TY - GEN
T1 - Towards Real-Time Contextual Touristic Emotion and Satisfaction Estimation with Wearable Devices
AU - Fedotov, Dmitrii
AU - Matsuda, Yuki
AU - Takahashi, Yuta
AU - Arakawa, Yutaka
AU - Yasumoto, Keiichi
AU - Minker, Wolfgang
PY - 2019/3/1
Y1 - 2019/3/1
N2 - Following the technical progress and growing touristic market, demand on guidance systems is constantly increasing. Current systems are not personalized, they usually provide only a general information on sightseeing spot and do not concern about the tourist's perception of it. To design more adjustable and context-aware system, we focus on collecting and estimating emotions and satisfaction level, those tourists experience during the sightseeing tour. We reducing changes in their behaviour by collecting two types of information: conscious (short videos with impressions) and unconscious (behavioural pattern recorded with wearable devices) continuously during the whole tour. We have conducted experiments and collected initial data to build the prototype system. For each sight of the tour, participants provided an emotion and satisfaction labels. We use them to train unimodal neural network based models, fuse them together and get the final prediction for each recording. As tourist himself is the only source of labels for such system, we introduce an approach of post-experimental label correction, based on paired comparison. Such system built together allows us to use different modalities or their combination to perform real-time tourist emotion recognition and satisfaction estimation in-the-wild, bringing touristic guidance systems to the new level.
AB - Following the technical progress and growing touristic market, demand on guidance systems is constantly increasing. Current systems are not personalized, they usually provide only a general information on sightseeing spot and do not concern about the tourist's perception of it. To design more adjustable and context-aware system, we focus on collecting and estimating emotions and satisfaction level, those tourists experience during the sightseeing tour. We reducing changes in their behaviour by collecting two types of information: conscious (short videos with impressions) and unconscious (behavioural pattern recorded with wearable devices) continuously during the whole tour. We have conducted experiments and collected initial data to build the prototype system. For each sight of the tour, participants provided an emotion and satisfaction labels. We use them to train unimodal neural network based models, fuse them together and get the final prediction for each recording. As tourist himself is the only source of labels for such system, we introduce an approach of post-experimental label correction, based on paired comparison. Such system built together allows us to use different modalities or their combination to perform real-time tourist emotion recognition and satisfaction estimation in-the-wild, bringing touristic guidance systems to the new level.
UR - http://www.scopus.com/inward/record.url?scp=85067921505&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85067921505&partnerID=8YFLogxK
U2 - 10.1109/PERCOMW.2019.8730864
DO - 10.1109/PERCOMW.2019.8730864
M3 - Conference contribution
T3 - 2019 IEEE International Conference on Pervasive Computing and Communications Workshops, PerCom Workshops 2019
SP - 358
EP - 360
BT - 2019 IEEE International Conference on Pervasive Computing and Communications Workshops, PerCom Workshops 2019
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2019 IEEE International Conference on Pervasive Computing and Communications Workshops, PerCom Workshops 2019
Y2 - 11 March 2019 through 15 March 2019
ER -