Dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data

Abstract

The invention discloses a dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data. The dynamic intelligent multi-module real-time content recommendation system comprises a basic tool module, a creation tool module, a content stream module, a creation agent module, a multi-channel data acquisition and analysis system and a cross-module dynamic linkage recommendation module. The method can accurately predict user demands and actively position related contents, greatly reduces user browsing and searching time, improves first screen content matching degree, improves content acquisition efficiency, can trigger and dynamically adjust content display strategies and function arrangement sequences of functional modules according to user behaviors, breaks barriers among the modules, and improves user experience. According to the method, intelligent linkage among multiple modules is achieved, more personalized creation experience is provided for a user, the core problems of content homogenization, module isolation and poor user experience of a current AI content creation platform can be effectively solved, and a dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data is provided for the user.

Description

Technical Field

[0001] The present invention relates to the technical field of content recommendation systems, and in particular to a dynamic, intelligent, multi-module, real-time content recommendation system based on multi-channel behavior data. Background Art

[0002] In the current digital content creation and consumption environment, users face the dual challenges of information overload and inefficient acquisition. Traditional content recommendation systems usually use fixed algorithm models and lack the ability to respond to users' comprehensive behaviors and preferences in real time, resulting in low recommendation accuracy and fragmented user experience. Although the AI-generated content platforms on the existing market have powerful content generation capabilities, they still have obvious deficiencies in content display and user interaction. In general, there are the following problems: 1. Homogeneity of content flow design styles: Mainstream AI-generated platforms basically adopt dual-stream or multi-stream content display formats, which lack innovation and differentiation. The user experience tends to be similar and it is difficult to meet the personalized needs of different users; 2. Isolated operation of functional modules: The content information flow and other functional modules of the product lack organic connection and operate independently. They cannot form a synergistic effect, which reduces the overall user efficiency; 3. Low content acquisition efficiency: Traditional recommendation systems cannot achieve accurate recommendations on the first or second screen, resulting in user loss in the early stages of browsing. It is difficult to obtain users' continued attention and engagement. Users usually need to turn pages multiple times or search frequently to find the content they need, which greatly reduces the user experience.

[0003] Based on the above, this application proposes a dynamic, intelligent, multi-module real-time content recommendation system based on multi-channel behavioral data. Summary of the Invention

[0004] Based on the technical problems existing in the background technology, the present invention proposes a dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data.

[0005] The dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavioral data proposed in the present invention includes a basic tool module, a creation tool module, a content flow module, a creation intelligent body module, a multi-channel data collection and analysis system and a cross-module dynamic linkage recommendation module. The multi-channel data collection and analysis system includes a multi-channel behavioral data collection module and a behavior analysis and modeling module.

[0006] Preferably, the basic tool module is located at the top of the product, providing a set of basic functions required for creation, and is used to provide basic editing functions of image processing, including picture editing, background erasing, high-definition restoration and lossless magnification functions.

[0007] Preferably, the creation tool module is used to provide multiple types of image or video creation functions according to different application scenarios, which are divided into the following different categories according to the usage scenarios:

[0008] Lifestyle: Contains portraits, couple portraits, pet portraits, social profile pictures, and other creation tools related to daily life;

[0009] Work category: including e-commerce background images, e-commerce materials, logo creation, and tattoo creation professional creation tools;

[0010] Popular categories: Recommend currently popular gameplay and user-preferred content.

[0011] Preferably, the content flow module adopts an up-and-down sliding layout with a quick positioning bar at the bottom, allowing users to quickly locate the content area of ​​interest. It is used to display recommended content and dynamically adjust the sorting and display strategy according to user behavior.

[0012] Preferably, the creative agent module exists in the form of a dialogue interface, and users can express their creative needs through dialogue. Based on the dialogue form, the agent receives user instructions, and can recommend suitable scenes, or directly complete the image / video generation task in the dialog box. The agent can actively provide creative suggestions, tool recommendations and operation guidance.

