A webpage self-adaptive method and system based on multi-modal emotion perception

Abstract

The application relates to the field of browser usability auxiliary technology in artificial intelligence, and discloses a webpage self-adaptive method and system based on multi-modal emotion perception. The method collects multi-source initial signals of a user, performs synchronization and pretreatment, carries out multi-modal fusion reasoning in a local edge model, obtains an emotion vector and a confidence degree, selects a strategy package S from a strategy library in combination with page context classification and individualized threshold value, executes webpage rendering adjustment on DOM, CSS and ARIA attributes through a browser content script, dynamically changes information density, readability of layout, dynamic effect and interaction complexity, records user feedback, and adopts online strategy updating with constraints on the local side; adaptive degradation is executed when resources are limited; only intermediate features are saved under privacy and permission management, and original audio and video are not externally transmitted. The application realizes real-time closed loop and individualized adaptation of emotion-strategy-page, and significantly improves the online experience of special groups such as people with attention disorders and people with anxiety.

Description

Technical Field

[0001] This application relates to the field of browser usability assistance technology in artificial intelligence, and in particular to a webpage adaptive method and system based on multimodal emotion perception. Background Technology

[0002] Most existing browsers offer fixed reading modes, night modes, or manual accessibility settings, lacking the ability to adapt in real time based on users' psychological and physiological states. While some studies can identify emotions based on camera, microphone, or keyboard and mouse behavior, they have failed to establish a stable closed loop of "emotion → strategy → page rendering," and most rely on cloud inference, which raises privacy and latency issues, and cannot effectively adapt to users' psychological and physiological states in real time. Summary of the Invention

[0003] To address the aforementioned technical issues, this application provides a webpage adaptive method and system based on multimodal emotion perception, which achieves real-time closed-loop and individualized adaptation of emotion → strategy → page, significantly improving the online experience for special groups such as those with attention deficit disorder and anxiety.

[0004] Firstly, this application provides a webpage adaptive method based on multimodal emotion perception, which adopts the following technical solution:

[0005] A webpage adaptation method based on multimodal emotion perception includes the following steps:

[0006] Signal acquisition involves collecting multiple initial signals from the user while they are browsing web pages in a browser. These initial signals include behavioral signals and initial physiological or facial signals.

[0007] Synchronization and preprocessing: Using behavioral signals as the time axis, physiological or facial expression signals are time-aligned within a sliding window to obtain multi-source aligned signals. Then, the multi-source aligned signals are preprocessed to obtain multi-source preprocessed signals.

[0008] Multimodal fusion inference extracts and fuses features from multi-source preprocessed signals to estimate the sentiment vector e and confidence level c.

[0009] Gating and steady-state control: When the confidence level c is below the threshold θ or the magnitude of the emotion change in adjacent time windows is less than δ and does not exceed the cooling-off time T. c When necessary, suppress the switching of strategies;

[0010] Page context classification: Input the page context into the page context classifier and output the page context classification result;

[0011] The strategy selection is based on the emotion vector e and the page context classification results, and the strategy package S that matches the individualization threshold and has the highest priority is retrieved from the strategy library;

[0012] Webpage rendering adjustments are made by generating corresponding browser content scripts based on strategy package S. These browser content scripts are then used to adjust the DOM, CSS, and ARIA properties of the target webpage in real time to change information density, layout readability, dynamic effects, and interaction complexity.

[0013] Feedback and updates: Record user interaction feedback r after the execution of policy package S, and use a constrained online policy weight update algorithm to fine-tune and update the policy library locally on the server.

[0014] Preferably, the behavioral signals include mouse trajectory, scrolling rhythm, typing rhythm, and pause-review mode;

[0015] The physiological or facial expression signals include facial expression frame features, acoustic prosodic statistical features, and limb movement features.

[0016] Preferably, the mouse trajectory includes mouse movement speed, mouse movement acceleration, density of mouse trajectory turning points, and mouse hover duration;

[0017] The rolling rhythm includes the average rolling speed and the beat variation coefficient;

[0018] The typing rhythm includes keystroke rate, mean and variance of interval, and key duration distribution;

[0019] The acoustic prosody includes speech MFCC features, speech rate, and dynamic changes in volume.

[0020] Preferably, the preprocessing of multi-source signals includes denoising, standardization, and missing information completion.

