Objective evaluation method for user experience emotion fused with multi-modal physiological signals

Abstract

The invention discloses a multi-modal physiological signal fused user experience emotion objective evaluation method, and relates to the technical field of human-computer interaction and user experience evaluation, and the method comprises the following steps: S1, constructing a VR emotion induction experiment environment, synchronously collecting an electroencephalogram signal, an electrocardiosignal and a skin electric signal based on an interaction task of a user and a product, and carrying out the interaction of the user and the product; and recording subjective evaluation data synchronized with the interaction event. According to the method, a complete standardized experiment system is formed by constructing a VR emotion inducing environment and synchronously collecting multi-modal physiological signals and combining standardized target emotion type definition, multi-dimensional task profile extraction and standardized signal collection parameter setting; therefore, the standardization and repeatability of the whole process from experimental design to data acquisition are ensured, the problems of large individual difference, inconsistent experimental conditions and difficulty in result reproduction in the traditional subjective evaluation method are effectively solved, and a solid foundation is laid for obtaining high-quality comparable physiological data.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and user experience evaluation technology, specifically to an objective evaluation method for user experience emotion that integrates multimodal physiological signals. Background Technology

[0002] In today's digital age, user-centered design has become a key factor in the success of products and services. The quality of user experience directly affects user satisfaction, product reputation, and market competitiveness. Emotion, as a core element influencing user experience, plays a crucial role in purchasing decisions, user stickiness, and product improvement. Currently, user experience evaluation mainly relies on subjective evaluation methods, including user interviews, focus groups, and standardized questionnaires. Although these methods can directly obtain user feedback, they have obvious limitations. First, they are easily affected by individual subjective bias, memory distortion, and social expectation effects. Second, they are difficult to quantify and standardize. Most importantly, they cannot capture the dynamic and continuous emotional changes of users during the interaction process.

[0003] The shortcomings of existing objective evaluation methods for user experience emotion are: 1. Patent document CN118690964A discloses a method and system for validating the subjective and objective evaluation data of automotive cockpit interaction experience. The document states that the method and system include: constructing a multi-dimensional subjective and objective joint evaluation system; collecting subjective and objective evaluation data from subjects based on the evaluation system; calculating the correlation between subjective and objective data under different dimensions to obtain correlation calculation results; obtaining a strongly correlated dataset based on the correlation calculation results; and performing subjective and objective index clustering analysis on the strongly correlated dataset to obtain a clustered dataset. This invention is applicable to different cockpit interaction objects, enabling effective verification of subjective and objective evaluation data of interaction experience, and providing support for researchers and industry workers to obtain objective and truthful product user evaluations in practical applications. However, the document lacks unified experimental standards and standardized procedures. Different studies employ significantly different experimental environments, task designs, and data collection parameters, making it difficult to directly compare and reproduce experimental results. These results are highly susceptible to individual differences, and the experimental conditions are not strictly controlled, failing to guarantee the consistency and reliability of data collection. This seriously hinders the accumulation and application of research results. Summary of the Invention

[0004] The purpose of this invention is to provide an objective evaluation method for user experience emotion that integrates multimodal physiological signals, so as to solve the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an objective evaluation method for user experience emotion that integrates multimodal physiological signals, comprising the following steps: S1. Construct a VR emotion-inducing experimental environment. Based on the user's interaction with the product, simultaneously collect EEG signals, ECG signals, and skin conductance signals, and record subjective evaluation data synchronized with the interaction events. S2. Extract the asymmetry index of the prefrontal alpha band from the EEG signal, extract the standard deviation and root mean square difference of the heart rate variability time domain index, the low-frequency to high-frequency ratio and sample entropy from the ECG signal, and extract the skin conductance level and non-specific skin conductance response frequency from the EEG signal. S3. Principal component analysis is used to reduce the dimensionality of the extracted features and fuse them, retaining the principal components with a cumulative variance contribution rate of ≥90%, and generating a low-dimensional fused feature vector. S4. Input the feature vector into a single-layer LSTM network in time sequence. The network has 96 hidden units and three fully connected layers with 128-64-32 neurons. Introduce an attention mechanism to calculate the weights of each time step and output the emotion state classification result. S5. Generate a dynamic emotional state curve based on the classification results, locate key experience moments through the attention weight distribution map, and associate them with the timestamps of interactive events.