[0013] Preferably, the multi-channel behavior data collection module is used to collect user behavior data in the basic tool module, the creation tool module, the content flow module and the creation agent module, which specifically includes the user's function usage frequency and sequence, creation tool preferences, content browsing track, dwell time, operation preferences, agent interaction records and instruction preferences, cross-module operation paths and time sequence relationships, and user preferred content style characteristics;

[0014] The behavior analysis and modeling module constructs user behavior profiles and preference models through multi-dimensional data fusion analysis and updates model parameters in real time to ensure that the recommendation system can dynamically respond to changes in user needs. The specific logical steps are as follows:

[0015] S101: Extract structured behavior features from the original behavior data collected by the multi-channel behavior data collection module to form a multi-dimensional vector , which includes Recent activity of basic tools, Distribution of creative preference types, Average duration of content flow Distribution of user active time periods, Recent agent dialogue keyword frequency, Click the content label frequency of the behavior, normalize the behavior feature vector and perform time weighting processing to obtain = ;

[0016] S102: Use a K-means clustering algorithm to generate a behavioral profile label for each user's behavioral feature vector. This label represents the content preference group to which the user belongs. This behavioral profile label serves as one of the input priors for the subsequent recommendation model. The behavioral profile labels include e-commerce-oriented users, creative experimental users, tool efficiency users, and beginner guidance users.

[0017] S103: Construct a user-content scoring function and train a recommendation scoring model. The formula is: ,in is the weight of user feature dimension i, Score the relevance of the user to the content in the i-th dimension (e.g., the degree of match between clicks, likes, favorites, and stay behaviors and content tags). is the matching score of user U to content c;

[0018] S104: Dynamic parameter update and adaptive learning: After each new user behavior occurs, the following logic is executed:

[0019] (1) Update the user behavior vector using the following formula: , where γ∈[0,1] is the historical weight coefficient;

[0020] (2) Adjusting the behavior weight distribution , increase the weight of recent frequent behaviors in recommendations;

[0021] (3) Adjust the recommendation ranking in real time and use behavioral feedback (clicks, bounces, etc.) as negative sampling / positive sampling signals to optimize the recommendation model;

[0022] (4) When user behavior significantly deviates from the group center, its profile category is updated, triggering the cross-module dynamic linkage recommendation module to execute the action;

[0023] S105: In any module, if the user behavior vector deviates significantly, , it triggers the cross-module dynamic linkage recommendation module to execute the action.

[0024] Preferably, the cross-module dynamic linkage recommendation module triggers dynamic adjustment of content and function recommendations based on the user's current behavior, thereby achieving real-time linkage between various function modules. The specific logical steps are as follows:

[0025] S201: The events collected by the multi-channel behavior data collection module are encapsulated into behavior event tuples: ,in The module where the current behavior occurs. is the operation behavior type, Behavioral parameters include time, content tags, and interaction words;

[0026] S202: Through the event processing engine Real-time semantic analysis is performed to identify the potential intention of the user's current operation. For example, if a user frequently uses "background erase" and "high-resolution restoration" → their goal is to infer e-commerce image optimization; if a user continuously browses "social avatar" images in the content stream → their interest is inferred to be inclined towards social scene creation. The user portrait model and historical behavior path are combined to calculate the behavior offset: , , it is regarded as a "behavior pattern mutation" and triggers cross-module dynamic linkage;

[0027] S203: Based on the behavioral intention and module dependency, determine the target module set to be linked: , and sort by priority;

[0028] S204: For each linked module , perform recommendation strategy update, and the formula used in the dynamic adjustment process is: ,in is the trigger response strength parameter, It is a trigger indicator function and takes 1 when the behavior deviation is significant;

[0029] S205: Once the linkage recommendation strategy takes effect, the UI interface immediately rearranges the module display order, and the linked modules quickly refresh the content or function display. If there is no user interaction delay, the current display rhythm is maintained;

[0030] S206: The system records the response effect of each linkage recommendation, including whether the user clicks on the recommended content, whether the length of stay after linkage increases, and whether the next action is triggered. The system updates the user model parameters based on these feedback data. , content matching rules , and linkage recommendation strategies , forming a self-learning closed loop of linkage recommendation.

