This paper reports an experiment for stress recognition in human-computer interaction. Thirty-one healthy participants performed five stressful HCI tasks and their skin conductance signals were monitored. The selected tasks were most frequently listed as stressful by 15 typical computer users who were involved in pre-experiment interviews asking them to identify stressful cases of computer interaction. The collected skin conductance signals were analyzed using seven popular machine learning classifiers. The best stress recognition accuracy was achieved by the cubic support vector machine classifier both per task (on average 90.8 %) and for all tasks (Mean = 98.8 %, SD = 0.6 %). This very high accuracy demonstrates the potentials of using physiological signals for stress recognition in the context of typical HCI tasks. In addition, the results allow us to move on a first integration of the specific stress recognition mechanism in PhysiOBS, a previously-proposed software tool that supports researchers and practitioners in user emotional experience evaluation.

HCI researchers and practitioners are increasingly using physiological data to measure User eXperience (UX) parameters. The dynamic nature of physiological data offers a continuous window for an in-depth understanding of users’ interaction experience. However, in order to be truly informative, physiological signals need to be linked to users’ interaction experience aspects, such as their emotional states, in a systematic and efficient way. Studies have shown that skin conductance is a physiological signal highly associated with stress. The main purpose of this paper is to present the validation study of our proposed stress detection mechanism which is integrated into a software named PhysiOBS. PhysiOBS is an observation analysis tool that can be used in the post-study analysis phase. PhysiOBS uses nonspecific skin conductance responses (NS-SCRs) in order to auto-report time periods that are probably associated with a problematic interaction. PhysiOBS can also combine multiple data sources. Hence, UX evaluators are able to further investigate a recorded session in order to reveal additional interaction flaws. The integrated stress assessment mechanism, which uses four trained classifiers, can be applied in the reported periods (auto/expert-reported) in order to classify them as stress or non-stress. For the purpose of the validation study, 24 users were recruited in order to participate in a lab experiment. Results showed that our stress assessment mechanism supports UX evaluators by accurately identifying stressful regions within an interaction scenario.

The current era of (e-)learning invests heavily on Massive Open Online Courses (MOOCs). Various dedicated MOOC platforms are used, such as Coursera, Udacity, and Open edX. Moodle, one of the most popular open-source Learning Management Systems (LMS), has been also used for MOOC implementations. Usability and user experience (UX) are very important for e-learning software. This paper presents an in-depth comparative UX evaluation of two platforms used for MOOC implementations: Open edX and Moodle. The reported evaluation study combines user interaction metrics and wearable biosensors. Participants performed four typical interaction tasks in both platforms. User performance data (task success, time on task), self-reported data (SUS, AttrakDiff), and stress-related physiological data (skin conductance, heart rate) were collected and analyzed. It was found that Open edX performs better than Moodle in almost every UX aspect we examined. However, the reported results do not imply that building and delivering a MOOC via Moodle is prohibitive. The choice of the platform to be used by an institution is affected by several factors of non-technological nature, such as students’ familiarity with an e-learning platform, available financial and human resources, as well as delivery deadlines.

Physiological measurements have been widely used by researchers and practitioners in order to address the stress detection challenge. So far, various datasets for stress detection have been recorded and are available to the research community for testing and benchmarking. The majority of the stress-related available datasets have been recorded while users were exposed to intense stressors, such as songs, movie clips, major hardware/software failures, image datasets, and gaming scenarios. However, it remains an open research question if such datasets can be used for creating models that will effectively detect stress in different contexts. This paper investigates the performance of the publicly available physiological dataset named WESAD (wearable stress and affect detection) in the context of user experience (UX) evaluation. More specifically, electrodermal activity (EDA) and skin temperature (ST) signals from WESAD were used in order to train three traditional machine learning classifiers and a simple feed forward deep learning artificial neural network combining continues variables and entity embeddings. Regarding the binary classification problem (stress vs. no stress), high accuracy (up to 97.4%), for both training approaches (deep-learning, machine learning), was achieved. Regarding the stress detection effectiveness of the created models in another context, such as user experience (UX) evaluation, the results were quite impressive. More specifically, the deep-learning model achieved a rather high agreement when a user-annotated dataset was used for validation.

This paper investigates gender differences in stress recognition in human computer interaction (HCI) for both objective (i.e., skin conductance) and subjective (i.e., valence-arousal VA ratings) metrics. To this end, 31 healthy participants, 18 females, performed five HCI tasks, while their skin conductance was recorded. These selected HCI tasks were the ones listed as the most stressful, by a group of typical computer users, who were involved in a face to face pre-experiment interview for the identification of stressful cases in computer interaction. After each task, participants rated their interaction experience using the valence-arousal scale. The collected data were split based on participants' gender. Skin conductance signals were analyzed using seven popular machine learning classifiers. In both groups the best stress recognition accuracy for all tasks was achieved by Linear Discriminant Analysis LDA; Males: Mean=94.8% and SD=1.5%, Females: Mean=98.9% and SD=0.3%. Self-reported data analysis revealed a significant difference on how both genders communicate their emotions using the arousal scale. Our findings tend to suggest that gender does not affect skin conductance data during subtle HCI tasks. However subjective ratings such as arousal of emotional experience must be utilized carefully.