[0006] Preferably, the VR emotion-inducing experimental environment in step S1 is constructed by including: The target emotional type is categorized into three types: positive, neutral, and negative. Typical task profiles are extracted from the dimensions of product appearance design approval, smoothness of use process, physical comfort, sense of technology and intelligence; The experimental procedure included baseline resting state recording, experience of multiple task segments, and subjective evaluation after each segment.

[0007] Preferably, the acquisition of multimodal physiological signals in step S1 specifically involves: acquiring EEG signals using a 32-lead EEG instrument, focusing on prefrontal electrode sites FP1, FP2, F3, F4, F7, and F8, with a sampling frequency of 512 Hz; The electrocardiogram signal was acquired using a three-lead configuration with a sampling frequency of 500Hz. Electrodermal signals were acquired using a multichannel physiological recorder at a sampling frequency of 500 Hz.

[0008] Preferably, the preprocessing of the EEG signal in step S2 includes rereference, downsampling, filtering, and removal of electrooculogram artifacts using independent component analysis. The preprocessing of the electrocardiogram signal uses a Butterworth low-pass filter, an FIR band-notch filter, and a zero-phase-shift linear filter to remove electromyographic artifacts, power line interference, and baseline drift, respectively. The preprocessing of the electrodermal signal includes 0.3Hz low-pass filtering and Z-score normalization.

[0009] Preferably, the prefrontal alpha band asymmetry index in step S2 is calculated by taking the alpha band signal power of the three pairs of co-originating electrodes FP1 and FP2, F3 and F4, and F7 and F8 in the prefrontal lobe, using the following formula:

[0010] in, and These represent the signal power of the right and left electrodes in the alpha band, respectively.

[0011] Preferably, in step S4, the LSTM network training uses a decreasing dropout rate, with the dropout rates for the three fully connected layers being 0.4, 0.3, and 0.2, respectively. The training optimizer is Adam, the initial learning rate is 0.001, and the training cycle is dynamically adjusted in conjunction with an early stopping mechanism.

[0012] Preferably, the accuracy of the emotion state classification result in step S4 is not less than 85%.

[0013] Preferably, the location of key experience moments in step S5 is specifically as follows: in the attention weight distribution map, the time period with a weight value higher than a preset threshold is identified as the emotional peak or trough moment. This time period is matched with the timestamp of the interaction event recorded in step S1, thereby locating the specific interaction event that triggers positive or negative emotions.

[0014] Preferably, in step S4, the LSTM network is trained using 5-fold cross-validation.

[0015] Preferably, it also includes an application phase: deploying the trained model on a user experience testing platform to perform real-time emotional evaluation of the new product interaction process and output a design optimization report that includes the location of key experience moments.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a VR emotion-inducing environment and simultaneously collects multimodal physiological signals. Combined with standardized target emotion type definitions, multi-dimensional task profile extraction, and standardized signal acquisition parameter settings, it forms a complete standardized experimental system. This ensures the standardization and repeatability of the entire process from experimental design to data acquisition, effectively solving the problems of large individual differences, inconsistent experimental conditions, and difficulty in reproducing results in traditional subjective evaluation methods. This lays a solid foundation for obtaining high-quality and comparable physiological data. 2. This invention significantly improves the quality of the original signal and the effectiveness of the feature parameters by combining specialized preprocessing methods for different physiological signals and feature extraction calculation formulas based on neuroscience principles on the basis of multimodal signal acquisition. Furthermore, through professional signal denoising, artifact elimination, and feature quantization, the original physiological signal is transformed into effective feature indicators that can accurately reflect the emotional state, providing high-quality input for subsequent emotion recognition and ensuring the scientificity and reliability of the entire technical solution from data acquisition to feature extraction. 3. This invention constructs a high-performance, high-reliability emotion recognition model by combining a regularization strategy with a decreasing dropout rate, clear performance metric requirements, and rigorous cross-validation methods on the basic LSTM-Attention architecture. This combination ensures that the model maintains high accuracy while possessing good generalization ability and stability through specific network parameter optimization, overfitting prevention measures, and statistical validation processes, effectively solving the technical problems of low accuracy and poor robustness of traditional emotion recognition methods. 4. This invention, by combining a key experience moment localization mechanism based on attention weights and a practical application deployment scheme within an overall methodological framework, forms a complete closed loop from technical implementation to industrial application. This combination can not only accurately identify emotional peaks and troughs in the user experience process through attention weight distribution, but also transform the analysis results into specific design optimization suggestions and realize real-time emotion evaluation functions, significantly enhancing the practical value and industrialization potential of the technology, and realizing the effective transformation of scientific research results into practical tools. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the evaluation process for the smart home control system of the present invention; Figure 3 This is a schematic diagram of the online education platform evaluation process of the present invention; Figure 4 This is a schematic diagram of the vehicle-mounted system evaluation process of the present invention. Detailed Implementation

