Self-assessment of affect-related events for physiological data collection in the wild based on appraisal theories

This paper addresses the need for collecting and labeling affect-related data in ecological settings. Collecting the annotations in the wild is a very challenging task, which, however, is crucial for the creation of datasets and emotion recognition models. We propose a novel solution to collect and annotate such data: a questionnaire based on the appraisal theory, that is accessible through an open-source mobile application. Our approach exploits a commercially available wearable physiological sensor connected to a smartphone. The app detects potentially relevant events from the physiological data, and prompts the users to report their emotions using a novel questionnaire based on the Ortony, Clore, and Collins (OCC) Model. The questionnaire is designed to gather information about the appraisal process concerning the significant event. The app guides a user through the reporting process by posing a series of questions related to the event. As a result, the annotated data can be used, e.g., to develop emotion recognition models. In the paper, we analyze users' reports. To validate the questionnaire, we asked 22 individuals to use the app and the sensor for a week. The analysis of the collected annotations shed new light on self-assessment in terms of appraisals. We compared a proposed method with two commonly used methods for reporting affect-related events: (1) a two-dimensional model of valence and arousal, and (2) a forced-choice list of 22 labels. According to the results, appraisal-based reports largely corresponded to the self-reported values of arousal and valence, but they differed substantially from the labels provided with a forced-choice list. In the latter case, when using the forced-choice list, individuals primarily selected labels of basic emotions such as anger or joy. However, they reported a greater variety of emotional states when using appraisal theory for self-assessment of the same events. Thus, proposed approach aids participants to focus on potential causes of their states, facilitating more precise reporting. We also found that regardless of the reporting mode (mandatory vs. voluntary reporting), the ratio between positive and negative reports remained stable. The paper concludes with a list of guidelines to consider in future data collections using self-assessment.