Graduate Student Xu Chen Defends Ph.D. Dissertation

Published July 7, 2021

Xu Chen defended his Ph.D. dissertation entitled TeamDNA: Automatic Measures of Effective Teamwork Processes from Unconstrained Team Meeting Recordings on June 29, 2021.

Committee Chair:

Dr. Ashu Sabharwal (Department Chair and Earnest Dell Butcher Professor of ECE)

Committee Members:

Dr. Margaret Beier (Professor of Rice Psychology)
Dr. Santiago Segarra (W. M. Rice Trustee Assistant Professor of ECE)
Dr. Daniel McDuff (from Microsoft Research)

The committee approved his candidacy for a Ph.D. in Electrical and Computer Engineering from Rice University.

His abstract is below.

The complexity of modern engineering tasks requires a diverse range of knowledge and skills that a single person cannot possess. Therefore, the teams formed by people from different backgrounds appear as standard working units, and the ability to work well in multidisciplinary teams has become a favorable skill. As such, in today’s engineering education, team-based learning is prevalent to better prepare future engineers for acquiring teamwork skills. A reliable sensor to gauge how people work together in teams is thus crucial for this educational purpose. Broadly categorized, there are mainly three classes of methods to measure teamwork: 1) “ask them” using self-report methods, where participants are asked to fill in standardized surveys with teamwork-related questions (e.g., team satisfaction, cohesion, etc.) while completing their projects, which is very interruptive; 2) “observe them” administered by a third party human observer who rates how people collaborate together via his/her perceptions or objective project outcomes, which is subjected to either observer biases or the weak relationship between project outcomes and actual teamwork quality; 3) “measure teams by sensors” that is intrusive, with the need for experimenters to wear on-body sensors throughout the study.

Effective teamwork in organization science is decomposed into six components, communication, coordination, contribution, mutual support, cohesion, and effort, in which communication is the most elementary aspect of great significance. However, indicative measures of team communication are not well-known in previous literature.

In this thesis, we propose TeamDNA that quantitatively measures the communication aspect of teamwork from a vital part of team processes — team meetings. TeamDNA can supplement self-report measures by providing objective and non-interruptive measurements, observer-based measures with team process-based analyses, and sensor-based measures with non-intrusive measurements. 

Our contributions are three-fold. First, we compile an unconstrained team meeting dataset — the TeamDNA dataset with actual teams working on real-world projects over a long period. The TeamDNA Dataset contains both a Zoom virtual meeting dataset of Rice engineering design teams across two semesters and an in-person team discussion meeting dataset. Second, we perform extensive analyses to understand what measures are related to the communication component of teamwork for both remote and in-person meeting datasets, from which we find informative speech and gaze features, such as speaking time, silence breaking probability, and the number k-turn interactions, gaze fixation time percentage on speakers and gaze exchange time percentage with speakers. Third, we build the automated system TeamDNA-InPerson gaze estimator to capture gaze dynamics, i.e., “who looks at whom and when” of each participant during multi-party in-person meetings. TeamDNA-InPerson gaze estimation yields over 75% of accuracy in a controlled setup and 67% of accuracy in challenging freeform meeting scenarios.

Congratulations, Xu!