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

[0019] Experimental environment setup First, a user experience emotion-inducing experimental platform was built. The experiment was conducted in a dedicated, soundproof laboratory room, which was electromagnetically shielded to minimize electromagnetic interference. The ambient temperature and humidity were maintained at constant levels of 22±1°C and 50±5%, respectively, background noise was controlled below 31dB, and background illumination was 2cd / m². 2 ; The experimental equipment is configured as follows: VR presentation device: PICO 4 Ultra VR headset, used to create an emotionally evoking environment; EEG acquisition equipment: TMSi SAGA 32-lead EEG analyzer; Multi-channel physiological recorder: BIOPAC MP160 system; Experimental control computer: Configured with the Unity engine for experimental program development; Synchronization device: All devices achieve strict time synchronization through the synchronization interface of the BIOPAC MP160 system, with a synchronization accuracy of ±1ms; Example 1: Please refer to Figure 1 and Figure 2 An objective evaluation method for user experience emotion that integrates multimodal physiological signals: Smart home control system user experience evaluation: 1. Experimental Design This embodiment selects a smart home control system as the evaluation object, focusing on examining the user's emotional experience in different interactive tasks; Target emotional type identified: Based on the two-dimensional emotional space model, three target emotional states are established: Positive emotions: high valence and high arousal (pleasure, excitement), valence score > 7, arousal score > 6; Negative emotions: low valence, high arousal (frustration, anxiety), valence score <3, arousal score >6; Neutral affect: moderate valence, low arousal (calm, natural), valence score 4-6, arousal score <4 Typical task profile extraction: Through desktop research, user interviews, and expert consultation, the following typical task profiles were extracted: Appearance design approval: system interface aesthetics, icon recognizability, and layout rationality; Smoothness and ease of use: Number of steps to complete the "Away Mode" setup and operation response time; Human comfort: fatigue after prolonged operation, and smoothness of interface switching; Technological sophistication and intelligence: accuracy of scene mode recommendations and recognition rate of voice control; Experimental Paradigm Design: The experimental procedure includes the following steps: Baseline recording: Baseline physiological signal data were collected during a 5-minute resting period; Task Experience: 18 task segments, each segment lasting 2 minutes; Subjective assessment: The SAM scale was scored on a 9-point scale immediately after each task segment; Recovery: Take a 2-minute break and play soothing music; 2. Subject Recruitment Twenty healthy participants were recruited, ten men and ten women, with an age range of 23.8 ± 2.4 years. All participants met the following criteria: Normal visual acuity or corrected visual acuity; No history of neurological, psychiatric, or cardiovascular diseases; A resting heart rate of 60-100 beats per minute is within the normal range. No history of VR motion sickness; Before the experiment, the emotional state was assessed as normal using the Short Form Mood Scale; 24 hours before the experiment, the subjects were required to maintain a regular routine and refrain from consuming stimulants such as alcohol and caffeine. On the day of the experiment, the subjects were required to wash their hair to ensure the quality of the EEG signal acquisition. 3. Data Collection Physiological signal acquisition configuration: EEG signals: Acquired using a 32-lead EEG instrument, focusing on electrode sites FP1, FP2, F3, F4, F7, and F8 in the prefrontal cortex region. The sampling frequency was set to 512Hz, AC acquisition mode, gain 1000 times, and bandpass filtering 0.05~100Hz. ECG signal: Three-lead configuration, disposable Ag-AgCl electrodes, sampling frequency 500Hz, bandpass filter 0.5~35Hz; Electrodermal signals: Acquired through the electrodermal module of the BIOPAC MP160 system, with a sampling frequency of 500Hz and a low-pass filter of 0.3Hz; All physiological signals were collected in strict synchronization with the VR task presentation, and timestamps of user interaction events were recorded simultaneously. 4. Signal Processing and Feature Extraction EEG signal processing: Preprocessing included re-reference (using bilateral mastoid processes as references), downsampling to 256 Hz, and bandpass filtering from 0.5 to 45 Hz. Independent component analysis was used to remove electrooculography artifacts. Feature extraction employs wavelet packet analysis to perform a 5-level decomposition of the three pairs of homologous electrodes in the prefrontal cortex. The asymmetry index is calculated based on the reconstructed Alpha band signal power.

