Predicting two facets of social verticality in meetings from five-minute time slices and nonverbal cues

Abstract

This paper addresses the automatic estimation of two aspects of social verticality (status and dominance) in small-group meetings using nonverbal cues. The correlation of nonverbal behavior with these social constructs have been extensively documented in social psychology, but their value for computational models is, in many cases, still unknown. We present a systematic study of automatically extracted cues -including vocalic, visual activity, and visual attention cues -and investigate their relative effectiveness to predict both the most-dominant person and the high-status project manager from relative short observations. We use five hours of task-oriented meeting data with natural behavior for our experiments. Our work suggests that, although dominance and role-based status are related concepts, they are not equivalent and are thus not equally explained by the same nonverbal cues. Furthermore, the best cues can correctly predict the person with highest dominance or role-based status with an accuracy of 70% approximately.

Content-based unsupervised segmentation of recurrent TV programs using grammatical inference

“…We use five measures to characterize each person cluster with the idea of finding the person that covers at best a significant amount of time in an episode [14]: total duration of appearance; total number of distinct appearances, i.e., number of non consecutive segments; duration of the longest segment in which the person appears; time range between the first and last occurrence; duration in which the speaker is engaged in a dialog. To account for varying episodes and program lengths, all five measures are scaled to [0,1].…”

Multimodal Detection of Engagement in Groups of Children Using Rank Learning

“…Moreover, we do not have prior knowledge of the proper size of a window for feature extraction in our study. In previous work [29,33,25], windows of between 0.5s and 5min long were used to predict engagement and dominance. To decide an optimal length of windows for detection experiments, we investigated the effect of different window length varying: {5s, 10s, 15s, 20s, and 25s}.…”

Mining Group Nonverbal Conversational Patterns Using Probabilistic Topic Models

“…The observations were speaking activity features and influence was estimated using a variation of the coupled HMM called the influence model. On the AMI corpus, two facets of social verticality [18], i.e., role-based status and dominance, was predicted, by employing speaking turns as well as visual attention based cues [21]. On the Augmented MultiParty Interaction with Distance Access (AMIDA) corpus, the remote participant in a remote meeting was predicted [45].…”

“…Social verticality in groups has been shown to be correlated to floor occupation related nonverbal cues [18]. Previous works have shown that the person with the highest speaking time correlates with the most dominant person [23], highest number of speaking turns correlates with role-based status [21], and highest number of successful interruptions signals real status and power [41]. In order to capture the leader's position in the group, we add three more words to the NVP vocabulary for each of the four sets of features to indicate whether the designated leader ("L") or someone else ("NL") is the one who has the maximum.…”

“…While modeling individuals allows to study dominance [23], [43], status [21], roles [11], [48], personality [38], modeling groups could reveal group interest [16], interactivity and centrality in groups [36], and identify cooperative groups against competitive groups [24]. The methods studied so far in the computational literature have mostly used supervised learning approaches.…”