Multi-dimensional constraint adaptive user interface layout method and device based on AI

By collecting device performance and user behavior data, and using lightweight deep learning models and multi-dimensional constraint models to dynamically adjust the user interface, the problem of real-time perception of device performance and user intent is solved, improving the user experience and interface smoothness of low-end devices.

CN121658137APending Publication Date: 2026-03-13GREE ELECTRIC APPLIANCE INC OF ZHUHAI
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202512051494.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-31
Publication Date
2026-03-13

AI Technical Summary

Technical Problem

Existing technologies cannot perceive device performance status and user operating habits in real time, resulting in lag on low-end devices and a decline in user experience. Furthermore, they lack dynamic degradation mechanisms and cannot dynamically adjust the interface layout based on user intent and device performance at runtime.

Method used

By collecting device performance and behavior data from user terminals, a lightweight deep learning model is used to predict user operation intentions. Combined with a multi-dimensional real-time constraint model, the optimal user interface layout strategy is solved, the rendering parameters and layout structure of interface elements are dynamically adjusted, and a closed-loop feedback mechanism is introduced for online parameter updates.

Benefits of technology

When device computing power is limited, it can accurately identify high-frequency operation areas and prioritize rendering quality, reduce the complexity of non-interested areas, solve the lag problem on low-end devices, and improve the smoothness of the user interface and interaction efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121658137A_ABST
    Figure CN121658137A_ABST
Patent Text Reader

Abstract

The invention relates to an AI-based multi-dimensional constraint adaptive user interface layout method and device, and the method comprises the steps: collecting equipment performance data and user behavior data of a user terminal; inputting the user behavior data into a pre-trained lightweight deep learning model, and predicting a next operation intention and a high-frequency operation area of the user; inputting equipment performance monitoring data and the high-frequency operation area into a preset multi-dimensional real-time constraint model to solve a balanced optimal solution among different constraints, and obtaining an optimal user interface layout strategy meeting a preset system operation fluency threshold value; calculating complexity quantitative indexes of interface elements according to the optimal user interface layout strategy; and dynamically adjusting rendering parameters and a layout structure of the user interface according to the complexity quantitative indexes of the interface elements. By adopting the method and the device, the dynamic adjustment process of the user interface layout can be smoother.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, and in particular to an AI-based multidimensional constraint adaptive user interface layout method and apparatus. Background Technology

[0002] With the rapid development of mobile internet and IoT technologies, the variety of user terminal devices is increasing, ranging from high-performance workstations to resource-constrained IoT (Internet of Things) devices, with significant differences in hardware performance such as CPU / GPU computing power, memory, and bandwidth. Traditional user interface design typically employs static layouts or responsive layouts based on media queries. While this development model can adjust the interface according to screen size, it fixes the interface structure and style during the development phase, failing to perceive the real-time performance status of the device or the user's specific operating habits at runtime. In existing technologies, to address the experience issues caused by differences in device performance, US patent application US20030046401A1 discloses a method for dynamically determining an appropriate computer user interface. This solution dynamically selects between plain text, audio, or complex 3D graphics interfaces based on the client device's bandwidth and processing power to adapt to device limitations. However, this technology primarily relies on preset hardware thresholds for coarse-grained mode switching, such as jumping between different modes, lacking fine-grained control over interface elements such as animation frame rate, image precision, and component complexity. More importantly, this solution only considers hardware constraints and neglects the personalized behavioral intentions of users during operation. This can easily lead to a compromised experience in core functional areas due to a "one-size-fits-all" simplification strategy. On the other hand, intelligent analysis technologies for user behavior are constantly evolving. US Patent 10929110B2 discloses an AI-assisted user experience design evaluation system that uses sequence models such as LSTM (Long Short-Term Memory) to analyze user interaction logs, identify typical operation paths, and evaluate design effectiveness. However, this technology is mainly applied in the offline design evaluation stage, aiming to assist designers in optimizing product prototypes. It does not involve real-time reconstruction of the interface layout or dynamic adjustment of rendering strategies based on predicted user intentions during application runtime. Existing layout systems lack a real-time closed-loop feedback mechanism that combines "user intention prediction" with "underlying performance monitoring." In summary, existing technologies lack an intelligent engine capable of simultaneously sensing device performance bottlenecks and predicting user behavioral intentions in real time, and dynamically solving for the optimal layout strategy during runtime. When running complex applications on low-end devices, traditional solutions often suffer from lag due to the lack of dynamic degradation mechanisms. Furthermore, in scenarios where user behavior changes frequently, fixed layout logic struggles to match user expectations, resulting in inefficient interaction. Therefore, there is an urgent need for an AI-based adaptive user interface layout solution that achieves a dynamic balance between performance and user experience through multi-dimensional constraint-based collaborative optimization. Summary of the Invention

[0003] To overcome the problems existing in related technologies, a first aspect of the present invention proposes an AI-based multidimensional constraint adaptive user interface layout method, wherein the method includes: Collect device performance data and user behavior data from user terminals; The user behavior data is input into a pre-trained lightweight deep learning model to predict the user's next operation intention and high-frequency operation area. The device performance monitoring data and the high-frequency operation area are input into a preset multidimensional real-time constraint model to solve for the optimal balance between different constraints, thereby obtaining the optimal user interface layout strategy that satisfies the preset system operation smoothness threshold; wherein, the multidimensional real-time constraint model includes at least: device performance constraints and user behavior constraints. Based on the optimal user interface layout strategy, calculate the complexity metrics of the interface elements. Based on the complex quantification metrics of the interface elements, the rendering parameters and layout structure of the user interface are dynamically adjusted; wherein, the rendering parameters include: the number, duration and frame rate of user interface animations, the accuracy and dynamic replacement strategy of image resources, and the component granularity.

[0004] Optionally, after the step of dynamically adjusting the rendering parameters and layout structure of the user interface based on the complexity metrics of the interface elements, the method further includes: After adjusting the user interface layout, device performance monitoring data and user behavior data of the user terminal are collected, and the parameters of the multidimensional real-time constrained deep learning model are updated online through a closed-loop feedback mechanism.

[0005] Optionally, the steps of collecting device performance monitoring data and user behavior data of the user terminal include: determining whether to enable the adaptive sampling mechanism; if enabled, determining the sampling frequency according to the system load, and collecting device performance monitoring data and user behavior data of the user terminal according to the sampling frequency; if not enabled, collecting device performance monitoring data and user behavior data of the user terminal using a system-preset fixed sampling frequency.

[0006] Optionally, the multidimensional real-time constraint model further includes: interface complexity constraints and business priority constraints.

[0007] Optionally, the device performance monitoring data includes: CPU utilization, GPU utilization, and memory usage; the user behavior data includes: click heatmap, dwell time, and operation path.

[0008] Optionally, the step of inputting the device performance monitoring data and the high-frequency operation area into a preset multi-dimensional real-time constraint model to solve for the optimal balance solution between different constraints, and obtaining the optimal user interface layout strategy that satisfies a preset system operation smoothness threshold, includes: The device performance monitoring data and the high-frequency operation area are input into a preset multi-dimensional real-time constraint model. When the device performance monitoring data indicates that the system load exceeds a preset threshold and the user behavior constraint conditions indicate the existence of a high-frequency operation area, a conflict resolution mechanism is activated. The conflict resolution mechanism includes: maintaining the rendering level and interaction response priority of the high-frequency operation area unchanged, and reducing the complexity of interface elements in non-high-frequency operation areas.