[0021] Preferably, the emotion vector e includes an attention index, anxiety level, fatigue index, and excitement level.

[0022] Preferably, the page context input includes page structure features, text semantic summaries, and real-time user interaction features;

[0023] The page context classification results include task-oriented, reading-oriented, and entertainment-oriented.

[0024] Preferably, real-time adjustments to the DOM, CSS, and ARIA properties of the target webpage include: adjusting font size, line spacing, and contrast; switching between low blue light and high contrast themes; disabling or weakening CSS animations and non-critical animations driven by requestAnimationFrame; expanding the hit area of ​​interactive controls to at least 44 pixels; enabling reading mode and segmented navigation for long articles; and disabling autoplay and floating recommendation bubbles for video pages.

[0025] Preferably, the strategy library uses a JSON structure to define priorities, conflict relationships, and upper limits for change magnitude, and applies the strategy in a step-by-step manner when the cumulative layout offset index exceeds the threshold.

[0026] Secondly, this application provides a webpage adaptive system based on multimodal emotion perception, which adopts the following technical solution:

[0027] A webpage adaptive system based on multimodal emotion perception includes:

[0028] The signal acquisition module is used for signal acquisition. When a user browses a webpage in a browser, it acquires the user's multi-source initial signals, which include behavioral signals and initial physiological or facial signals.

[0029] The synchronization and preprocessing module is used for synchronization and preprocessing. It performs time alignment of physiological or facial expression signals within a sliding window with behavioral signals as the time axis to obtain multi-source aligned signals. Then, it performs preprocessing on the multi-source aligned signals to obtain multi-source preprocessed signals.

[0030] The multimodal fusion inference module is used for multimodal fusion inference, which extracts and fuses features from multi-source preprocessed signals, and estimates the sentiment vector e and confidence level c.

[0031] The gating and steady-state control module is used for gating and steady-state control. This is implemented when the confidence level c is below a threshold θ or the magnitude of the emotion change in adjacent time windows is less than δ and does not exceed the cooling-off time T. c When necessary, suppress the switching of strategies;

[0032] The page context classification module is used for page context classification. It takes the page context as input to the page context classifier and outputs the page context classification results.

[0033] The strategy selection module selects strategies based on the emotion vector e and the page context classification results, and retrieves the strategy package S that matches the individualized threshold and has the highest priority from the strategy library.

[0034] The webpage rendering adjustment module is used for webpage rendering adjustments. Based on the strategy package S, it generates corresponding browser content scripts and uses these scripts to adjust the DOM, CSS, and ARIA properties of the target webpage in real time to change information density, layout readability, dynamic effects, and interaction complexity.

[0035] The feedback and update module is used for feedback and updates. It records user interaction feedback r after the execution of the policy package S and uses a constrained online policy weight update algorithm to fine-tune and update the policy library locally on the server.

[0036] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0037] A computer-readable storage medium includes: a computer program that can be loaded by a processor and executed to implement a webpage adaptation method based on multimodal emotion perception as shown in any possible implementation of the first aspect.

[0038] In summary, this application includes the following beneficial technical effects:

[0039] 1. This application achieves real-time closed-loop and personalized adaptation of emotion → strategy → page, significantly improving the online experience for special groups such as those with attention deficit disorder and anxiety.

[0040] 2. This application introduces page semantic classification to enable the same emotion to adopt differentiated strategies in different task contexts.

[0041] 3. This application employs local inference and feature desensitization, which allows for smooth degradation when resources are limited, thus balancing real-time performance and security.

[0042] 4. This application uses DOM / CSS / ARIA and a reading extractor to finely control readability, dynamic effects, and interaction complexity, and sets layout stability limits.

[0043] 5. This application collects user behavioral signals and initial physiological or facial signals, and selects appropriate strategy packages based on the user's real-time emotions. This allows for dynamic adjustments to the layout and typography of web pages to adapt to the user's real-time emotions, improve user experience, and provide significant relief for individuals with attention deficit disorder, anxiety, or other special needs. Attached Figure Description

[0044] Figure 1 This is a system diagram of the webpage adaptive method in the embodiments of this application.

[0045] Figure 2 This is a block diagram of the webpage adaptive method in the embodiments of this application.

[0046] Figure 3 This is a schematic diagram of time alignment and hysteresis gating in an embodiment of this application. Detailed Implementation

[0047] The present application will be further described in detail below with reference to the accompanying drawings.

