A Ranking Model for Evaluation of Conversation Partners Based on Rapport Levels

Our proposed ranking model ranks conversation partners based on self-reported rapport levels for each participant. The is important tasks that recommend interaction user built in past interactions, such as matchmaking between a student and teacher one-to-one online language classes. To rank partners, we can use regression predicts ratings. It is, however, challenging to learn the mapping from participants’ behavior their associated ratings because subjective scale may vary across different participants. Hence, propose trained via preference learning (PL). avoids bias predict ordinal relations two reported by same input of multimodal (acoustic linguistic) features extracted behaviors an interaction. Since there no publicly available dataset validating model, created new composed dyadic (person-to-person) interactions participant several partners. We compare with using evaluation metrics ranking. experimental results show more suitable approach Furthermore, investigate effect modality stages development performance.