[0009] Optionally, the step of dynamically adjusting the rendering parameters and layout structure of the user interface based on the complexity metrics of the interface elements includes: Based on the complexity metrics of the interface elements, adjust the number, duration, and frame rate of user interface animations, the loading accuracy of image resources, the dynamic replacement strategy of image resources, and the rendering granularity of components; determine the complexity score corresponding to different areas in the user interface based on the complexity metrics of the interface elements, and determine the rendering level based on the complexity score corresponding to each area.

[0010] Optionally, the step of determining the sampling frequency based on the system load includes: calculating the current system load coefficient based on the CPU utilization and memory usage in the device performance monitoring data; increasing the data acquisition frequency to obtain a high-precision user behavior path when the system load coefficient is lower than a first threshold; and decreasing the data acquisition frequency to reduce the occupation of device resources by monitoring behavior when the system load coefficient is higher than a second threshold.

[0011] Optionally, the step of calculating the complexity metrics of interface elements according to the optimal user interface layout strategy includes: determining the DOM node complexity, texture memory complexity, animation rendering complexity, interaction response complexity, and network resource complexity according to the optimal user interface layout strategy; and calculating the complexity metrics of interface elements based on the DOM node complexity, texture memory complexity, animation rendering complexity, interaction response complexity, and network resource complexity.

[0012] A second aspect of this invention provides an AI-based multidimensional constraint adaptive user interface layout device, wherein the device includes: The data acquisition module is used to collect device performance data and user behavior data from user terminals. The prediction module is used to input the user behavior data into a pre-trained lightweight deep learning model to predict the user's next operation intention and high-frequency operation area. The solution module is used to input the device performance monitoring data and the high-frequency operation area into a preset multidimensional real-time constraint model to solve for the optimal balance between different constraints, thereby obtaining the optimal user interface layout strategy that satisfies the preset system operation smoothness threshold; wherein, the multidimensional real-time constraint deep learning model includes at least: device performance constraints and user behavior constraints; The calculation module is used to calculate the complexity metrics of the interface elements according to the optimal user interface layout strategy. The adjustment module is used to dynamically adjust the rendering parameters and layout structure of the user interface based on the complex quantification indicators of the interface elements; wherein, the rendering parameters include: the number, duration and frame rate of user interface animations, the accuracy and dynamic replacement strategy of image resources, and the component granularity.

[0013] The third aspect of the present invention provides an electronic device, wherein the electronic device performs user interface layout using any one of the methods described in the above method embodiments, or the AI-based multidimensional constraint adaptive user interface layout device described in the above device embodiments.

[0014] The technical solution of this invention, in its first aspect, introduces a multi-dimensional real-time constraint model and conflict resolution mechanism. When device computing power is limited, it can accurately identify and prioritize the rendering quality of frequently accessed user areas based on AI prediction, while only degrading non-focused areas. This solves the lag problem on low-end devices and avoids the damage to the core business process experience caused by traditional "one-size-fits-all" degradation. In its second aspect, it provides refined adaptive control capabilities, specifically through the use of complex metric indicators to finely control interface elements, such as animation frame rate and component granularity, making the dynamic adjustment process of the user interface layout smoother and improving the user's perceived fluidity.

[0015] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0017] Figure 1 This is a flowchart illustrating an AI-based multidimensional constraint adaptive user interface layout method according to an exemplary embodiment. Figure 2 This is a structural block diagram of an AI-based multidimensional constraint adaptive user interface layout device according to an exemplary embodiment. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] According to an embodiment of this application, an AI-based multidimensional constraint adaptive user interface layout method is provided, referring to... Figure 1 The methods include: 100: Collect device performance data and user behavior data from user terminals.

[0021] In this embodiment, the AI-based multidimensional constraint adaptive user interface layout method is applied to an electronic device. The storage medium in the electronic device stores a computer program related to the AI-based multidimensional constraint adaptive user interface layout, and the processor in the electronic device executes the computer program during runtime.

[0022] Device performance monitoring data may include, but is not limited to: CPU utilization, GPU utilization, and memory usage. When collecting device performance monitoring data, the Performance API or Web Vitals tool can be used in conjunction with these metrics. User behavior data may include, but is not limited to: click heatmaps, dwell time, and operation paths.

[0023] In the specific implementation process, an adaptive sampling mechanism can also be introduced at this stage to calculate the system load coefficient based on CPU utilization and memory usage, thereby determining whether to increase or decrease the sampling frequency.

[0024] In one optional embodiment, the method of collecting device performance monitoring data and user behavior data of user terminals may include the following sub-steps: Sub-step 1: Determine whether to enable the adaptive sampling mechanism; if yes, start sub-step 2; if no, start sub-step 3.

[0025] Sub-step 2: If enabled, determine the sampling frequency based on the system load, and collect device performance monitoring data and user behavior data of the user terminal according to the sampling frequency; Sub-step 3: If not enabled, use the system's preset fixed sampling frequency to collect device performance monitoring data and user behavior data from user terminals.

[0026] One feasible way to determine the sampling frequency based on system load is to calculate the current system load coefficient based on the CPU utilization and memory usage in the device performance monitoring data; when the system load coefficient is lower than a first threshold, increase the data acquisition frequency to obtain a high-precision user behavior path; when the system load coefficient is higher than a second threshold, decrease the data acquisition frequency to reduce the occupation of device resources by monitoring behavior.

[0027] It should be noted that the specific values ​​of the first threshold and the second threshold can be flexibly set by those skilled in the art, and no specific restrictions are imposed on them in the embodiments of this application.

[0028] The system load factor can be calculated using a dynamic weighting algorithm. The specific calculation formula is: System load factor = (CPU utilization × 0.4 + GPU utilization × 0.3 + memory utilization × 0.3) × 100%.

[0029] The frequency adjustment rules for the adaptive sampling mechanism can be set as follows: when the system load factor is ≤40%, the sampling frequency is set to 2 times / second; when 40% < system load factor ≤70%, the sampling frequency is set to 1 time / second; when the system load factor is >70%, the sampling frequency is set to 1 time / 5 seconds.

[0030] 102: Input user behavior data into a pre-trained lightweight deep learning model to predict the user's next operation intention and high-frequency operation area.

[0031] The lightweight deep learning model in this embodiment can be a Long Short-Term Memory (LSTM) network, but it is not limited to this and can also be any other suitable deep learning model.

[0032] By predicting the user's next action intention, the area of ​​the user's next action can be determined, and this area can also be classified as a high-frequency operation area.

[0033] 104: Input the equipment performance monitoring data and high-frequency operation area into the preset multi-dimensional real-time constraint model to solve the optimal solution of the balance between different constraints, and obtain the optimal user interface layout strategy that meets the preset system operation smoothness threshold.

[0034] The multidimensional real-time constraint model includes at least: device performance constraints and user behavior constraints. It simultaneously evaluates the current system performance status and user operational intentions, intelligently outputting a unified layout strategy.

[0035] Device performance constraints may include, but are not limited to: minimum FPS requirements, memory limits, etc.; user behavior constraints may include, but are not limited to: critical path response latency.

[0036] In an optional embodiment, the multidimensional real-time constraint model further includes interface complexity constraints and business priority constraints. When performing predictions, this optional multidimensional real-time constraint model constructs a real-time constraint vector: C = {P, U, B, L}; P = performance metrics (CPU, GPU, memory, FPS); U = User behavior intent and hotspot weight; B = Business element priority, such as important entry points and promotional slots; L = Current layout complexity; The multidimensional real-time constraint model generates the optimal user interface layout strategy by solving the function: Layout_Strategy = f(P, U, B, L).