[0048] This application discloses a webpage adaptive method based on multimodal emotion perception.

[0049] Reference Figure 1 and Figure 2 This webpage adaptation method based on multimodal emotion perception includes the following steps:

[0050] S1, Signal Acquisition;

[0051] When a user browses a webpage in a browser, the browser collects the user's multi-source initial signals, which include behavioral signals and initial physiological or facial signals.

[0052] The aforementioned signal acquisition can utilize user-installed, configured, and authorized cameras, microphones, and pre-installed devices such as built-in models, keyboards, and mice on the host computer.

[0053] Among them, behavioral signals include mouse trajectory, scrolling rhythm, typing rhythm, and hover-look-back pattern; mouse trajectory includes mouse movement speed, mouse movement acceleration, density of mouse trajectory turning points, and mouse hover duration; scrolling rhythm includes average scrolling speed and beat variation coefficient; typing rhythm includes keystroke rate, mean and variance of intervals, and keystroke duration distribution.

[0054] Physiological or facial expression signals include facial expression frame features, acoustic prosodic statistical features, and body movement features; acoustic prosodic features include speech MFCC features, speech rate, and dynamic changes in volume.

[0055] S2, Synchronization and Preprocessing;

[0056] Reference Figures 1 to 3 Using behavioral signals as the time axis, physiological or facial expression signals are time-aligned within a sliding window. Specifically, using the behavioral event sequence as the main axis and a window length of W seconds, dynamic time warping is performed on camera frame embedding and acoustic prosody to obtain multi-source aligned signals. Then, the multi-source aligned signals are preprocessed to obtain multi-source preprocessed signals.

[0057] Preprocessing of multi-source signals includes denoising, standardization, and missing data completion.

[0058] S3, Multimodal Fusion Reasoning;

[0059] Feature extraction and fusion are performed on multi-source preprocessed signals to estimate the sentiment vector e and confidence level c;

[0060] The emotion vector e includes attention index, anxiety level, fatigue index, and excitement level.

[0061] Each component is a normalized or probability-based score between 0 and 1:

[0062] e attention Attention index: This indicates the user's level of focus.

[0063] e anxiety Anxiety level reflects the degree of tension / unease.

[0064] e fatigue Fatigue index: a quantitative measure of fatigue or decreased energy.

[0065] e arousal Excitement level reflects the state of activation or activity.

[0066] Each of the above components is obtained by weighted combination of relevant features.

[0067] Specifically, all features of the multi-source preprocessed signal are normalized, including z-score or Min-Max normalization, to ensure that the input features are of the same magnitude.

[0068] By using weighted linear combination or deep fusion networks, such as MLP / Transformer, multimodal features are concatenated or weighted and then input into the fusion network to output a sentiment vector e.

[0069] The convergence process of the converged network is as follows:

[0070] ;

[0071] Among them, F fusion The feature to be fused is the emotion vector e, and concat is the fusion function f. mouse For the mouse trajectory, ω mouse f mouse The weight, f scroll For mouse scrolling rhythm, ω scroll f scroll The weights for features such as typing rhythm, pause-and-look-back pattern, facial expression frame features, acoustic prosodic statistical features, and body movement features are substituted into the above formula in the same way.

[0072] S4, Gating and Steady-State Control;

[0073] When the confidence level c is lower than the threshold θ, or the magnitude of the mood change in adjacent time windows is less than δ and does not exceed the cooling-off time T. c When necessary, suppress the switching of strategies;

[0074] This embodiment employs a dual-threshold hysteresis (θ) on >θ off ) and cooling timer T c When T c <θ on or |ee prev |<δ and not past T c Maintain the strategy and avoid critical jitter.

[0075] S5. Page Context Classification;

[0076] Input the page context into the page context classifier and output the page context classification result;

[0077] Page contextual input includes page structure features, text semantic summarization, and real-time user interaction features;

[0078] By using structural and semantic signals, the page context classification results can be divided into task-oriented, reading-oriented, or entertainment-oriented categories, so as to adopt different strategies under the same mood.

[0079] The semantic signals include text length, media density, form / button density, and scrolling mode.

[0080] Triggering conditions: Anxiety confidence ≥ 0.7 and c ≥ θ on ctx = reading type;

[0081] Implementation strategy: Reduce information density (hide secondary sidebar and recommendation area), line spacing 1.8, font size +2px, disable animation, disable autoplay, display single button (cooldown 10 minutes), CLS limit 0.1.

