Improving Automated Group Assignments in an Academic Setting

Petra Bonfert-Taylor; Christopher Miller

Download Paper | Permalink

Conference: 2020 ASEE Virtual Annual Conference Content Access
Location: Virtual On line
Publication Date: June 22, 2020
Start Date: June 22, 2020
End Date: June 26, 2021
Conference Session: Faculty Development Research
Tagged Division: Faculty Development Division
Page Count: 13
DOI: 10.18260/1-2--34791
Permanent URL: https://peer.asee.org/34791
Download Count: 409

Request a correction

Paper Authors

biography

Petra Bonfert-Taylor Dartmouth College

visit author page

Petra Bonfert-Taylor is a Professor and an Instructional Designer at the Thayer School of Engineering at Dartmouth College. She received her Ph.D. in Mathematics from Technical University of Berlin (Germany) in 1996 and subsequently spent three years as a postdoctoral fellow at the University of Michigan before accepting a tenure-track position in the Mathematics Department at Wesleyan University. She left Wesleyan as a tenured full professor in 2015 for her current position at Dartmouth College. Petra has published extensively and lectured widely to national and international audiences. Her work has been recognized by the National Science Foundation with numerous research grants. She is equally passionate about her teaching and has recently designed and created a seven-MOOC Professional Certificate on C-programming for edX for which her team won the “2019 edX Prize for Exceptional Contributions in Online Teaching and Learning”. Previously she designed a MOOC “Analysis of a Complex Kind” on Coursera. The recipient of the New Hampshire High Tech Council 2018 Tech Teacher of the Year Award, the Binswanger Prize for Excellence in Teaching at Wesleyan University and the Excellence in Teaching Award at the Thayer School of Engineering, Petra has a strong interest in broadening access to high-quality higher education and pedagogical innovations that aid in providing equal opportunities to students from all backgrounds.

visit author page

author page

Christopher Miller Dartmouth College

Download Paper | Permalink

Abstract

Introduction In this research paper we introduce an automated method to form optimal, diverse student groups for active learning in academic settings, called the Group Assignment Tool (GAT). Input data for the GAT consists of student survey data, where the survey is tailored to the specific class based on learning goals and instructor preferences. The tool resolves the challenge of effective group formation by handling the optimization of groups automatically, while allowing instructors the freedom to define and rank the attributes that they find most important for successful groups. Instructors can select the relative importance of each attribute and also indicate whether to optimize for heterogeneity or homogeneity, avoid isolating students by any attribute, or fulfill student preferences for a given question. We show that our approach outperforms existing methods in both efficiency and effectiveness.

Background and Motivation Increasing numbers of college courses rely on group work. Group work can promote collaborative learning, improve information retention, and provide students with team skills which are vital for the modern workplace [Hansen 2006]. However, groups can also suffer from issues such as lack of shared scheduling availability, lack of diverse skill sets, and marginalization of at-risk students [Bacon 1999, Dasgupta 2015]. This means that forming effective groups - which we define as groups characterized by sufficient diversity of experience and knowledge, shared scheduling availability, lack of isolation of at-risk students, and fulfilled student preferences - is an important task. The three primary styles of group formation are self-selection (students choose their own groups), random assignment (students are sorted into random groups), and instructor selection (course instructors assign students to groups).

Literature suggests that random and self-selected groups are often ineffective [Feichtner 1984 , Bacon 1999]. Studies of such groups reveal that students are more likely to rate their group experience poorly, and indicate that performance issues arise with random and self-selected groups.

Instructor selected groups are the most effective way for producing positive group outcomes. However, manual group formation can be difficult and time consuming for the instructor. Group formation is a high dimensionality constrained optimization problem: maximizing heterogeneity of group members with respect to certain attributes while minimizing heterogeneity with respect to other criteria as well as adhering to various constraints. Such constraints might include avoiding the pairing of certain students or avoiding isolating certain students (for example by gender or ethnicity). Research supports the importance of such considerations [Sullivan 2018, Dasgupta 2015].