[0020] The electrode pairs include FP1 and FP2, F3 and F4, and F7 and F8; ECG signal processing: Preprocessing employed a 0.5-40Hz bandpass filter. The Pan-Tompkins algorithm was used to detect the R wave and construct the RR interval sequence. Feature extraction included: Time-domain characteristics: standard deviation (standard deviation of all normal RR intervals), root mean square difference (root mean square of the difference between adjacent RR intervals); Frequency domain characteristics: low-frequency power, high-frequency power, and the ratio of low-frequency power to high-frequency power; Nonlinear characteristics: Sample entropy (a measure of the complexity of RR interval sequences); Electrodermal signal processing: Preprocessing includes 0.3Hz low-pass filtering and Z-score normalization; feature extraction includes: Skin conductivity levels (mean, standard deviation); Nonspecific skin conductance response frequency; 5. Feature Fusion and Model Building Feature dimensionality reduction: Principal component analysis was performed on the extracted 11-dimensional original features, and principal components with a cumulative variance contribution rate ≥ 90% were retained. Selecting the first 6 principal components retained 92.2% of the cumulative variance, generating a 6-dimensional fused feature vector.

[0021] Model building: A hybrid architecture of principal component analysis, LSTM, and attention is constructed, with the following parameters: LSTM layer: Single layer, 96 hidden units, processes the input sequence to generate a hidden state sequence; Attention mechanism: An additive attention model is used, and the calculation process is as follows: Suppose that the LSTM network processes a length of After processing the physiological signal sequence, output the hidden state vector. ,in Representing the Feature vectors at each time step; First, calculate each time step. Unnormalized attention score :

[0022] in: It is the first The LSTM hidden states at each time step (96 dimensions). It is a trainable weight matrix (dimensions). ×96). It is a trainable bias vector (dimension) ), It is a trainable context vector (dimension 1) ), It is the hyperbolic tangent activation function; Subsequently, the scores at all time steps were normalized into attention weights using the Softmax function. :

[0023] This weight satisfy This reflects the first The relative importance of physiological signal features at each time step to the final emotion classification decision; Finally, a context vector incorporating global temporal information is obtained through weighted summation. :

[0024] Fully connected layer: a three-layer structure with 128, 64, and 32 neurons respectively; Drop rate: A decreasing strategy is adopted, with values ​​of 0.4, 0.3, and 0.2 respectively; Output layer: 3 neurons, corresponding to three emotional states: positive, neutral, and negative; Model training: The Adam optimizer was used with an initial learning rate of 0.001. Dynamic learning rate adjustment and early stopping mechanism were combined, and 5-fold cross-validation was used. The training set, validation set and test set were divided in a ratio of 7:1.5:1.5. 6. Results Analysis and Application Model Performance: After training, the model achieved an average classification accuracy of 88.06% in the user experience evaluation of smart home control systems. The performance of each category is as follows: Positive sentiment: Precision 84.75%, Recall 83.33%, F1 score 84.03%; Neutral sentiment: Precision 86.89%, Recall 88.33%, F1 score 87.60%; Negative sentiment: Precision 92.50%, Recall 92.50%, F1 score 92.50%; Key experience moment location: By visualizing the attention weight distribution map, multiple key experience moments were successfully located. The specific location method is as follows: In the attention weight distribution map, the time period when the weight value is higher than the preset threshold (set to 0.15 in this embodiment) is identified as the emotional peak or trough moment. The identified high-weight moments include: High-weight positive moments: When the intelligent scene is successfully started with one click (weight value 0.89); When voice control is accurately recognized and executed (weight value 0.82); High-weight negative moments: When device connection fails and manual reconnection is required (weight value 0.91); When the interface menu hierarchy is too deep, causing confusion in operation (weight value 0.87).

[0025] Performance comparison and verification: The following four representative existing technologies were selected as benchmarks for comparison: (1) Single-modal EEG signal method Technical basis: Emotion recognition using only brainwave signals; Feature extraction: Extracting features such as relative power and differential asymmetry in the four frequency bands of the prefrontal cortex: Alpha, Beta, Theta, and Gamma. Classification model: Support Vector Machine classifier; Parameter settings: Radial basis function, penalty factor C=1.0, kernel function parameter γ=0.1.

[0026] (2) Traditional subjective scale method Technical basis: Utilizing subjective assessments based on user self-reporting; Assessment tools: System usability scale and user experience questionnaire; Implementation method: Complete a standardized questionnaire immediately after completing the task; Data analysis: Calculate scores for each dimension, and perform descriptive statistics and correlation analysis.