[0037] In one optional embodiment, the method of inputting device performance monitoring data and high-frequency operation areas into a preset multi-dimensional real-time constraint model to solve for the optimal balance solution between different constraints, thereby obtaining the optimal user interface layout strategy that satisfies a preset system smoothness threshold, may include: Input the equipment performance monitoring data and high-frequency operation areas into the preset multi-dimensional real-time constraint model; when the equipment performance monitoring data indicates that the system load exceeds the preset threshold and the user behavior constraint conditions indicate the existence of a high-frequency operation area, activate the conflict resolution mechanism. The conflict resolution mechanism includes: maintaining the rendering level and interaction priority of high-frequency operation areas unchanged, and reducing the complexity of UI elements in low-frequency operation areas. Methods for reducing UI element complexity may include, but are not limited to: reducing animation frame rate, reducing image resource resolution, or replacing them with placeholder components.

[0038] The conflict resolution mechanism set in this optional embodiment is not only to predict the user's operation intention and detect the device performance, but also has a clear handling principle when the two conflict, namely, to maintain the high-frequency operation area and sacrifice the non-high-frequency area.

[0039] In the specific implementation process, the conflict resolution mechanism triggering conditions and weighted calculation method of the multidimensional real-time constraint model can be as follows: Triggering conditions: When the system load factor is ≥70% and the user interaction frequency is ≥5 times / minute, the conflict resolution between performance constraints and behavioral constraints is triggered; when the matching degree between the user behavior prediction result and the business guidance goal is ≤50%, the conflict resolution between behavioral constraints and business priority constraints is triggered.

[0040] 106: Calculate the complexity metrics of interface elements based on the optimal user interface layout strategy.

[0041] In this step, the UI, or user interface, is transformed from a subjective experience into a calculable object, and the complexity of UI elements is quantified and used as a basis for control.

[0042] In one alternative embodiment, the complexity metric of interface elements can be calculated according to the optimal user interface layout strategy as follows: Based on the optimal user interface layout strategy, determine the complexity of DOM nodes, texture memory, animation rendering, interaction response, and network resources; and calculate the complexity metrics of interface elements based on the complexity of DOM nodes, texture memory, animation rendering, interaction response, and network resources.

[0043] In actual implementation, the weight of DOM node complexity can be set to 30%. DOM node complexity = actual number of DOM nodes / preset DOM threshold. The preset DOM threshold for mobile devices is 500, and the preset DOM threshold for desktop devices is 1000. The weight of texture memory complexity can be set to 25%. Texture memory complexity = Σ (texture memory of a single image × number of images) / available video memory capacity of the device, where texture memory of a single image = image width × image height × pixel bit depth / 8; The weight of animation rendering complexity can be set to 20%, and the animation rendering complexity = (animation frame rate × number of animation elements × animation duration) / 1000; The weight of interaction response complexity can be set to 15%, and the interaction response complexity = actual interaction response time / 100ms, where the actual interaction response time is the time interval between the user operation and the interface feedback. The weight of network resource complexity can be set to 10%, and network resource complexity = current page resource loading size / (device network bandwidth × 1000). The final value of the complexity metric is calculated as follows: 0.3 × DOM node complexity + 0.25 × texture memory complexity + 0.2 × animation rendering complexity + 0.15 × interaction response complexity + 0.1 × network resource complexity. The value range is [0, 10]. When the value is ≥ 7, UI simplification is triggered.

[0044] 108: Dynamically adjust the rendering parameters and layout structure of the user interface based on the complexity metrics of the interface elements.

[0045] The rendering parameters include: the number, duration, and frame rate of user interface animations; the precision and dynamic replacement strategy of image resources; and component granularity. Component granularity includes, but is not limited to: whether placeholders, skeleton screens, and lazy loading are enabled. In this embodiment, the entire user interface display is not downgraded; instead, high-frequency areas are preserved, and low-frequency (i.e., non-critical) areas are simplified. The core principle is regional differentiation.

[0046] In one alternative embodiment, the complexity metric involves controlling the following rendering parameters. That is, the way to dynamically adjust the rendering parameters and layout structure of the user interface based on the complexity metric of the interface elements can be: adjusting the number, duration, and frame rate of user interface animations, the loading accuracy of image resources, the dynamic replacement strategy of image resources, and the rendering granularity of components based on the complexity metric of the interface elements; determining the complexity score corresponding to different areas in the user interface based on the complexity metric of the interface elements, and determining the rendering level based on the complexity score corresponding to each area.

[0047] In actual implementation, the system assigns different complexity scores to different interface areas based on the optimal user interface layout strategy. When the area score exceeds the upper limit allowed by the current performance status, it automatically performs degraded rendering.

[0048] In one optional embodiment, after dynamically adjusting the rendering parameters and layout structure of the user interface based on the complexity metrics of the interface elements, the following process is also included: After adjusting the user interface layout, device performance monitoring data and user behavior data of user terminals are collected, and online parameter updates are performed on the multidimensional real-time constrained deep learning model through a closed-loop feedback mechanism.

[0049] This closed-loop feedback mechanism enables online parameter updates for multidimensional real-time constrained deep learning models, thereby continuously improving the model's prediction accuracy.

[0050] The triggering timing for online parameter updates of multidimensional real-time constrained deep learning models can include timed triggering and threshold triggering. In specific implementations, those skilled in the art can flexibly set these timings according to actual needs; this application embodiment does not impose specific limitations on this.

[0051] Among them, the timed trigger can be set to a full update every 24 hours, and when the system load coefficient is ≤40%, it can be advanced to once every 12 hours; the scenario trigger mechanism is a targeted update when the user scenario changes; the threshold trigger includes an immediate incremental update when the prediction error rate of the multi-dimensional real-time constrained deep learning model reaches 15%.

[0052] The data alignment method for online parameter updates of multidimensional real-time constrained deep learning models can be set as follows: the edge only performs lightweight preprocessing and compression of the feedback data for transmission, the cloud performs batch training based on a distributed framework, extracts parameter increment packets of ≤100KB and sends them out, the edge completes the update in the background, and a breakpoint resume mechanism is used to ensure the success rate.

[0053] User behavior data includes, but is not limited to: click heatmap coordinate sequence, page dwell time, and operation path vector; collected user terminal device performance indicators include, but are not limited to: page load time, first content rendering time (FCP), and frames per second (FPS).

[0054] The AI-based adaptive user interface layout method provided in this application has two main aspects. First, by introducing a multi-dimensional real-time constraint model and conflict resolution mechanism, when device computing power is limited, it can accurately identify and prioritize the rendering quality of frequently accessed user areas based on AI prediction, while only degrading non-focused areas. This can solve the lag problem on low-end devices and avoid the damage to the core business process experience caused by traditional "one-size-fits-all" degradation. Second, it provides refined adaptive control capabilities, specifically by using complex metric indicators to finely control interface elements, such as animation frame rate and component granularity, making the dynamic adjustment process of the user interface layout smoother and improving the user's perceived smoothness.

[0055] This application also provides an AI-based adaptive user interface layout system. This system mainly includes a data acquisition module, an AI decision-making module, a layout adjustment module, and a feedback optimization module. These modules work collaboratively to achieve intelligent adaptive control of the user interface (UI) layout. First, the data acquisition module collects real-time device performance monitoring data and user behavior data from the user terminal. Then, the AI ​​decision-making module uses a pre-trained deep learning model, such as a Long Short-Term Memory (LSTM) network, to process the collected data and predict the user's next operational intention and high-frequency operation areas. Simultaneously, a multi-dimensional real-time constraint model is pre-constructed in the system. This model includes performance constraints and user behavior constraints. Combining real-time device performance and predicted high-frequency areas, an algorithm is used to find the optimal user interface layout strategy that satisfies a preset system smoothness threshold.