Method and Evaluation Automatic group formation is an optimization problem with large enough dimensionality so as to cause typical hill climbing algorithms (which can get caught in local extrema) to perform poorly. Prior tools attempt to address this issue by seeding the algorithm with a number of random initializations to optimize and evaluate. This approach is inefficient, and we introduce the use of epsilon greedy ascent and locally greedy initializations to address the issue more effectively and efficiently.

Evaluation of the GAT in undergraduate courses at [redacted] and analysis of results from simulated data show that the Group Assignment Tool is significantly more efficient than prior hill climbing algorithms. The GAT outperforms alternative methods of automated group formation, and produces groups which score within a small factor of optimal scores. These results will be presented through a combination of a traditional presentation and a mini-demonstration.

Citation
Format

Bonfert-Taylor, P., & Miller, C. (2020, June), Improving Automated Group Assignments in an Academic Setting Paper presented at 2020 ASEE Virtual Annual Conference Content Access, Virtual On line . 10.18260/1-2--34791

TY - CPAPER
AB - Introduction
In this research paper we introduce an automated method to form optimal, diverse student groups for active learning in academic settings, called the Group Assignment Tool (GAT). Input data for the GAT consists of student survey data, where the survey is tailored to the specific class based on learning goals and instructor preferences. The tool resolves the challenge of effective group formation by handling the optimization of groups automatically, while allowing instructors the freedom to define and rank the attributes that they find most important for successful groups. Instructors can select the relative importance of each attribute and also indicate whether to optimize for heterogeneity or homogeneity, avoid isolating students by any attribute, or fulfill student preferences for a given question. We show that our approach outperforms existing methods in both efficiency and effectiveness.

Background and Motivation
Increasing numbers of college courses rely on group work. Group work can promote collaborative learning, improve information retention, and provide students with team skills which are vital for the modern workplace [Hansen 2006]. However, groups can also suffer from issues such as lack of shared scheduling availability, lack of diverse skill sets, and marginalization of at-risk students [Bacon 1999, Dasgupta 2015]. This means that forming effective groups - which we define as groups characterized by sufficient diversity of experience and knowledge, shared scheduling availability, lack of isolation of at-risk students, and fulfilled student preferences - is an important task. The three primary styles of group formation are self-selection (students choose their own groups), random assignment (students are sorted into random groups), and instructor selection (course instructors assign students to groups).

Literature suggests that random and self-selected groups are often ineffective [Feichtner 1984 , Bacon 1999]. Studies of such groups reveal that students are more likely to rate their group experience poorly, and indicate that performance issues arise with random and self-selected groups.

Instructor selected groups are the most effective way for producing positive group outcomes. However, manual group formation can be difficult and time consuming for the instructor. Group formation is a high dimensionality constrained optimization problem: maximizing heterogeneity of group members with respect to certain attributes while minimizing heterogeneity with respect to other criteria as well as adhering to various constraints. Such constraints might include avoiding the pairing of certain students or avoiding isolating certain students (for example by gender or ethnicity). Research supports the importance of such considerations [Sullivan 2018, Dasgupta 2015].

Method and Evaluation
Automatic group formation is an optimization problem with large enough dimensionality so as to cause typical hill climbing algorithms (which can get caught in local extrema) to perform poorly. Prior tools attempt to address this issue by seeding the algorithm with a number of random initializations to optimize and evaluate. This approach is inefficient, and we introduce the use of epsilon greedy ascent and locally greedy initializations to address the issue more effectively and efficiently.

Evaluation of the GAT in undergraduate courses at [redacted] and analysis of results from simulated data show that the Group Assignment Tool is significantly more efficient than prior hill climbing algorithms. The GAT outperforms alternative methods of automated group formation, and produces groups which score within a small factor of optimal scores. These results will be presented through a combination of a traditional presentation and a mini-demonstration.

AU - Petra Bonfert-Taylor
AU - Christopher Miller
CY - Virtual On line
DA - 2020/06/22
PB - ASEE Conferences
TI - Improving Automated Group Assignments in an Academic Setting
UR - https://peer.asee.org/34791
DO - 10.18260/1-2--34791
ER -