[0082] Triggering conditions: Distraction score ≥ 0.6 and c ≥ θ on ctx = task type;

[0083] Implementation strategy: Focus mode (main task path only), de-emphasize non-critical buttons, provide step-by-step guidance, change site colors to cool tones, and allow a 5-minute cooldown period.

[0084] S6, Strategy Selection;

[0085] Based on the emotion vector e and the page context classification results, retrieve the policy package S that matches the individualization threshold and has the highest priority from the policy library;

[0086] The strategy library uses a JSON structure to define priorities, conflict relationships, and upper limits for change magnitude, and applies the strategy in a step-by-step manner when the cumulative layout offset metric exceeds the threshold.

[0087] The strategy package parameters include: font size, line spacing, character spacing, contrast, animation intensity, information density, control hit area, reading mode, segmented navigation, video autoplay and floating notification, etc.; define priority, amplitude limit and mutual exclusion relationship; apply in stages when the cumulative layout offset (CLS) exceeds the threshold.

[0088] When in use, resource load can be sensed from browser performance API, power consumption and temperature readings; when resource load exceeds the threshold, camera and audio sampling are reduced, high-cost subnets are disabled, and behavior is degraded to a single modality to maintain real-time performance.

[0089] Specifically, this implementation can also set up resource-constrained degradation:

[0090] Triggering conditions: CPU > 80% or battery < 15% or temperature alarm;

[0091] Implementation strategy: Disable face and acoustic subnets; single-modality behavior; halve the sampling rate; only toggle layout / animation on / off, without making significant layout changes.

[0092] S7, Webpage rendering adjustments;

[0093] Based on strategy package S, corresponding browser content scripts are generated. The browser content scripts are used to adjust the DOM, CSS and ARIA properties of the target webpage in real time to change information density, typographic readability, dynamic effects and interactive complexity.

[0094] Real-time adjustments to the DOM, CSS, and ARIA properties of the target webpage include: adjusting font size, line spacing, and contrast; switching between low blue light and high contrast themes; disabling or weakening CSS animations and non-critical animations driven by requestAnimationFrame; expanding the hit area of ​​interactive controls to at least 44 pixels; enabling reading mode and segmented navigation for long articles; and disabling autoplay and floating recommendation bubbles for video pages.

[0095] Specifically, the following operations can be performed on a webpage through browser content scripts:

[0096] Layout: Font size + Δ, line spacing 1.6~1.9, character spacing 0.02em, contrast conforms to WCAG 2.2 AA;

[0097] Dynamic: Apply prefers-reduced-motion to disable non-critical animations and autoplay;

[0098] Interaction: The control's hit area is ≥44px, it is visible with focus, and it is reachable by the keyboard;

[0099] Information density: Hide the secondary sidebar, reduce the recommendation feed, enable reading mode, segmented navigation and reading progress.

[0100] Disable autoplay and floating bubble notifications on video pages to reduce interruptions.

[0101] S8, Feedback and Updates;

[0102] Record user interaction feedback r after the execution of policy package S, and use a constrained online policy weight update algorithm to fine-tune and update the policy library locally on the server.

[0103] Feedback and updates employ an online bandit algorithm with an intervention load budget. When interruptive interventions exceed a threshold within a unit of time or users continuously refuse interventions, the algorithm automatically degrades to a passive optimization level.

[0104] The webpage adaptation method based on multimodal emotion perception disclosed in this embodiment is that all the raw data, subsequent calculation data, strategy library data, and generated browser content scripts involved are stored locally. It is mainly performed through a local edge computing module. Furthermore, without the user setting to upload to a cloud platform, all the user-related privacy data applied or collected do not need to be uploaded to the cloud platform, but can be implemented entirely locally, thus protecting user privacy.

[0105] This application also discloses a webpage adaptive system based on multimodal emotion perception, which is applied to the webpage adaptive method based on multimodal emotion perception disclosed in the above embodiments. Specifically, the webpage adaptive system based on multimodal emotion perception includes the following modules or units:

[0106] The signal acquisition module is used for signal acquisition. When a user browses a webpage in a browser, it acquires the user's multi-source initial signals, which include behavioral signals and initial physiological or facial signals.

