محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| التنسيق: | Recurso digital |
| اللغة: | |
| منشور في: |
Zenodo
2025
|
| الموضوعات: | |
| الوصول للمادة أونلاين: | https://doi.org/10.5281/zenodo.15870704 |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
جدول المحتويات:
- <p>AMUSED (Affective Metrics & Usability Smell Evaluation Dataset) is a comprehensive multimodal dataset designed to advance research in Human-Computer Interaction (HCI), usability evaluation, and emotion recognition. This dataset combines user interaction data with physiological signals and emotional responses to provide insights into how users experience usability issues.</p> <p>GitHub: <a href="https://github.com/flassantos/amused" rel="noopener">https://github.com/flassantos/amused</a></p> <h3>Dataset Highlights</h3> <ul> <li><strong>70 participants</strong> with equal gender distribution, ages 18-61</li> <li><strong>24 hours 46 minutes</strong> of total recorded interaction time</li> <li><strong>20,050 user interaction events</strong> (clicks, scrolls, URL changes, input changes, keypress)</li> <li><strong>Multiple data modalities</strong> synchronized in time</li> <li><strong>Expert annotations</strong> of 11 types of usability smells by 20 HCI experts</li> <li><strong>Three social networking platforms</strong> with usability issues (prototypes)</li> </ul> <h3>Data Modalities</h3> <ul> <li><strong>User Interaction Logs:</strong> The dataset captures comprehensive user interaction events including clicks, scrolls, navigation, and text input, totaling over 20,050 events across all participants. The interaction data is structured in JSON format with detailed DOM elements and XPath information, organized temporally as episodes and tasks.</li> <li><strong>Physiological Signals (56 participants): </strong>EEG data was recorded using 2-channel forehead placement at 1000Hz sampling rate with extracted frequency bands (theta, alpha, beta, gamma) for cognitive state analysis. Blood Volume Pulse (BVP) signals were captured via PPG sensor to measure heart rate variability, providing insights into users' physiological responses during interactions.</li> <li><strong>Facial Emotion Recognition (50 participants):</strong> Facial expressions were analyzed using a pre-trained VGG16 CNN model achieving 84.6% validation accuracy to classify emotions into 7 Ekman categories: joy, sadness, anger, surprise, fear, disgust, and contempt. Video recordings at 30 FPS enabled frame-by-frame emotion classification synchronized with user actions.</li> <li><strong>Self-Assessment Reports (29 participants):</strong> Participants completed Self-Assessment Manikin (SAM) questionnaires reporting valence and arousal dimensions after each task. These subjective emotional assessments enable comparison with objective physiological and facial emotion measurements, providing a comprehensive view of user emotional states.</li> </ul> <h3>Usability Smell Annotations</h3> <p>The dataset features expert annotations of usability smells conducted by 20 HCI evaluators with varying levels of experience, ensuring comprehensive coverage of usability issues. </p> <p>The annotation schema operates at two levels: task-level smells that impact overall task completion (Laborious Task, Cyclic Task, Too Many Layers, Missing Task Feedback, High Interaction Distance, Repetition in Text Fields, and Late Validation) and action-level smells that focus on specific user actions (Undescriptive Element, Missing Action Feedback, Unnecessary Action, and Misleading Action). The resulting annotations comprise 1,873 task-level smell instances and 3,336 action-level smell instances with precise timestamps, revealing detailed co-occurrence patterns that demonstrate how multiple usability issues compound to create user frustration.</p> <p> </p>