Browsing by Author "Henriques, Nuno Andrade da Cruz"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- SensAI+Expanse : Prediction of Emotional Valence Changes on Humans in Context by an Artificial Agent Towards EmpathyPublication . Henriques, Nuno Andrade da Cruz; Coelho, Helder Manuel Ferreira; Garcia-Marques, LeonelThe field of Cognitive Science is broader enough on the interdisciplinary study of the brain, mind, and intelligence with a scientific research community gaining momentum over the last few decades. Specifically, joining the two fields of psychology and artificial intelligence (AI) one may envision agents, embodied or not, human-like or wearable, with the ability to significantly change the way humans live. This research conceive the artificial agent as a non-anthropomorphic with adaptive empathy for human-agent interaction (HAI) synergy towards better companionship. Therefore, the main objectives of this research are (a) to build a predictive model for each human user on context-based emotional valence changes; and (b) to study the age, gender, and human behaviour neutrality and robustness of the artificial agent regarding the prediction ability. The context include geographically located data from sensors, text sentiment analysis, and human emotional valence self-report, all timestamped events, using a common mobile device such as a smartphone. Also, to analyse and discuss the results on how to leverage such a model to adapt interaction strategies in order to foster higher levels of empathy between a non-anthropomorphic agent and its interacting human. For these goals SensAI+Expanse is developed where SensAI acts as an embodied nearby agent and Expanse encompass the machine learning resources in efficient manner, i.e., a distributed, fault-tolerant, mobile and Cloud-based platform from scratch as a research tool to continuously, online, gather and process data towards automated machine learning (AutoML) and prediction. The study is designed with a methodology in place to avoid the Western, Educated, Industrialized, Rich, and Democratic (WEIRD) societies bias. This goal is accomplished by collecting data in the wild and worldwide by making use of the publicly accessible Google Play repository for the Android™ SensAI smartphone application. Eligible participants are diverse in age, gender, and behaviour on self-reporting emotional valence. In order to balance the gender distribution by age a dichotomy approach using age median (M = 34) is used. Regarding participation duration, two thirds (33/49) of the eligible individuals for analysis remain interacting for the required minimum of four weeks. The analysis of the results show evidence of significant behaviour differences between some age and gender combinations regarding self-reported emotional valence. Furthermore, the results from a comparison study between state-of-the-art algorithms revealed Extreme Gradient Boosting on average the best model for prediction (F1 = 0:91) with efficient energy use, and explainable using feature importance inspection. Moreover, the artificial agent remained neutral regarding human demographics and, simultaneously, able to reveal individual idiosyncrasies. Therefore, this research contributions include results with evidence, restricted to population and data samples available, of differences in behaviour amongst some combinations of age ranges versus gender. The main contribution is a novel platform for studies regarding human emotional valence changes in context. This system may complement and supersede (eventually) traditional long-list self-appraisal questionnaires. The SensAI+Expanse platform contributes with several parts such as a mobile device application (SensAI) able to adapt and learn in order to predict emotional valence states with high performance, a cloud computing (Cloud) service (SensAI Expanse) with ready-to-action analysis and processing modules towards AutoML. Additionally, smartphone sensing add a contribution for continuous, non-invasive and personalised health check. In the future, developments about human-agent relationships regarding affective interactions are foreseen. Further, the measurement of empathetic reactions and evaluating outcomes may be used to verify and validate health status thus improving care and significantly change the way humans live.