[0107] The signal acquisition module includes:

[0108] The camera is used to capture the user's facial expression frames and body movement features;

[0109] The microphone is used to collect the user's acoustic prosodic statistical features, including speech MFCC features, speech rate, and dynamic changes in volume.

[0110] The mouse is used to collect the user's mouse trajectory, scrolling rhythm, and hover-back mode.

[0111] The keyboard is used to capture the user's typing rhythm;

[0112] The synchronization and preprocessing module is used for synchronization and preprocessing. It performs time alignment of physiological or facial expression signals within a sliding window with behavioral signals as the time axis to obtain multi-source aligned signals. Then, it performs preprocessing on the multi-source aligned signals to obtain multi-source preprocessed signals.

[0113] The synchronization unit is used to time-align physiological or facial signals within a sliding window with behavioral signals as the time axis, to obtain multi-source aligned signals.

[0114] The preprocessing unit is used to preprocess the multi-source aligned signal to obtain the multi-source preprocessed signal;

[0115] The preprocessing includes denoising, standardization, and missing data completion.

[0116] The multimodal fusion inference module is used for multimodal fusion inference, which extracts and fuses features from multi-source preprocessed signals, and estimates the sentiment vector e and confidence level c.

[0117] The gating and steady-state control module is used for gating and steady-state control. This is implemented when the confidence level c is below a threshold θ or the magnitude of the emotion change in adjacent time windows is less than δ and does not exceed the cooling-off time T. c When necessary, suppress the switching of strategies;

[0118] The page context classification module is used for page context classification. It takes the page context as input to the page context classifier and outputs the page context classification results.

[0119] The strategy selection module selects strategies based on the emotion vector e and the page context classification results, and retrieves the strategy package S that matches the individualized threshold and has the highest priority from the strategy library.

[0120] The webpage rendering adjustment module is used for webpage rendering adjustments. Based on the strategy package S, it generates corresponding browser content scripts and uses these scripts to adjust the DOM, CSS, and ARIA properties of the target webpage in real time to change information density, layout readability, dynamic effects, and interaction complexity.

[0121] The feedback and update module is used for feedback and updates. It records user interaction feedback r after the execution of the policy package S and uses a constrained online policy weight update algorithm to fine-tune and update the policy library locally on the server.

[0122] This application also provides an electronic device, which includes a processor and a memory. The processor and memory are connected, for example, via a bus. Optionally, the electronic device may also include a transceiver. It should be noted that in practical applications, the transceiver is not limited to one unit, and the structure of this electronic device does not constitute a limitation on the embodiments of this application.

[0123] The processor can be a central processing unit (CPU), a general-purpose processor, a data signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0124] A bus can include a pathway for transmitting information between the aforementioned components. The bus can be a peripheral component interconnect standard PCI bus or an extended industry standard structure EISA bus, etc. Buses can be categorized as address buses, data buses, control buses, etc.

[0125] The memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a read-only optical disc (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0126] The memory stores the application code that executes the scheme of this application, and the processor controls its execution. The processor executes the application code stored in the memory to implement the content shown in an embodiment of a webpage adaptation method based on multimodal emotion perception.

[0127] Electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Servers can also be included.

[0128] This application also provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute corresponding content in an embodiment of a webpage adaptation method based on multimodal emotion perception.

[0129] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

Claims

1. A webpage adaptive method based on multimodal emotion perception, characterized in that, Includes the following steps: Signal acquisition involves collecting multiple initial signals from the user while they are browsing web pages in a browser. These initial signals include behavioral signals and initial physiological or facial signals. Synchronization and preprocessing: Using behavioral signals as the time axis, physiological or facial expression signals are time-aligned within a sliding window to obtain multi-source aligned signals. Then, the multi-source aligned signals are preprocessed to obtain multi-source preprocessed signals. Multimodal fusion inference extracts and fuses features from multi-source preprocessed signals to estimate the sentiment vector e and confidence level c. Gating and steady-state control: When the confidence level c is below the threshold θ or the magnitude of the emotion change in adjacent time windows is less than δ and does not exceed the cooling-off time T. c When necessary, suppress the switching of strategies; Page context classification: Input the page context into the page context classifier and output the page context classification result; The strategy selection is based on the emotion vector e and the page context classification results, and the strategy package S that matches the individualization threshold and has the highest priority is retrieved from the strategy library; Webpage rendering adjustments are made by generating corresponding browser content scripts based on strategy package S. These browser content scripts are then used to adjust the DOM, CSS, and ARIA properties of the target webpage in real time to change information density, layout readability, dynamic effects, and interaction complexity. Feedback and updates: Record user interaction feedback r after the execution of policy package S, and use a constrained online policy weight update algorithm to fine-tune and update the policy library locally on the server.