[0031] Preferably, in S204, the content of executing the recommendation strategy update is as follows:

[0032] (1) Content flow module: Adaptive changes in content flow layout: record user long-term preferences, automatically locate the user's frequently visited preferred section each time logging in, and dynamically adjust the default sorting method, content density and display format of the content flow according to user usage habits; Personalized adjustment of content flow image layout: automatically adjust the image layout in the content flow according to the user's preferred style, and the content template of the preferred style is enhanced in visual presentation, and support dynamic adjustment of image display ratio according to user browsing habits;

[0033] (2) Adaptive layout of functional modules: The layout of the basic tool module and the creative tool module is dynamically adjusted according to the frequency of user use. High-frequency functions are automatically displayed in the front, and low-frequency function modules are automatically collapsed, saving interface space and improving user experience. At the same time, it supports customizable sorting of functional modules, and the system provides default sorting suggestions based on user habits;

[0034] (3) Dynamic adaptation of the personality of the intelligent agent module: The system analyzes the user's interaction style and response preferences, and intelligently adjusts the communication style and personality characteristics of the AI ​​Agent. For efficiency-oriented users, the intelligent agent module adopts a simple and efficient communication method, reduces redundant hierarchical guidance, and directly provides solutions. For creativity-oriented users, the intelligent agent module shows more imaginative and humorous personality characteristics. For learning-oriented users, the intelligent agent actively provides more tutorials and guidance information.

[0035] (4) Multi-terminal synchronization and scene continuation: User preferences and settings can be synchronized across devices to ensure a consistent personalized experience on different terminal devices. It supports cross-device continuation of creative tasks. The system automatically records the creative progress and environmental parameters. It also makes intelligent recommendations based on time and scene, giving priority to professional content during work hours and favoring entertainment creation during leisure time.

[0036] (5) Evolution of interaction mode: Analyze user operation habits, automatically adjust the response priority of the interaction method, dynamically adjust the frequency and detail of operation prompts according to user proficiency, and support customized interaction processes. The system can learn the user's unique operation habits and optimize the response method.

[0037] Compared with the existing technology, the beneficial effects of the present invention are:

[0038] 1. Accurately predict user needs and proactively locate relevant content, significantly reducing user browsing and search time, improving the matching of home screen content, increasing content acquisition efficiency, and reducing user churn;

[0039] 2. Through the setting of cross-module dynamic linkage recommendation modules, the content display strategy and function arrangement order of each functional module can be dynamically adjusted according to user behavior triggers, breaking down the barriers between modules and realizing intelligent linkage between modules, so that all parts of the product form an organic whole, breaking through the homogeneity limitations of traditional AI content platforms. Through innovative multi-module linkage mechanisms, unique product differentiation advantages are established to provide users with a more personalized creation experience. In addition, by continuously learning user behavior patterns, it can gradually adjust to a form that is more in line with the user's personal habits, providing truly personalized services. Through intelligent guidance and precise recommendations, users' trust and dependence on the product are enhanced, the usage time is extended, and the user retention rate is improved.

[0040] The present invention can accurately predict user needs and actively locate relevant content, greatly reducing the time users spend browsing and searching, improving the matching degree of content on the first screen, and improving content acquisition efficiency. It can also dynamically adjust the content display strategy and function arrangement order of each functional module according to user behavior triggers, break down the barriers between modules, and realize intelligent linkage between multiple modules, providing users with a more personalized creation experience. It can effectively solve the core problems of content homogeneity, module isolation, and poor user experience faced by current AI content creation platforms, and provide users with a dynamic, intelligent, multi-module real-time content recommendation system based on multi-channel behavioral data, realizing a truly intelligent, personalized, and efficient creation and content consumption experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a block diagram of the dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavioral data proposed by the present invention;

[0042] Figure 2 This is a flow chart of the dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavioral data proposed by the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further explained below with reference to specific embodiments.

