Recommendations in visual analytics using emotions : a mixed-initiative interaction approach
Abstract
The thesis demonstrates an idea for helping users in visual analytic tasks by investigating some critical steps required for providing recommendations. The proposed model uses mixed-initiative interaction approach by detecting users’ negative emotions, caused by the visual analytic tasks, as a cue to generate useful guidance. For building a negative emotion detection classifier, I have created a dataset from 28 participants carrying out intentionally difficult visualization tasks and collected their emotional responses using multiple biosensors. I used this dataset to built a real-time emotion detection model which predicts mental state in every 4s. Next, the visualization tool uses the detected emotions to generate a recommendation and decide when to intervene. Additionally, the system also adapts intrusion level by analyzing long-term emotions, and decide the best way to show the help. Finally, I have concluded this work by discussing the design space of interventions for providing just-in-time assistance in visual analytics.