Fig 1.
Triandis’ Theory of Interpersonal Behaviour.
(Reference: Jackson [22]).
Fig 2.
The field nature is visible in the positioning of similar behaviours next to each other. (Extracted from Orford [49]).
Table 1.
Summary of data-driven models for engagement prediction, their strengths and weaknesses.
Fig 3.
Representation of the proposed pipeline: From multimodal data to behaviour inferences.
E: Extraversion, O: Openness. A: Agreeableness, C: Conscientiousness, N: Neuroticism.
Table 2.
The upper, lower, and mean angles used for projecting the Big Five traits over the IPC.
(Extracted from Du et al. [21]).
Fig 4.
Relating emotion and attitude to behaviour using IPC.
(a) IPC for attitude. (Extracted from Dermouche and Pelachaud [28].) (b) IPC for emotion. (Extracted from Russell [25].).
Table 3.
Predicting the engagement index from the 30 physiological and 185 first-person vision features of 18 users in HHI and HRI settings.
An SVM classifier is used in each case. The values are the F1 scores obtained after performing 18-fold cross-validation. The F1 scores of Celiktutan et al. [18] are presented in brackets for comparison, with improvements shown in bold.
Table 4.
F1 scores for the Big Five traits predicted from the multimodal features in both HHI and HRI settings, with self and acquaintance labels.
The F1 scores by Celiktutan et al. [18] for the high class are presented in brackets for comparison. The scores which are an improvement are shown in bold.
Table 5.
The Pearson Correlation coefficient quantifies the correlation between ground truth engagement indices and the predicted Big Five trait scores in the HRI setting.
Table 6.
Correlation matrix of the IPC components with the engagement index using ground truth self and acquaintance labels in the HRI setting.
Table 7.
Correlation values of the IPC components with the engagement index using ground truth self and acquaintance labels for introverts and extroverts in the HRI setting.
Table 8.
Observations from correlation patterns of engagement with attitude and emotion, after segregation into extroverts and introverts.
A highly engaging individual would show a positive correlation with attitude and emotion scores.
Fig 5.
The horizontal component of the IPC embedding gives a measure of emotion and the vertical component, a measure of attitude.
(a) Projection of the personality embedding (Big Five trait scores) onto the IPC for user U003 in the HRI setting. (b) Representation of the relation of engaging behaviour to attitude and emotion for user U003, whose engagement index is 5 (out of 10), in the HRI setting.