[0056] After obtaining the optimal user interface strategy, the layout adjustment module calculates the complexity metrics of interface elements based on this strategy. These metrics include the number / duration / frame rate of UI animations, the loading accuracy / dynamic replacement strategy of image resources, and the rendering granularity of components, such as skeleton screens / lazy loading. The system assigns complexity score caps to different interface areas according to the strategy. When the existing score of an area exceeds the cap, downgraded rendering is automatically performed. Specifically, when the device load is high and there are high-frequency operation areas, a conflict resolution mechanism is triggered: the rendering level and interaction priority of high-frequency areas remain unchanged, while the complexity of non-high-frequency areas is forcibly reduced, such as reducing animation frame rates or replacing them with placeholders.

[0057] Finally, the feedback optimization module collects adjusted performance metrics and user interaction data, and updates the AI ​​model's parameters online through a closed-loop feedback mechanism to achieve continuous evolution of the layout strategy.

[0058] The following describes the AI-based multidimensional constraint adaptive user interface layout method provided in this application, with reference to specific embodiments.

[0059] Example 1: Adaptive Layout Optimization in High-Load Mobile Scenarios This embodiment simulates a scenario where a complex e-commerce application runs on a resource-constrained low-to-mid-range mobile device. The system first collects the device's current CPU utilization and memory usage data via the Performance API. When the system detects a high-load state, such as CPU usage >80%, it activates a performance-aware layout optimization process. At this point, a user behavior prediction model based on LSTM analyzes the user's operation path over the past 30 seconds, identifying the "Add to Cart" button and "Product Details Image" as high-frequency operation areas, while the "Recommended Product Carousel" at the bottom is a low-frequency area.

[0060] The multi-dimensional real-time constraint deep learning model identifies the conflict between performance bottlenecks and interaction requirements, outputting a layout strategy of "preserving local excellence and degrading global performance." The layout adjustment module then calculates the complexity metrics for each area. For high-frequency areas, the original high-definition image resources and real-time response logic are maintained; for low-frequency areas, a degradation operation is performed: the carousel component is replaced with a static placeholder or a low-resolution thumbnail, and all CSS3 transition animations in that area are turned off. After the adjustment is complete, the system continuously monitors FPS changes and feeds the performance data before and after the operation back to the cloud-based AI model as samples.

[0061] The adaptive layout optimization method for high-load mobile scenarios provided in this specific embodiment, while preserving the core business function of smooth shopping and ordering experience, increases the average FPS of high-load devices from 18 to 35 by sacrificing the visual effects of non-critical areas, which can effectively avoid application crashes or severe lag.

[0062] Specifically, the high-load mobile e-commerce scenario implementation example is based on an Android phone with a quad-core CPU and 4GB of RAM, running an e-commerce promotion application. The specific process is as follows: Data collection phase The data acquisition module collects performance data based on the Performance API: CPU load 85%, GPU load 72%, memory usage 75%, system load coefficient = (85×0.4+72×0.3+75×0.3)×100%≈79.1%>70%, and the adaptive sampling unit reduces the frequency from 1 time / second to 1 time / 5 seconds. Behavioral data collection shows that the user clicked the "618 Limited-Time Offer" button 3 times within 5 minutes, with each click lasting over 10 seconds. The user interaction frequency = (3×0.6+2×0.4) / 5 = 0.8 times / minute, and the user swiped 2 times.

[0063] AI decision-making stage The LSTM model analyzes behavioral data and predicts that the user will click on the product card under "Limited-Time Offer" next. If the prediction probability for this area is 92%, then this area is determined to be a high-frequency operation area. The constraint solving unit identifies conflicts: 79.1% system load performance constraint triggers + high-frequency area is clearly defined (behavioral constraint trigger, priority weight = 79.1% × 0.5 + 0.8 × 0.5 ≈ 0.795; decision: maintain rendering of the high-frequency area and reduce the complexity of other areas).

[0064] User interface layout adjustment phase The complexity calculation unit calculates the metrics for each area: The "Limited-Time Flash Sale" area has 80 DOM nodes (complexity 80 / 500 = 0.16) and 30MB of texture memory (30 / 512 ≈ 0.058), resulting in a final metric of 0.3 × 0.16 + 0.25 × 0.058 + ... ≈ 1.2; the bottom "Historical Orders" area has 120 DOM nodes and 40MB of texture memory, with a metric of ≈ 4.8. Adjustments were made: the entry point was changed from "icon + text" to plain text, the carousel animation was disabled, and lazy loading was enabled. The adjustments used partial refresh, taking ≤200ms with no stuttering.

[0065] Feedback Optimization Phase Feedback module data collection: Page FPS increased from 18 to 38, click-through rate of the "Limited-Time Offer" area increased from 12% to 34%, and user conversion rate increased from 8% to 26%. Model prediction error rate = |Actual Clicks - Predicted Clicks| / Actual Clicks = 5% < 15%, incremental update will not be triggered for now; after 24 hours, the system load drops to 55%, triggering a full update. After cloud training, an incremental parameter package (85KB) is sent out, and the update is completed on the device when the phone screen is off, with a 100% update success rate.

[0066] The linear programming expression for the multidimensional real-time constrained deep learning model is as follows: Let the layout strategy variable be X = [x1, x2, ..., x n [where x1 represents the number of UI animations, x2 represents the image resolution level, x3 represents the component granularity level, ..., x] n This represents the priority of the high-frequency operation area layout. The objective function is used to represent maximizing the overall benefit of the layout. The objective function is as follows: Max F (X) = α×P (X) + β×U (X) + γ×B (X) – δ×C (X) P(X): Performance benefit function, P(X)=1 - (system load coefficient × 0.8 + page load time / preset load threshold × 0.2), value range [0,1]; U(X): User experience benefit function, U(X) = User interaction frequency × 0.7 + Click-through rate improvement rate × 0.3, with a value range of [0,1]. B(X): Business revenue function, B(X) = Business target achievement rate × 0.9 + User conversion rate × 0.1, with a value range of [0,1]; C(X): Layout adjustment cost function, C(X) = Layout refactoring time / Preset refactoring threshold × 0.6 + Resource reload amount / Preset resource threshold × 0.4, with a value range of [0,1]. α, β, γ, and δ are weighting coefficients, and α+β+γ+δ=1. The default configuration is α=0.35, β=0.3, γ=0.25, and δ=0.1.

[0067] Constraints: Performance constraint: System load factor ≤ preset load threshold (e.g., 70%), i.e. (CPU utilization × 0.4 + GPU utilization × 0.3 + memory utilization × 0.3) × 100% ≤ 70; Behavioral constraints: Response latency in high-frequency operation areas ≤ preset response threshold (e.g., 100ms), i.e., x n The corresponding regional interaction response time is ≤100; Interface complexity constraint: Complexity score ≤ preset complexity threshold (e.g., 6 points out of 10), i.e., 0.3×DOM complexity + 0.25×texture complexity + ... + 0.1×network complexity ≤ 6; Business priority constraints: Layout priority x in core business areas (such as the "flash sale" area in e-commerce applications) n ≥Preset priority threshold (e.g., 8 levels, out of 10).

