Weighting Model Based on Group Dynamics to Measure Convergence in Multi-party Dialogue

Abstract

This paper proposes a new weighting method for extending a dyad-level measure of convergence to multi-party dialogues by considering group dynamics instead of simply averaging. Experiments indicate the usefulness of the proposed weighted measure and also show that in general a proper weighting of the dyadlevel measures performs better than nonweighted averaging in multiple tasks.

Entrainment2Vec: Embedding Entrainment for Multi-Party Dialogues

“…Individually taken self-reported pre-and post-game surveys are available for both sessions, including: (1) favorable social outcome measures (perceptions of cohesion, satisfaction, potency/efficacy and perceptions of shared cognition), and (2) conflict measures (task, process, and relationship conflicts). Following prior works using the Teams corpus (Rahimi and Litman 2018;Yu and Litman 2019), we created a team-level Favorable measure by z-scoring and averaging all the highly correlated favorable measures and averaging them for each team. We followed the prior work and z-scored the process conflict measure and averaged it in the groups to construct a team-level Conflict measure.…”

“…Evidence of entrainment has been found for multiple aspects of speech, including lexical choice (Brennan and Clark 1996;Metzing and Brennan 2003;Niederhoffer and Pennebaker 2002;Danescu-Niculescu-Mizil, Gamon, and Dumais 2011;Gonzales, Hancock, and Pennebaker 2010;Pennebaker, Francis, and Booth 2001;Beňuš, Levitan, and Hirschberg 2012;Rahimi et al 2017;Friedberg, Litman, and Paletz 2012) in both human-human and human-computer dialogues. In addition, the strength of entrainment has been shown to be associated with numerous social and conversational qualities, such as the cohesiveness of speech (Lubold and Pon-Barry 2014;Natale 1975;Rahimi and Litman 2018;Beňuš et al 2014;Danescu-Niculescu-Mizil et al 2012).…”

“…While entrainment studies have largely focused on twoparty conversations, recent studies have estimated entrainment in multi-party conversations by averaging pairs' en-trainment values, or by averaging individuals' entrainment to the group (Gonzales, Hancock, and Pennebaker 2010;Nenkova, Gravano, and Hirschberg 2008;Friedberg, Litman, and Paletz 2012;Danescu-Niculescu-Mizil et al 2012;Rahimi and Litman 2018). However, the entrainment structure of multi-party groups consists of many pairwise relations with different strengths and directions.…”

“…To the best of our knowledge, almost no existing multi-party entrainment measures utilize the structure of pairwise entrainment relations in groups. Measuring convergence, Rahimi and Litman (2018) introduced a weighting based on entrainment behaviors (converging, diverging, or maintaining) in groups to decrease the influence of outliers. Although this is an attempt to utilize the structure of group entrainment relations, it encodes only one aspect of these relations.…”

“…To encode the structure of small graphs/groups (only about 4 nodes/speakers), we propose to initialize the node and graph embeddings by applying a set of graph algorithms where each encodes a distinctive property of the graph. As the supervision, we propose to employ the domain-specific task of predicting real versus randomly permuted conversations, which has been utilized in the entrainment domain to verify the validity of measures (De Looze et al 2014;Lee et al 2011;Jain et al 2012;Rahimi and Litman 2018). Experimental evaluations demonstrate that the group entrainment embedding improves performance for the downstream task of predicting group outcomes compared to the state-of-the-art methods.…”