[0027] (3) Traditional machine learning methods Feature extraction: The same multimodal physiological features as in this invention; Feature processing: Use the original features directly without performing principal component analysis for dimensionality reduction; Classification model: Support Vector Machine classifier; Parameter settings: Determine the optimal parameters through grid search.

[0028] (4) Random Forest Method Feature extraction: The same multimodal physiological features as in this invention; Classification model: Random Forest classifier; Parameter settings: Number of trees = 100, maximum depth = 10, minimum sample split = 2.

[0029] Table 1: Performance Comparison of Different Methods in Smart Home Evaluation

[0030] Furthermore, we will enhance the stability and response speed of intelligent scenarios, optimize the accuracy of voice recognition algorithms, and simplify the device connection process.

[0031] Example 2: Please refer to Figure 1 and Figure 3 An objective evaluation method for user experience emotion that integrates multimodal physiological signals: Online education platform learning experience evaluation: Experimental design: The course learning module of an online education platform was selected for evaluation, with a focus on examining changes in emotional experience during the learning process; Task profile: Attractiveness of video tutorial content: video quality, clarity of explanation; The appropriateness of the difficulty level of interactive exercises: gradient of question difficulty, and timeliness of feedback; Timely feedback on learning progress: progress reminders and achievement feedback; A sense of accomplishment from knowledge acquisition: test results, learning reports; Adaptive adjustments to technical solutions: The experimental design was optimized to suit the characteristics of educational settings: Pre- and post-tests of learning outcomes were added, and a correlation analysis between emotional state and learning outcomes was established. The task duration has been adjusted, with each learning segment extended to 3 minutes. Added subjective assessment items related to focus; Experimental results: The model achieved a classification accuracy of 86.42%. Attention weight analysis revealed that: The interactive Q&A session during video playback elicited a significant positive response (peak weight 0.85). The sudden increase in difficulty of practice questions led to significant negative emotions (weight peak 0.88). Timely feedback on learning outcomes effectively enhances learning motivation (weight remains high at 0.78-0.82).

[0032] Example 3: Please refer to Figure 1 and Figure 4 An objective evaluation method for user experience emotion that integrates multimodal physiological signals: Automotive in-vehicle information system evaluation: Application Scenarios Expansion: To further demonstrate the broad applicability of the method, evaluation cases for automotive in-vehicle information systems have been added, with a focus on: Distraction assessment during driving; Efficiency and satisfaction of voice interaction; The impact of interface design on driving safety; Special technical handling: To address the specific characteristics of driving scenarios, the following additions were made to the experimental design: The construction of a simulated driving environment includes driving simulators and road scenarios; The experimental procedure design prioritizes safety and includes an emergency braking test component. Emergency system response testing evaluates the system's performance under stress. Experimental results: The model achieved a classification accuracy of 84.73% in driving scenarios. Key findings include: Complex menu operations led to a 42% increase in the driver distraction index (weight 0.89). The decreased recognition rate of voice control in poor road conditions leads to frustration (weight 0.86). A clean and simple interface design improves driving safety (positive weight 0.81).

[0033] The working principle involves constructing a VR emotion-inducing environment and simultaneously acquiring multimodal physiological signals. Combined with standardized target emotion type definitions, multi-dimensional task profile extraction, and standardized signal acquisition parameter settings, a complete standardized experimental system is formed. This ensures the standardization and repeatability of the entire process from experimental design to data acquisition, effectively solving the problems of large individual differences, inconsistent experimental conditions, and difficulty in reproducing results inherent in traditional subjective evaluation methods. This lays a solid foundation for obtaining high-quality, comparable physiological data. Based on multimodal signal acquisition, and combined with specialized preprocessing methods for different physiological signals and feature extraction calculation formulas based on neuroscience principles, the quality of the original signals and the effectiveness of feature parameters are significantly improved. Furthermore, through professional signal denoising, artifact removal, and feature quantization, the original physiological signals are transformed into effective feature indicators that accurately reflect emotional states, providing high-quality input for subsequent emotion recognition and ensuring the scientific rigor of the entire technical solution from data acquisition to feature extraction. To improve reliability, a high-performance, high-reliability emotion recognition model was constructed by combining a regularization strategy with a decreasing dropout rate, clear performance metric requirements, and rigorous cross-validation methods on the basic LSTM-Attention architecture. This combination ensures that the model maintains high accuracy while possessing good generalization ability and stability through specific network parameter optimization, overfitting prevention measures, and statistical validation processes. It effectively solves the technical problems of low accuracy and poor robustness of traditional emotion recognition methods. By combining a key experience moment localization mechanism based on attention weights and a practical application deployment scheme within the overall methodological framework, a complete closed loop from technical implementation to industrial application is formed. This combination can not only accurately identify the emotional peaks and troughs in the user experience process through attention weight distribution, but also transform the analysis results into specific design optimization suggestions and realize real-time emotion evaluation functions, significantly improving the practical value and industrialization potential of the technology and realizing the effective transformation of scientific research results into practical tools.