[0068] The loss function for the LSTM model is designed as follows: The loss function used for training the LSTM model is the multi-task loss function, as shown in the following formula: Loss = λ1×Loss p + λ2×Loss + λ3×Loss

[0069] Loss p Performance prediction loss, calculated using mean squared error (MSE), Loss p =1 / N×Σ(Predicted system load factor - Actual system load factor)², where N is the sample size; Loss User behavior prediction loss is calculated using cross-entropy loss. =-1 / N×Σ(y i ×log (p i ) + (1-y i )×log (1-p i )), y i A label indicating whether the user clicked the button (1 = clicked, 0 = not clicked), p i Predict click probabilities for the model; Loss Loss on business objectives, calculated using MAE (Maximum Absolute Error). =1 / N×Σ|Predicted business target achievement rate - Actual business target achievement rate|; λ1, λ2, and λ3 are the loss weights, with the default configuration being λ1=0.4, λ2=0.35, and λ3=0.25.

[0070] The process of using a multidimensional real-time constrained deep learning model to solve for the optimal user interface layout strategy is as follows: Input layer: Receives device performance data from the data acquisition module: CPU / GPU load, memory usage; user behavior data: click heatmap, interaction frequency; business target data: business metric target values; Preprocessing layer: Calculates system load coefficient, user interaction frequency, and complex quantification indicators, normalizes the data, and maps the data to the [0,1] interval; Constraint Decision Layer: Determines whether performance constraints, behavioral constraints, interface complexity constraints, and business priority constraints are met. If all are met, the current layout strategy is output. If there are conflicts, the conflict resolution layer is entered. Conflict resolution layer: Calculates the priority weight of each constraint according to the preset weighting rules, and prioritizes satisfying constraints with higher weights. For example, if the performance constraint has a higher weight, the complexity of non-high frequency areas will be reduced. Optimization solution layer: Based on the linear programming objective function Max F(X), the gradient descent algorithm is used to solve for the optimal solution of the layout strategy variable X; Output layer: Outputs the optimal layout strategy to the layout adjustment module.

[0071] Timing of online parameter updates for multidimensional real-time constrained deep learning models: Scheduled Trigger: The default configuration is to trigger a full parameter update every 24 hours. If the device is under low load (system load factor ≤ 40%), the trigger will be moved to once every 12 hours. Threshold trigger: When the "model prediction error rate" in the feedback data is greater than or equal to the preset error threshold (e.g., 15%), the incremental parameter update is triggered immediately; where the model prediction error rate = 1 / N×Σ|predicted user behavior - actual user behavior|×100%, and N is the sample size of the feedback data; Scene trigger: When a user scene change is detected, such as switching from "browsing scene" to "order scene", a targeted parameter update is triggered. During the update, only the model parameters related to the current scene are updated, such as the prediction parameters of the high-frequency operation area in the order scene.

[0072] Data alignment: Edge-side data preprocessing: For low-computing devices such as Android phones with 4-core CPUs and 4GB of RAM, only lightweight preprocessing such as data deduplication and outlier removal is performed on the feedback data. No model training is performed. The preprocessed data is compressed to 30% of its original size using a compression algorithm to reduce network transmission overhead. Cloud-based training and parameter distribution: The cloud server receives preprocessed data from multiple devices and performs batch model training, such as using the distributed training framework TensorFlow Distributed. After training, the incremental package of model parameters is extracted. The incremental package can contain only the changed parameters and its size is controlled within 100KB. End-side parameter reception and update: After receiving the parameter incremental packet, the device updates the parameters when the screen is off or the application is running in the background. The update process adopts the "breakpoint resume" mechanism. If the network is interrupted, the update will continue when the network is connected again to ensure the success rate of the update. Data privacy protection: Before data transmission on the device side, a federated learning approach of "local encryption + parameter aggregation" is used to upload only the model gradients rather than the original user data, thus avoiding privacy leaks.

[0073] The difference between edge and cloud Computing capability adaptation: The edge only performs data acquisition, preprocessing and parameter reception, without model training to avoid consuming CPU / GPU resources, while the cloud undertakes large-scale data training and parameter optimization, which meets the hardware limitations of low computing power devices; Update efficiency optimization: The device side adopts "incremental update" instead of "full update", and the parameter package size is only 5%-10% of the full model. For example, the full model is 200MB, and the incremental package is 10-20MB, which reduces device storage usage and network transmission time. Real-time performance guarantee: Cloud training adopts a "rolling training" mechanism to process new data once per hour to ensure the real-time update of parameters. On the edge side, a dual triggering mechanism of "timed + threshold" is used to avoid frequent updates that consume resources and to ensure model accuracy.

[0074] Example 2: Refined Experience Enhancement in High-Performance Desktop Scenarios This embodiment simulates a scenario where a data visualization dashboard is running on a high-performance desktop browser equipped with a dedicated graphics card. The system detects that both the device's CPU and GPU loads are below 20%, determining that current resources are sufficient. The adaptive sampling mechanism automatically increases the data acquisition frequency from the default 1Hz to 10Hz to capture extremely subtle user mouse hovering behavior.

[0075] Based on high-frequency sampling data, the AI ​​model accurately predicts that users will soon zoom in on the "Sales Trend Chart" component. The constraint model solves to determine the currently permissible extremely high interface complexity. The layout adjustment module then increases the component's rendering level, preloads high-definition vector graphics resources, and enables particle effects animations. Simultaneously, complexity quantification allows for globally enabled Gaussian blur backgrounds and real-time lighting and shadow rendering. A closed-loop feedback mechanism records the increase in user dwell time in this high-fidelity mode, positively incentivizing the model parameters.

[0076] The method provided in this embodiment utilizes idle system resources to deliver an exceptional visual experience, extending the average user page dwell time by 20% and increasing the interaction depth by 15%.

[0077] Example 3: Coordinated Adjustment under Network Fluctuations and Dynamic Service Priorities This embodiment focuses on a complex scenario where network bandwidth is unstable and business objectives change. In this case, "network bandwidth constraints" and "business priority constraints" are added to the multi-dimensional constraint model. Assume the system detects that the network downlink speed has dropped to 200KB / s, while the backend is configured with a high business priority strategy for the "Double 11 Shopping Festival".

[0078] While users may be accustomed to clicking on "Order History," business constraints require guiding users to focus on the "Main Promotion Venue." The constraint solver calculates the optimal solution: prioritizing lightweight resources such as WebP format images for the "Main Promotion Venue," while ensuring page loading speed but limited by bandwidth, and collapsing or delaying the loading of the "Order History" entry. Even with different user habits, the user interface layout balances business needs and user habits through visual guidance such as highlighting the main venue, while strictly controlling the total package size to accommodate low bandwidth.

[0079] The method provided in this embodiment reduces the first-screen rendering time (FCP) by 40% compared to traditional solutions in weak network environments, while the click-through rate of promotional activities remains stable under poor network conditions.

[0080] Comparative Example 1: This example uses an adaptive layout system and a mid-range smartphone configured with a quad-core CPU and 4GB of RAM, running a test application integrating the AI ​​engine from Example 1. It simulates user actions for one hour, including browsing, clicking, and rapid swiping, while simulating random CPU load fluctuations in the background. This example achieved an average FPS of 52, 12 junks, an average page load time of 1.2 seconds, and a user task completion rate of 95%.

[0081] Compared to the traditional responsive layout system in Comparative Example 2, this embodiment sets the experimental conditions as follows: the hardware and test scripts are exactly the same as in Comparative Example 1. It uses traditional CSS media queries and fixed JavaScript logic, lacks AI prediction and real-time performance constraint solving capabilities, and has a fixed UI complexity adapted to medium configurations. This embodiment achieves an average FPS of 38, 85 stutters, an average page load time of 1.8 seconds, and a user task completion rate of 78%.