[0044] Example

[0045] Reference Figure 1-2 This embodiment proposes a dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data, including a basic tool module, an authoring tool module, a content flow module, an authoring agent module, a multi-channel data collection and analysis system, and a cross-module dynamic linkage recommendation module. The multi-channel data collection and analysis system includes a multi-channel behavior data collection module and a behavior analysis and modeling module.

[0046] The Basic Tools module is located at the top of the product and provides a collection of basic functions required for creation. It is used to provide basic editing functions such as image processing, including picture editing, background erasing, high-definition restoration, and lossless enlargement.

[0047] The creation tool module is used to provide various image or video creation functions according to different application scenarios. It is divided into the following categories according to the usage scenarios:

[0048] Lifestyle: Contains portraits, couple portraits, pet portraits, social profile pictures, and other creation tools related to daily life;

[0049] Work category: including e-commerce background images, e-commerce materials, logo creation, and tattoo creation professional creation tools;

[0050] Popular categories: Recommend currently popular gameplay and user-preferred content;

[0051] The content flow module adopts an up-and-down sliding layout with a quick location bar at the bottom, allowing users to quickly locate the content area of ​​interest. It is used to display recommended content and dynamically adjust the sorting and display strategy based on user behavior.

[0052] The creative agent module exists in the form of a dialogue interface. Users can express their creative needs through dialogue. Based on the user's instructions in the form of dialogue, the agent can recommend suitable scenes or complete the image / video generation task directly in the dialog box. The agent can actively provide creative suggestions, tool recommendations and operation guidance.

[0053] The multi-channel behavior data collection module is used to collect user behavior data in the basic tool module, creation tool module, content flow module and creation agent module, which specifically includes the frequency and order of user function usage, creation tool preferences, content browsing trajectory, dwell time, operation preferences, agent interaction records and instruction preferences, cross-module operation paths and time sequence relationships, and user preferred content style characteristics;

[0054] The behavior analysis and modeling module builds user behavior profiles and preference models through multi-dimensional data fusion analysis and updates model parameters in real time to ensure that the recommendation system can dynamically respond to changes in user needs. The specific logical steps are as follows:

[0055] S101: Extract structured behavior features from the original behavior data collected by the multi-channel behavior data collection module to form a multi-dimensional vector , which includes Recent activity of basic tools, Distribution of creative preference types, Average duration of content flow Distribution of user active time periods, Recent agent dialogue keyword frequency, Click the content label frequency of the behavior, normalize the behavior feature vector and perform time weighting processing to obtain = ;

[0056] S102: Use a K-means clustering algorithm to generate a behavioral profile label for each user's behavioral feature vector. This label represents the content preference group to which the user belongs. This behavioral profile label serves as one of the input priors for the subsequent recommendation model. The behavioral profile labels include e-commerce-oriented users, creative experimental users, tool efficiency users, and beginner guidance users.

[0057] S103: Construct a user-content scoring function and train a recommendation scoring model. The formula is: ,in is the weight of user feature dimension i, Score the relevance of the user to the content in the i-th dimension (e.g., the degree of match between clicks, likes, favorites, and stay behaviors and content tags). is the matching score of user U to content c;

[0058] S104: Dynamic parameter update and adaptive learning: After each new user behavior occurs, the following logic is executed:

[0059] (1) Update the user behavior vector using the following formula: , where γ∈[0,1] is the historical weight coefficient;

[0060] (2) Adjusting the behavior weight distribution , increase the weight of recent frequent behaviors in recommendations;

[0061] (3) Adjust the recommendation ranking in real time and use behavioral feedback (clicks, bounces, etc.) as negative sampling / positive sampling signals to optimize the recommendation model;

[0062] (4) When user behavior significantly deviates from the group center, its profile category is updated, triggering the cross-module dynamic linkage recommendation module to execute the action;

[0063] S105: In any module, if the user behavior vector deviates significantly, , then the cross-module dynamic linkage recommendation module is triggered to execute the action;

[0064] The cross-module dynamic linkage recommendation module triggers dynamic adjustments to content and function recommendations based on the user's current behavior, achieving real-time linkage between functional modules. The specific logical steps are as follows:

[0065] S201: The events collected by the multi-channel behavior data collection module are encapsulated into behavior event tuples: ,in The module where the current behavior occurs. is the operation behavior type, Behavioral parameters include time, content tags, and interaction words;

[0066] S202: Through the event processing engine Real-time semantic analysis is performed to identify the potential intention of the user's current operation. For example, if a user frequently uses "background erase" and "high-resolution restoration" → their goal is to infer e-commerce image optimization; if a user continuously browses "social avatar" images in the content stream → their interest is inferred to be inclined towards social scene creation. The user portrait model and historical behavior path are combined to calculate the behavior offset: , , it is regarded as a "behavior pattern mutation" and triggers cross-module dynamic linkage;

[0067] S203: Based on the behavioral intention and module dependency, determine the target module set to be linked: , and sort by priority;

[0068] S204: For each linked module , perform recommendation strategy update, and the formula used in the dynamic adjustment process is: ,in is the trigger response strength parameter, It is a trigger indicator function and takes 1 when the behavior deviation is significant;

[0069] The content of its recommended strategy update is as follows:

[0070] (1) Content flow module: Adaptive changes in content flow layout: record user long-term preferences, automatically locate the user's frequently visited preferred section each time logging in, and dynamically adjust the default sorting method, content density and display format of the content flow according to user usage habits; Personalized adjustment of content flow image layout: automatically adjust the image layout in the content flow according to the user's preferred style, and the content template of the preferred style is enhanced in visual presentation, and support dynamic adjustment of image display ratio according to user browsing habits;

[0071] (2) Adaptive layout of functional modules: The layout of the basic tool module and the creative tool module is dynamically adjusted according to the frequency of user use. High-frequency functions are automatically displayed in the front, and low-frequency function modules are automatically collapsed, saving interface space and improving user experience. At the same time, it supports customizable sorting of functional modules, and the system provides default sorting suggestions based on user habits;

[0072] (3) Dynamic adaptation of the personality of the intelligent agent module: The system analyzes the user's interaction style and response preferences, and intelligently adjusts the communication style and personality characteristics of the AI ​​Agent. For efficiency-oriented users, the intelligent agent module adopts a simple and efficient communication method, reduces redundant hierarchical guidance, and directly provides solutions. For creativity-oriented users, the intelligent agent module shows more imaginative and humorous personality characteristics. For learning-oriented users, the intelligent agent actively provides more tutorials and guidance information.

[0073] (4) Multi-terminal synchronization and scene continuation: User preferences and settings can be synchronized across devices to ensure a consistent personalized experience on different terminal devices. It supports cross-device continuation of creative tasks. The system automatically records the creative progress and environmental parameters. It also makes intelligent recommendations based on time and scene, giving priority to professional content during work hours and favoring entertainment creation during leisure time.

[0074] (5) Evolution of interaction mode: Analyze user operation habits, automatically adjust the response priority of the interaction mode, dynamically adjust the frequency and detail of operation prompts according to user proficiency, and support customized interaction processes. The system can learn the user's unique operation habits and optimize the response method;

[0075] S205: Once the linkage recommendation strategy takes effect, the UI interface immediately rearranges the module display order, and the linked modules quickly refresh the content or function display. If there is no user interaction delay, the current display rhythm is maintained;

[0076] S206: The system records the response effect of each linkage recommendation, including whether the user clicks on the recommended content, whether the length of stay after linkage increases, and whether the next action is triggered. The system updates the user model parameters based on these feedback data. , content matching rules , and linkage recommendation strategies , forming a self-learning closed loop of linkage recommendation;

[0077] This embodiment can accurately predict user needs and proactively locate relevant content, significantly reducing user browsing and searching time, improving the matching degree of first-screen content, and enhancing content acquisition efficiency. It can also dynamically adjust the content display strategy and function arrangement order of each functional module based on user behavior triggers, break down barriers between modules, and achieve intelligent linkage between multiple modules, providing users with a more personalized creation experience. It can effectively solve the core problems of content homogeneity, module isolation, and poor user experience faced by current AI content creation platforms, and provide users with a dynamic, intelligent, multi-module real-time content recommendation system based on multi-channel behavioral data, achieving a truly intelligent, personalized, and efficient creation and content consumption experience.