[0034] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An objective evaluation method for user experience emotion that integrates multimodal physiological signals, characterized in that: Includes the following steps: S1. Construct a VR emotion-inducing experimental environment. Based on the user's interaction with the product, simultaneously collect EEG signals, ECG signals, and skin conductance signals, and record subjective evaluation data synchronized with the interaction events. S2. Extract the asymmetry index of the prefrontal alpha band from the EEG signal, extract the standard deviation and root mean square difference of the heart rate variability time domain index, the low-frequency to high-frequency ratio and sample entropy from the ECG signal, and extract the skin conductance level and non-specific skin conductance response frequency from the EEG signal. S3. Principal component analysis is used to reduce the dimensionality of the extracted features and fuse them, retaining the principal components with a cumulative variance contribution rate of ≥90%, and generating a low-dimensional fused feature vector. S4. Input the feature vector into a single-layer LSTM network in time series. The network has 96 hidden units and connects three fully connected layers with a 128-64-32 structure. Introduce an attention mechanism to calculate the weights of each time step and output the sentiment state classification result. S5. Generate a dynamic emotional state curve based on the classification results, locate key experience moments through the attention weight distribution map, and associate key experience moments with interaction event timestamps.

2. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: The VR emotion-inducing experimental environment constructed in step S1 includes: The target emotional type is categorized into three types: positive, neutral, and negative. Typical task profiles are extracted from the dimensions of product appearance design approval, smoothness of use process, physical comfort, sense of technology and intelligence. The experimental procedure included baseline resting state recording, experience of multiple task segments, and subjective evaluation after each segment.

3. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: The acquisition of EEG signals, ECG signals, and skin conductance signals in step S1 specifically involves: The EEG signals were acquired using a 32-lead EEG instrument, focusing on the prefrontal cortex electrode sites FP1, FP2, F3, F4, F7, and F8, with a sampling frequency of 512 Hz. The electrocardiogram signal was acquired using a three-lead configuration with a sampling frequency of 500Hz. Electrodermal signals were acquired using a multichannel physiological recorder at a sampling frequency of 500 Hz.

4. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: The preprocessing of the EEG signal in step S2 includes rereference, downsampling, filtering, and removal of electrooculogram artifacts using independent component analysis. The preprocessing of the electrocardiogram signal uses a Butterworth low-pass filter, an FIR band-notch filter, and a zero-phase-shift linear filter to remove electromyographic artifacts, power line interference, and baseline drift, respectively. The preprocessing of the electrodermal signal includes 0.3Hz low-pass filtering and Z-score normalization.

5. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: In step S2, the prefrontal alpha band asymmetry index is calculated by examining the alpha band signal power of the three pairs of co-originating electrodes in the prefrontal cortex: FP1 and FP2, F3 and F4, and F7 and F8. The formula is as follows: ； in, and These represent the signal power of the right and left electrodes in the alpha band, respectively.

6. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: In step S4, the LSTM network training uses a decreasing dropout rate. The dropout rates for the three fully connected layers are 0.4, 0.3, and 0.2, respectively. The training optimizer is Adam, the initial learning rate is 0.001, and the training cycle is dynamically adjusted using an early stopping mechanism.

7. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: The accuracy of the emotional state classification results in step S4 is no less than 85%.

8. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: The location of key experience moments in step S5 is specifically as follows: in the attention weight distribution map, the time period with a weight value higher than a preset threshold is identified as the emotional peak or trough moment. This time period is matched with the timestamp of the interaction event recorded in step S1, thereby locating the specific interaction event that triggers positive or negative emotions.

9. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: In step S4, the LSTM network is trained using 5-fold cross-validation.

10. The objective evaluation method for user experience emotion that integrates multimodal physiological signals according to claim 1, characterized in that: It also includes the application phase: deploying the trained model on the user experience testing platform to perform real-time emotional evaluation of the interaction process of new products and output a design optimization report that includes the location of key experience moments.

Images