Ngoc Doan*, Dan Hudson*, Travis J. Wiltshire, Philia Lijdsman, Stijn Wever, Martin Atzmueller

*Both authors contributed equally to this research

 

Data-driven methods enable the use of finer temporal timescales and behavior modeling with fewer assumptions. This work focuses on analyzing face-to-face interactions in a data-driven way. We present HINTS, an analytics system that examines meeting dynamics and visualizes the results in an easily interpretable way. This can help to answer difficult questions in group work, e. g., how to quantify the behavior of individual users, their interactions, and groups/teams and, ideally, to find differences between high-performing/low-performing teams. Analyzing the behavior of groups, e. g., in the context of teamwork itself is an important topic because teams rather than individuals are needed to handle the complex problems encountered in modern society, especially where solutions are work-intensive and demand multiple types of expertise.

 

System and methods 

The data is collected and streaming from the Sociometric Badges, specifically the Rhythm Badges, unobtrusive wearable sensors that record the ongoing speech volume, proximity to other badges, and movement acceleration. The accumulated data is analyzed as well as visualized by HINTs at the end of the meeting to construct a meeting report. At the moment, HINTs only provide the report in the end. Future works aim to develop HINTs as a more interactive dashboard.

 An application workflow diagram

 

Our analytics system aims to characterize the dynamics of communication. First, it calculates the ‘dynamic complexity’ measure, which quantifies how disordered values are at different times in a time series. Dynamic complexity is visualized through heatmaps, showing the complexity values for all team members over the course of an interaction. The phase transitions can be derived from the shifts in dynamic complexity. This provides a temporal structuring dividing the overall recording period into the moments between critical instabilities.

Within each communication phase, we can then visualize the patterns of communication within the team by extracting and analyzing the turn-taking behavior. The turn-taking behavior is measured according to two metrics: the frequency with which each team member talks (energy), and the pairwise frequency with which two team members respond to one another (engagement). The metrics are visualized in the format of network diagrams. In the diagram, the size of each node is the speaker's speaking frequency, and the width of the edges linking speakers representing the inter-speaker interaction frequency. The visualizations provide a fast impression of the balance of the conversation(s) during each phase of a meeting.

A heatmap visualisation, a network visualisation and visualisations of demographics

 

Initial case study and future directions

We applied our system in an initial case study to analyze team dynamics during a decision-making task. The task is the Lost at Sea paper-based activity in which they collaboratively ranked the importance of items on a list, given a disaster scenario. The main goal of the case study was to investigate team dynamics in a meeting-like setting where group decision-making and seeking consensus are central. This initial case study gives preliminary evidence that HINTS is capable of finding meaningful temporal structure in social interactions, which can then be used to visualize speech dynamics at different phases of a meeting. Unfortunately, more experimentation has been delayed due to the ongoing pandemics.

In the future, we hope to conduct more experimentation, testing, and refinement of the system. In addition, we hope to further develop HINTs to be robust to multiple teams of different nature. We also aim to complement the data using further sensors and analytical options.