2. The webpage adaptive method based on multimodal emotion perception according to claim 1, characterized in that, The behavioral signals include mouse trajectory, scrolling rhythm, typing rhythm, and pause-review mode; The physiological or facial expression signals include facial expression frame features, acoustic prosodic statistical features, and limb movement features.

3. The webpage adaptive method based on multimodal emotion perception according to claim 2, characterized in that, The mouse trajectory includes mouse movement speed, mouse movement acceleration, density of mouse trajectory turning points, and mouse hover duration; The rolling rhythm includes the average rolling speed and the beat variation coefficient; The typing rhythm includes keystroke rate, mean and variance of interval, and key duration distribution; The acoustic prosody includes speech MFCC features, speech rate, and dynamic changes in volume.

4. The webpage adaptive method based on multimodal emotion perception according to claim 1, characterized in that, Preprocessing of multi-source signals includes denoising, standardization, and missing data completion.

5. The webpage adaptive method based on multimodal emotion perception according to claim 1, characterized in that, The emotion vector e includes attention index, anxiety level, fatigue index, and excitement level.

6. The webpage adaptive method based on multimodal emotion perception according to claim 1, characterized in that, The page context input includes page structure features, text semantic summarization, and real-time user interaction features; The page context classification results include task-oriented, reading-oriented, and entertainment-oriented.

7. The webpage adaptive method based on multimodal emotion perception according to claim 1, characterized in that, Real-time adjustments to the DOM, CSS, and ARIA properties of the target webpage include: adjusting font size, line spacing, and contrast; switching between low blue light and high contrast themes; disabling or weakening CSS animations and non-critical animations driven by requestAnimationFrame; expanding the hit area of ​​interactive controls to at least 44 pixels; enabling reading mode and segmented navigation for long articles; and disabling autoplay and floating recommendation bubbles for video pages.

8. The webpage adaptive method based on multimodal emotion perception according to claim 1, characterized in that, The strategy library uses a JSON structure to define priorities, conflict relationships, and upper limits for change magnitude, and applies the strategy in a step-by-step manner when the cumulative layout offset index exceeds the threshold.

9. A webpage adaptive system based on multimodal emotion perception, characterized in that, include: The signal acquisition module is used for signal acquisition. When a user browses a webpage in a browser, it acquires the user's multi-source initial signals, which include behavioral signals and initial physiological or facial signals. The synchronization and preprocessing module is used for synchronization and preprocessing. It performs time alignment of physiological or facial expression signals within a sliding window with behavioral signals as the time axis to obtain multi-source aligned signals. Then, it performs preprocessing on the multi-source aligned signals to obtain multi-source preprocessed signals. The multimodal fusion inference module is used for multimodal fusion inference, which extracts and fuses features from multi-source preprocessed signals, and estimates the sentiment vector e and confidence level c. The gating and steady-state control module is used for gating and steady-state control. This is implemented when the confidence level c is below a threshold θ or the magnitude of the emotion change in adjacent time windows is less than δ and does not exceed the cooling-off time T. c When necessary, suppress the switching of strategies; The page context classification module is used for page context classification. It takes the page context as input to the page context classifier and outputs the page context classification results. The strategy selection module selects strategies based on the emotion vector e and the page context classification results, and retrieves the strategy package S that matches the individualized threshold and has the highest priority from the strategy library. The webpage rendering adjustment module is used for webpage rendering adjustments. Based on the strategy package S, it generates corresponding browser content scripts and uses these scripts to adjust the DOM, CSS, and ARIA properties of the target webpage in real time to change information density, layout readability, dynamic effects, and interaction complexity. The feedback and update module is used for feedback and updates. It records user interaction feedback r after the execution of the policy package S and uses a constrained online policy weight update algorithm to fine-tune and update the policy library locally on the server.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements a webpage adaptive method based on multimodal emotion perception as described in any one of claims 1 to 8.

Images