[0082] Example 4: High-load mobile e-commerce scenario This embodiment is based on an Android phone with a quad-core CPU and 4GB of RAM, running an e-commerce promotion application. The specific process is as follows: 1. Data Acquisition Phase The data acquisition module collects performance data based on the Performance API: CPU load 85%, GPU load 72%, memory usage 75%, system load coefficient = (85×0.4+72×0.3+75×0.3)×100%≈79.1%>70%, and the adaptive sampling unit reduces the frequency from 1 time / second to 1 time / 5 seconds. Behavioral data collection shows that the user clicked the "618 Limited-Time Offer" button 3 times within 5 minutes, with each click lasting over 10 seconds. The user interaction frequency = (3×0.6+2×0.4) / 5 = 0.8 times / minute, and the user swiped 2 times.

[0083] 2. AI Decision-Making Stage The LSTM model analyzes behavioral data and predicts that the user will click on the product card under "Limited-Time Offer" next, determining that this area is a high-frequency operation zone. The constraint solving unit identifies conflicts: system load 79.1% (performance constraint triggered) + clear high-frequency area (behavioral constraint triggered), priority weight = 79.1% × 0.5 + 0.8 × 0.5 ≈ 0.795, decision: maintain rendering of the high-frequency area and reduce the complexity of other areas.

[0084] 3. Layout Adjustment Phase The complexity calculation unit calculates the metrics for each area: The "Limited-Time Flash Sale" area has 80 DOM nodes (complexity 80 / 500 = 0.16) and 30MB of texture memory, resulting in a final metric of 0.3 × 0.16 + 0.25 × 0.058 + ... ≈ 1.2; the bottom "Historical Orders" area has 120 DOM nodes and 40MB of texture memory, with a metric of approximately 4.8. Adjustments were made: the entry point was changed from "icon + text" to plain text, the carousel animation was disabled, and lazy loading was enabled. The adjustments used partial refresh, taking ≤200ms with no stuttering.

[0085] 4. Feedback and Optimization Phase Feedback module data collection: Page FPS increased from 18 to 38, click-through rate of the "Limited-Time Offer" area increased from 12% to 34%, and user conversion rate increased from 8% to 26%. Model prediction error rate = |Actual Clicks - Predicted Clicks| / Actual Clicks = 5% < 15%, incremental update will not be triggered for now; after 24 hours, the system load drops to 55%, triggering a full update. After cloud training, an 85KB parameter increment package is sent out, and the update is completed on the device when the phone screen is off, with a 100% update success rate.

[0086] Example 5 This embodiment is based on an AI-driven adaptive user interface layout engine, which achieves dynamic adaptation through the collaboration of four major modules. The specific technical solution is as follows: 1. Data Acquisition Module This module collects performance data such as CPU / GPU load and memory usage based on the Performance API, and collects user data such as click heatmaps and operation paths through behavioral tracking. The collection method is compatible with the Web Vitals third-party tool. The core innovation lies in the adaptive sampling unit: it calculates the system load coefficient (CPU×0.4+GPU×0.3+Memory×0.3) through a dynamic weighting algorithm, and adjusts the sampling frequency in three levels according to the coefficient (2 times / second, 1 time / second, and 1 time / 5 seconds respectively), balancing data accuracy and resource consumption. For example, when the CPU load of a low-end mobile phone reaches 85%, the sampling frequency automatically drops to 1 time / 5 seconds to avoid consuming too much computing power.

[0087] 2. AI Decision Module As the core unit, this module comprises two main sub-units: Model training and inference unit: An optimized lightweight LSTM model is used, with the number of model parameters controlled within 5 million to adapt to low computing power devices. It is pre-trained with 100,000 performance data and 200,000 behavioral data to predict the user's next operation intention, such as predicting the user to click "limited time flash sale" in e-commerce scenarios, with an accuracy of over 85%.

[0088] Constraint Solving Unit: Constructs a multi-dimensional constraint model including performance, behavior, interface complexity, and business priority, and uses linear programming to solve for the optimal solution. The objective function is Max F(X) = 0.35P(X) + 0.3U(X) + 0.25B(X) - 0.1C(X), where P(X) = 1 - (System load × 0.8 + Loading time / Threshold × 0.2), U(X) = Interaction frequency × 0.7 + Click-through rate × 0.3, B(X) = Business achievement rate × 0.9 + Conversion rate × 0.1, and C(X) = Refactoring time / Threshold × 0.6 + Resource quantity / Threshold × 0.4. Constraints include system load ≤ 70%, high-frequency area response ≤ 100ms, and complexity index ≤ 7, ensuring the global optimization of the layout strategy.

[0089] 3. Layout Adjustment Module This module works in conjunction with a cross-platform execution unit through a complexity calculation unit: Complexity calculation unit: The ComplexityScore (0-10 points) is calculated based on 5 dimensions, including the number of DOM nodes and texture memory. For example, when there are 400 DOM nodes on a mobile page, the DOM complexity = 400 / 500 = 0.8. When the texture memory usage is 200MB, the texture complexity = 200 / 512 ≈ 0.39. The final index = 0.3×0.8 + 0.25×0.39 + ... ≈ 3.2, which does not need to be simplified. If the index is ≥7, an adjustment is triggered.

[0090] Cross-platform execution unit: Supports adaptation to mobile, desktop, and embedded devices. For mobile devices, image resolution is prioritized for compression, reducing it from 2K to 1080P. For desktop devices, 30fps animations are retained. Embedded devices enable skeleton screens. Adjustments utilize partial refresh technology, updating only changed areas, such as replacing only the "Historical Orders" entry style, to avoid page redraw stuttering.

[0091] 4. Feedback Optimization Module This module enables closed-loop optimization of edge-cloud collaboration: Data acquisition unit: Collects adjusted data such as FPS, FCP, and click-through rate, such as recording changes in click-through rate of the "limited-time sale" area in e-commerce scenarios.

[0092] Model update unit: Adopting the "device-side preprocessing + cloud-based training" mode, the device side only performs data deduplication and compression, while the cloud uses the TensorFlow Distributed framework for batch training, extracting parameter increment packets of ≤100KB and distributing them. Update triggering timings include timed updates, thresholds, and scene switching. The device side completes the update in the background, with a breakpoint resume success rate of over 99%.

[0093] The AI-based multidimensional constraint adaptive user interface layout method provided in this specific embodiment has the following beneficial effects: 1. Dual improvement in performance and user experience: Through performance-aware optimization, page loading time on low-end devices is reduced by an average of 30%, and FPS is increased to over 38 frames; adjustments to high-frequency areas shorten the interaction path by 60%, and user conversion rate is increased by 15%-18%.

[0094] 2. Global Constraint Optimization: Multi-dimensional constraint deep learning models resolve the conflict between "performance-behavior-business". For example, in a promotional scenario, when the CPU load is 85%, the "limited-time purchase" area is rendered in high definition, while the complexity of non-critical areas is reduced, thus balancing performance and business objectives.

[0095] 3. Low-computing-power device adaptation: Lightweight model + lightweight processing on the device side. When running on a mobile phone with a 4-core CPU and 4GB of memory, the CPU utilization rate is ≤15% and the memory usage is ≤200MB.

[0096] 4. Continuous iterative optimization: The closed-loop feedback mechanism improves the model prediction accuracy by 2%-3% per month, and the layout strategy evolves dynamically with user habits and device status.

[0097] Example 6 This embodiment provides an AI-based multi-dimensional constraint adaptive user interface layout engine, deployed in a cross-platform e-commerce application. The application needs to be compatible with low-end Android phones, mid-to-high-end smartphones, and desktop devices, and needs to cope with the high-frequency changes in user behavior during major promotional periods.

