-structured content template library, where each branch node corresponds to a user selection path, i.e., a branch path; b: After the user selects an option: Match a pre-defined template based on the semantics of the options; If a match fails, the GPT model is invoked to generate coherent copy based on the option content in real time. c: Embed tracking tags in the branch copy, and subsequent conversion behaviors will be attributed to that branch path.

4. The AI-based intelligent marketing method for new media according to claim 1, characterized in that: The specific steps for generating cross-platform compensation content are as follows: a: Extract a subset of the spatiotemporal behavior matrix of users who failed in marketing on platform P, and label the reasons for failure; b: Input a subset of the behavior matrix into the content generation model of platform Q, and perform differential deconstruction: Compare the differences in users' historical behavioral patterns on the P and Q platforms; Identify unmet potential interest dimensions of users in platform Q; c: Generate compensatory content gene vectors: Enhance visual elements that are unique to the Q platform; Weaken copywriting elements related to the reasons for failure; d: Generate compensating content through an adversarial network and push it to the user the next time they log in to the platform Q; The platform preference identification in the differential deconstruction includes: a: Extract behavioral feature dimensions unique to platform Q; b: Calculate the deviation of the user from the platform average level in this dimension; c: If the deviation exceeds the significance threshold, add this dimension to the compensation content enhancement list.

5. The AI-based intelligent marketing method for new media according to claim 2, characterized in that: The determination of the dynamic threshold includes: a: Establish a user gaze stability model: Extract the fluctuation characteristics of the duration of user gaze in their browsing history; The standard deviation of the line-of-sight trajectory is calculated as the baseline fluctuation. b: Real-time monitoring of gaze movement acceleration; if the acceleration exceeds a multiple of the baseline fluctuation, threshold adaptation is triggered. The larger the multiplier, the smaller the threshold, in order to respond quickly to anomalies; The smaller the multiplier, the larger the threshold should be to avoid false triggering.

Images