[0078] In this embodiment, during the user operation process, the system's multi-channel behavior data collection module collects in real time the user's behavior data in the basic tool module (such as background erasure, high-definition restoration), creation tool module (such as e-commerce templates, social avatars), content flow module (such as browsing, staying, liking) and creation agent module (such as command input, dialogue feedback), and constructs a behavior feature vector containing operation sequence, usage frequency, content preference, and creation style dimensions through the behavior analysis and modeling module. Subsequently, the behavior analysis and modeling module performs intent recognition and user portrait construction based on the vector, and predicts the user's current behavior through clustering algorithm and weighted scoring function. possible creative needs in the past; when the system detects that there is a significant deviation between user behavior and historical patterns, the cross-module dynamic linkage recommendation module is triggered. The cross-module dynamic linkage recommendation module dynamically adjusts the content display strategy and function arrangement order of each functional module according to the behavior trigger, such as automatically jumping to the content flow section, displaying specific tools in advance, and evoking personalized suggestions from intelligent agents; the recommendation score is adjusted using time weighting, and the recommendation parameters are continuously updated according to user clicks and feedback data, achieving millisecond-level response to user operations and multi-module collaborative recommendations, and ultimately realizing a highly personalized content and function recommendation process, improving creation efficiency and system usage experience.

[0079] For example:

[0080] The user selects and uses the "Background Erase" function in the Basic Tools module to process a product image;

[0081] The system analyzes the operation characteristics in real time and, combined with historical usage data, determines that the user may have e-commerce background optimization needs, and immediately triggers cross-module dynamic linkage and recommendation module linkage. The specific linkage is:

[0082] (1) The creation tool module automatically adjusts the displayed content and puts e-commerce related tools (such as e-commerce background images and e-commerce materials) in the "work" category in the front, improving the efficiency of users in discovering related tools;

[0083] (2) If the user further selects e-commerce related tools, the system will increase the prediction weight of e-commerce needs;

[0084] (3) The content flow module responds synchronously. When the user scrolls down to browse the content, the system automatically locates the content to the e-commerce related content area of ​​the "Work" section, and the "Work" option is highlighted in the bottom positioning bar;

[0085] (4) The intelligent agent module perceives the user's behavior path and actively pops up prompts to recommend optimization suggestions for e-commerce images to users, such as "You seem to be processing e-commerce images. Do you need recommendations for some background templates suitable for e-commerce?" or "I can help you directly generate product display images that meet e-commerce standards."