[0098] The AI-based multidimensional constraint adaptive user interface layout engine includes: a data acquisition module, an AI decision-making module, a layout adjustment module, and a feedback optimization module, with each module forming a closed-loop management system across the entire chain. The data acquisition module is used to acquire device performance monitoring data and user behavior data, and is equipped with an adaptive sampling mechanism that can dynamically adjust the sampling frequency according to the system load. The AI ​​decision-making module uses a lightweight deep learning model to fuse and analyze the collected multi-source data and output layout adjustment strategies. The layout adjustment module receives strategies from the AI ​​decision-making module and executes dynamic adjustments to interface elements, supporting partial refresh and overall reconstruction; The feedback optimization module collects performance metrics and user behavior data after layout adjustments and feeds them back to the AI ​​decision-making module for model parameter updates.

[0099] Optionally, the device performance monitoring data acquired by the data acquisition module includes CPU load, GPU load, and memory usage, while user behavior data includes click heatmaps, dwell time, and operation paths. The data acquisition method is based on Performance API or Web Vitals to ensure the real-time performance and accuracy of the data.

[0100] Optionally, the lightweight deep learning model in the AI ​​decision-making module is an optimized LSTM model. This model has two core capabilities: first, it determines whether UI elements need to be simplified based on the device performance status; second, it predicts the user's next operation based on user behavior and adjusts the layout priority of high-frequency areas. Moreover, the model adopts an online learning mechanism, combining historical data and real-time feedback to continuously optimize prediction accuracy.

[0101] Optionally, the AI ​​decision-making module introduces a multi-dimensional real-time constraint model, which includes performance constraints, behavioral constraints, interface complexity constraints, and business priority constraints. The AI ​​model solves the optimal balance between different constraints in real time to generate a globally adaptable layout strategy.

[0102] Optionally, the layout adjustment module is configured with an adaptive UI complexity control algorithm. This algorithm achieves quantitative management of the global UI through complexity quantification indicators. It can finely adjust the following UI parameters according to the performance status: the number, duration, and frame rate of UI animations; and the accuracy and dynamic replacement strategy of image resources. Component granularity, including whether placeholders, skeleton screens, and lazy loading are enabled.

[0103] Optionally, the UI element simplification operations performed by the layout adjustment module include reducing animations, lowering resolution, and compressing image resources, thereby reducing system resource consumption and improving interface responsiveness and smoothness.

[0104] Optionally, the feedback optimization module collects performance metrics such as page load time and FPS through a closed-loop mechanism, and collects user behavior data such as click-through rate and dwell time. The collected data is used for online training and parameter optimization of AI models to achieve continuous evolution of layout strategies.

[0105] Optionally, the adaptive sampling mechanism of the data acquisition module is configured with a dynamic weighting algorithm, which can intelligently adjust the sampling period based on the current CPU utilization and memory usage indicators, balancing performance monitoring accuracy and system overhead.

[0106] The AI-based multidimensional constraint adaptive user interface layout engine shown in Example 6 is performing... The specific implementation process for adaptive user interface layout is as follows: 1. Deployment of the data acquisition module Device performance data is collected based on the Performance API, including metrics such as CPU load, GPU load, memory usage, page load time, and FPS. User behavior data is collected using the application's built-in behavior tracking tool, including product click heatmaps, page dwell time, and order placement paths. An adaptive sampling mechanism is also enabled, setting a CPU load threshold of 70%. When the load is below the threshold, the sampling frequency is set to 1 time / second; when the load is above the threshold, the sampling frequency automatically decreases to 1 time / 5 seconds, balancing monitoring accuracy and resource consumption.

[0107] 2. AI Decision-Making Module Training and Deployment A lightweight LSTM model is used as the core decision-making model. The model training dataset contains 100,000 device performance data points and 200,000 user behavior data points. After training, the model can make two core decisions: When the device's CPU load is detected to be consistently above 70%, a UI simplification strategy is triggered. When a user is detected to have clicked the "Special Offers" entry on the homepage three times consecutively, the entry will be moved to the high-frequency interaction area at the top of the homepage.

[0108] Simultaneously, a multi-dimensional real-time constraint model is configured, and business priority weights are set: during major promotional periods, the "promotional activities" component has a higher priority than ordinary product components. When user behavior prediction conflicts with activity guidance, priority is given to ensuring the display position and interaction smoothness of the activity component.

[0109] 3. Layout adjustment module execution: When the AI ​​decision-making module outputs a UI simplification instruction, the layout adjustment module performs the following operations through an adaptive UI complexity control algorithm: turning off the floating animation of product cards, reducing the resolution of product images from 2K to 1080P, and enabling lazy loading and skeleton screens for products not on the first screen; when outputting a high-frequency area adjustment instruction, the "Special Offers" entry is moved from the middle of the homepage to below the top navigation bar, achieving a partial interface refresh without the need for a complete page reconstruction.

[0110] 4. Feedback optimization module execution: The feedback module continuously collects adjusted performance and behavioral data: on low-end Android devices, page loading time has decreased from 5 seconds to 3.5 seconds, and FPS has increased from 25 frames to 40 frames; the average path length for users to click on "special offer items" has been shortened from 3 steps to 1 step, and the order conversion rate has increased by 18%. The module feeds the above data back to the AI ​​decision module to fine-tune the LSTM model online, optimizing the accuracy of behavioral prediction and the performance adaptation threshold.

[0111] According to an embodiment of this application, an AI-based multidimensional constraint adaptive user interface layout device is provided. Figure 2 The AI-based multidimensional constraint adaptive user interface layout device includes the following functional modules: The data acquisition module 200 is used to collect device performance data and user behavior data of the user terminal. Prediction module 202 is used to input the user behavior data into a pre-trained lightweight deep learning model to predict the user's next operation intention and high-frequency operation area. The solution module 204 is used to input the device performance monitoring data and the high-frequency operation area into a preset multidimensional real-time constraint model to solve for the optimal balance between different constraints, thereby obtaining the optimal user interface layout strategy that satisfies the preset system operation smoothness threshold; wherein, the multidimensional real-time constraint model includes at least: device performance constraints and user behavior constraints; The calculation module 206 is used to calculate the complexity quantification index of the interface elements according to the optimal user interface layout strategy. The adjustment module 208 is used to dynamically adjust the rendering parameters and layout structure of the user interface based on the complex quantification indicators of the interface elements; wherein, the rendering parameters include: the number, duration and frame rate of user interface animations, the accuracy and dynamic replacement strategy of image resources, and the component granularity.

[0112] Optionally, in one implementation of this embodiment, the apparatus further includes: The parameter update module is used to collect device performance monitoring data and user behavior data of the user terminal after the user interface layout is adjusted, and to update the parameters of the multidimensional real-time constrained deep learning model online through a closed-loop feedback mechanism.

[0113] Optionally, in one implementation of this embodiment, the acquisition module is specifically used for: Determine whether to enable the adaptive sampling mechanism; If enabled, the sampling frequency is determined based on the system load, and the device performance monitoring data and user behavior data of the user terminal are collected according to the sampling frequency. If not enabled, the system will use a preset fixed sampling frequency to collect device performance monitoring data and user behavior data from user terminals.

[0114] Optionally, in one implementation of this embodiment, the multidimensional real-time constraint model further includes: interface complexity constraints and business priority constraints.

[0115] Optionally, in one implementation of this embodiment, the device performance monitoring data includes: CPU utilization, GPU utilization, and memory usage; the user behavior data includes: click heatmap, dwell time, and operation path.

[0116] Optionally, in one implementation of this embodiment, the solving module is specifically used for: The equipment performance monitoring data and the high-frequency operation area are input into a preset multi-dimensional real-time constraint model; When the device performance monitoring data indicates that the system load exceeds a preset threshold, and the user behavior constraints indicate the existence of a high-frequency operation area, the conflict resolution mechanism is activated. The conflict resolution mechanism includes: maintaining the rendering level and interaction response priority of the high-frequency operation area unchanged, and reducing the complexity of interface elements in the low-frequency operation area.

[0117] Optionally, in one implementation of this embodiment, the adjustment module is specifically used for: Based on the complex quantification indicators of the interface elements, adjust the number, duration, and frame rate of user interface animations, the loading accuracy of image resources, the dynamic replacement strategy of image resources, and the rendering granularity of components. The complexity score of different areas in the user interface is determined based on the complexity quantification index of the interface elements, and the rendering level is determined based on the complexity score of each area.

[0118] Optionally, in one implementation of this embodiment, when the acquisition module determines the sampling frequency based on the system load, it is specifically used for: Calculate the current system load coefficient based on the CPU utilization and memory usage in the device performance monitoring data; When the system load factor is lower than the first threshold, the data acquisition frequency is increased to obtain a high-precision user behavior path. When the system load factor is higher than the second threshold, the data acquisition frequency is reduced to reduce the occupation of equipment resources by monitoring activities.

[0119] Optionally, in one implementation of this embodiment, the calculation module is specifically used for: Based on the optimal user interface layout strategy, determine the complexity of DOM nodes, texture memory, animation rendering, interaction response, and network resources. Based on the DOM node complexity, texture memory complexity, animation rendering complexity, interaction response complexity, and network resource complexity, a quantification of the complexity of UI elements is calculated.

[0120] about Figure 2 For a more detailed description of the relevant operations, please refer to the description in the method embodiments above, which will not be repeated here.

[0121] One embodiment of the present invention also provides a product that implements various method embodiments of the present invention. This product can be a computer software product, or an electronic device or hardware product that has the computer software product installed or integrated. The electronic device can perform the methods shown in any of the above method embodiments to adjust the user interface layout.

[0122] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0123] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0124] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0125] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0126] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0127] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0128] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0129] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A multi-dimensional constraint adaptive user interface layout method based on AI, characterized in that, The method includes: Collect device performance data and user behavior data from user terminals; The user behavior data is input into a pre-trained lightweight deep learning model to predict the user's next operation intention and high-frequency operation area. The device performance monitoring data and the high-frequency operation area are input into a preset multidimensional real-time constraint model to solve for the optimal balance between different constraints, thereby obtaining the optimal user interface layout strategy that satisfies the preset system operation smoothness threshold; wherein, the multidimensional real-time constraint model includes at least: device performance constraints and user behavior constraints. Based on the optimal user interface layout strategy, calculate the complexity metrics of the interface elements; Based on the complex quantification metrics of the interface elements, the rendering parameters and layout structure of the user interface are dynamically adjusted; wherein, the rendering parameters include: the number, duration and frame rate of user interface animations, the accuracy and dynamic replacement strategy of image resources, and the component granularity.

2. The method according to claim 1, characterized in that, After the step of dynamically adjusting the rendering parameters and layout structure of the user interface based on the complexity metrics of the interface elements, the method further includes: After adjusting the user interface layout, device performance monitoring data and user behavior data of the user terminal are collected, and the parameters of the multidimensional real-time constrained deep learning model are updated online through a closed-loop feedback mechanism.

3. The method according to claim 1, characterized in that, The steps for collecting device performance monitoring data and user behavior data from user terminals include: Determine whether to enable the adaptive sampling mechanism; If enabled, the sampling frequency is determined based on the system load, and the device performance monitoring data and user behavior data of the user terminal are collected according to the sampling frequency. If not enabled, the system will use a preset fixed sampling frequency to collect device performance monitoring data and user behavior data from user terminals.

4. The method according to claim 1, characterized in that, The multidimensional real-time constraint model also includes: interface complexity constraints and business priority constraints.

5. The method according to claim 1, characterized in that, in, The device performance monitoring data includes: CPU utilization, GPU utilization, and memory usage; the user behavior data includes: click heatmap, dwell time, and operation path.

6. The method according to claim 1, characterized in that, The steps of inputting the device performance monitoring data and the high-frequency operation area into a preset multi-dimensional real-time constraint model to solve for the optimal balance solution between different constraints, and obtaining the optimal user interface layout strategy that satisfies the preset system operation smoothness threshold, include: The equipment performance monitoring data and the high-frequency operation area are input into a preset multi-dimensional real-time constraint model; When the device performance monitoring data indicates that the system load exceeds a preset threshold, and the user behavior constraints indicate the existence of a high-frequency operation area, the conflict resolution mechanism is activated. The conflict resolution mechanism includes: maintaining the rendering level and interaction response priority of the high-frequency operation area unchanged, and reducing the complexity of interface elements in the low-frequency operation area.

7. The method according to claim 1, characterized in that, The steps for dynamically adjusting the rendering parameters and layout structure of the user interface based on the complexity metrics of the interface elements include: Based on the complex quantification indicators of the interface elements, adjust the number, duration, and frame rate of user interface animations, the loading accuracy of image resources, the dynamic replacement strategy of image resources, and the rendering granularity of components. The complexity score of different areas in the user interface is determined based on the complexity quantification index of the interface elements, and the rendering level is determined based on the complexity score of each area.

8. The method according to claim 3, characterized in that, The steps for determining the sampling frequency based on system load include: Calculate the current system load coefficient based on the CPU utilization and memory usage in the device performance monitoring data; When the system load factor is lower than the first threshold, the data acquisition frequency is increased to obtain a high-precision user behavior path. When the system load factor is higher than the second threshold, the data acquisition frequency is reduced to reduce the occupation of equipment resources by monitoring activities.

9. The method according to claim 1, characterized in that, The steps for calculating the complexity metrics of interface elements based on the optimal user interface layout strategy include: Based on the optimal user interface layout strategy, determine the complexity of DOM nodes, texture memory, animation rendering, interaction response, and network resources. Based on the DOM node complexity, texture memory complexity, animation rendering complexity, interaction response complexity, and network resource complexity, a quantification of the complexity of UI elements is calculated.

10. A multi-dimensional constraint adaptive user interface layout device based on AI, characterized in that, The device includes: The data acquisition module is used to collect device performance data and user behavior data from user terminals. The prediction module is used to input the user behavior data into a pre-trained lightweight deep learning model to predict the user's next operation intention and high-frequency operation area. The solution module is used to input the device performance monitoring data and the high-frequency operation area into a preset multi-dimensional real-time constraint model to solve for the optimal balance between different constraints, thereby obtaining the optimal user interface layout strategy that satisfies the preset system operation smoothness threshold; wherein, the multi-dimensional real-time constraint model includes at least: device performance constraints and user behavior constraints; The calculation module is used to calculate the complexity metrics of the interface elements according to the optimal user interface layout strategy. The adjustment module is used to dynamically adjust the rendering parameters and layout structure of the user interface based on the complex quantification indicators of the interface elements; wherein, the rendering parameters include: the number, duration and frame rate of user interface animations, the accuracy and dynamic replacement strategy of image resources, and the component granularity.

11. An electronic device, characterized in that, The electronic device uses the method described in any one of claims 1-9, or has an AI-based multidimensional constraint adaptive user interface layout device as described in claim 10, to perform user interface layout.

Citation Information

Patent Citations

  • AI-assisted UX design evaluation

    US10929110B2

  • Dynamically determing appropriate computer user interfaces

    US20030046401A1