[0086] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A dynamic, intelligent, multi-module, real-time content recommendation system based on multi-channel behavioral data, characterized by: It includes a basic tool module, an authoring tool module, a content flow module, an authoring agent module, a multi-channel data collection and analysis system, and a cross-module dynamic linkage recommendation module. The multi-channel data collection and analysis system includes a multi-channel behavior data collection module and a behavior analysis and modeling module. The behavior analysis and modeling module constructs a user behavior profile and preference model through multi-dimensional data fusion analysis and updates the model parameters in real time to ensure that the recommendation system can dynamically respond to changes in user needs. The specific logical steps are as follows: S101: Extract structured behavior features from the original behavior data collected by the multi-channel behavior data collection module to form a multi-dimensional vector , which includes Recent activity of basic tools, Distribution of creative preference types, Average duration of content flow Distribution of user active time periods, Recent agent dialogue keyword frequency, Click the content label frequency of the behavior, normalize the behavior feature vector and perform time weighting processing to obtain = ; S102: Use a K-means clustering algorithm to generate a behavioral profile label for each user's behavioral feature vector. This label represents the content preference group to which the user belongs. This behavioral profile label serves as one of the input priors for the subsequent recommendation model. The behavioral profile labels include e-commerce-oriented users, creative experimental users, tool efficiency users, and beginner guidance users. S103: Construct a user-content scoring function and train a recommendation scoring model. The formula is: ,in is the weight of user feature dimension i, Score the relevance of the user to the content in the i-th dimension, is the matching score of user U to content c; S104: Dynamic parameter update and adaptive learning: After each new user behavior occurs, the following logic is executed: (1) Update the user behavior vector using the following formula: , where γ∈[0,1] is the historical weight coefficient; (2) Adjusting the behavior weight distribution , increase the weight of recent frequent behaviors in recommendations; (3) Adjust the recommendation ranking in real time and use behavioral feedback as negative sampling / positive sampling signals to optimize the recommendation model; (4) When user behavior significantly deviates from the group center, its profile category is updated, triggering the cross-module dynamic linkage recommendation module to execute the action; S105: In any module, if the user behavior vector deviates significantly, , then the cross-module dynamic linkage recommendation module is triggered to execute the action; The cross-module dynamic linkage recommendation module triggers dynamic adjustment of content and function recommendations based on the user's current behavior, realizing real-time linkage between functional modules. The specific logical steps are as follows: S201: The events collected by the multi-channel behavior data collection module are encapsulated into behavior event tuples: ,in The module where the current behavior occurs. is the operation behavior type, Behavioral parameters include time, content tags, and interaction words; S202: Through the event processing engine Perform real-time semantic analysis to identify the potential intention of the user's current operation, and calculate the behavior offset by combining the user portrait model and historical behavior path: , , it is regarded as a "behavior pattern mutation", triggering cross-module dynamic linkage; S203: Based on the behavioral intention and module dependency, determine the target module set to be linked: , and sort by priority; S204: For each linked module , perform recommendation strategy update, and the formula used in the dynamic adjustment process is: ,in is the trigger response strength parameter, It is a trigger indicator function and takes 1 when the behavior deviation is significant; S205: Once the linkage recommendation strategy takes effect, the UI interface immediately rearranges the module display order, and the linked modules quickly refresh the content or function display. If there is no user interaction delay, the current display rhythm is maintained; S206: The system records the response effect of each linkage recommendation, including whether the user clicks on the recommended content, whether the length of stay after linkage increases, and whether the next action is triggered. The system updates the user model parameters based on these feedback data. , content matching rules , and linkage recommendation strategies , forming a self-learning closed loop of linkage recommendation.

2. The dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data according to claim 1 is characterized in that: The basic tool module is located at the top of the product and provides a set of basic functions required for creation. It is used to provide basic editing functions of image processing, including picture editing, background erasing, high-definition restoration and lossless magnification functions.

3. The dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data according to claim 1 is characterized in that: The creation tool module is used to provide multiple types of image or video creation functions according to different application scenarios. It is divided into the following different categories according to the usage scenarios: Lifestyle: Contains portraits, couple portraits, pet portraits, social profile pictures, and other creation tools related to daily life; Work category: including e-commerce background images, e-commerce materials, logo creation, and tattoo creation professional creation tools; Popular categories: Recommend currently popular gameplay and user-preferred content.

4. The dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data according to claim 1 is characterized in that: The content flow module adopts an up-and-down sliding layout with a quick positioning bar at the bottom, allowing users to quickly locate the content area of ​​interest. It is used to display recommended content and dynamically adjust the sorting and display strategies based on user behavior.

5. The dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data according to claim 1 is characterized in that: The creative agent module exists in the form of a dialogue interface. Users can express their creative needs through dialogue and receive user instructions based on the dialogue form. The agent can recommend suitable scenes or complete the image / video generation task directly in the dialog box. The agent can actively provide creative suggestions, tool recommendations and operation guidance.

6. The dynamic intelligent multi-module real-time content recommendation system based on multi-channel behavior data according to claim 1 is characterized in that: The multi-channel behavior data collection module is used to collect user behavior data in the basic tool module, creation tool module, content flow module and creation agent module, which specifically includes the frequency and order of user function usage, creation tool preferences, content browsing tracks, dwell time, operation preferences, agent interaction records and instruction preferences, cross-module operation paths and timing relationships, and user-preferred content style characteristics.

Images