Work in Progress: Developing Disambiguation Methods for Large-Scale Educational Network Data

Adam Steven Weaver; Jack Elliott

Download Paper | Permalink

Conference: 2022 ASEE Zone IV Conference
Location: Vancouver
Publication Date: May 12, 2022
Start Date: May 12, 2022
End Date: May 14, 2022
Conference Session: Student Success and Interactions
Tagged Topic: Conference Submission
Page Count: 16
DOI: 10.18260/1-2--44760
Permanent URL: https://peer.asee.org/44760
Download Count: 52

Request a correction

Paper Authors

biography

Adam Steven Weaver Utah State University

visit author page

Adam Weaver is a B.S. Mechanical Engineering student at Utah State University. His research is focused on developing explicit disambiguation methods for large-scale social network studies. In addition, he works with applications of Particle Image Velocimetry (PIV), and wrote curriculum using PIV to teach energy conservation to high school students.

visit author page

biography

Jack Elliott Utah State University

visit author page

Jack Elliott is a concurrent M.S. in Engineering (mechanical) and Ph.D. in Engineering Education student at Utah State University. His M.S. research is in fluid dynamics including the application of PIV, and his Ph.D. work examines student collaboration in engineering education.

visit author page

Download Paper | Permalink

Abstract

This work in progress paper presents for review ongoing efforts developing and disambiguating large-scale interaction networks to improve engineering education research. Upon completion, we will present this work as a final paper at the ASEE Annual Conference 2023. The potential for social interactions to enhance engineering education is well founded in social learning theories, and research confirms many engineering students rely on their social networks in coursework, for persistence in engineering, and even as a part of their career choices. When decomposed, these networks are comprised of a variety of individuals, including peers, family members, etc. In addition, these networks vary significantly in size and traits (e.g., interaction frequency, reason for development, communication type). Because understanding should precede application, educators must develop a broad understanding of students’ interactions before they can effectively foster the growth of positive networks. Social Network Analysis (SNA) is a well-suited research method that quantitatively maps traditionally qualitative social networks, enabling the numerical representation of ties (interactions) between actors (individuals in a network). Further, researchers may plot these networks in graphs called sociograms. Together, these methods allow researchers to conduct statistical and visual analysis of relationships between networks and traits—like sub-network homophily vs. student grades. However, the difficulties associated with gathering accurate interaction data have veered those using SNA for engineering education toward oversimplified social environments. For example, researchers often observe students’ online interactions, where they can collect participant information concurrent with interaction data. Similarly, studies in face-to-face contexts are typically bounded to single classrooms, which confines the number of participants’ observed ties. Examining these easily monitored environments is a critical first step towards deploying SNA in educational fields but bars a complete depiction of students’ actual social networks. Hence, educators hoping to analyze authentic social networks will find existing research insufficient. To better understand these networks, our research group is currently conducting an SNA study comparing all participating freshmen and sophomore engineering students’ peer interactions to academic outcomes at a large, public land-grant university. This study produced over a thousand survey responses, and because the interaction data is open response, we found difficulty attributing many names to their correct nodes. Therefore, this paper presents our recommendations for best practices and ongoing efforts in disambiguating large scale educational network data. To begin, we organized the overarching network development task into discrete stages to filter responses according to unique name-ambiguity circumstances. These stages begin with simple spelling checks and end with a sub-network comparison process. To complete these stages, our first iteration relied on manual substitution, which yielded a best estimate of the complete network. This strategy proved effective in matching many ambiguous names with the correct node, but also highlighted procedures that we can refine through modern clustering methods. Therefore, we present a faster, automated approach using agglomerative hierarchical clustering. The results of this study demonstrate first steps toward more efficient and repeatable disambiguation methods for educational network data.

Citation
Format

Weaver, A. S., & Elliott, J. (2022, May), Work in Progress: Developing Disambiguation Methods for Large-Scale Educational Network Data Paper presented at 2022 ASEE Zone IV Conference, Vancouver. 10.18260/1-2--44760

TY - CPAPER
AB - This work in progress paper presents for review ongoing efforts developing and disambiguating large-scale interaction networks to improve engineering education research. Upon completion, we will present this work as a final paper at the ASEE Annual Conference 2023.
The potential for social interactions to enhance engineering education is well founded in social learning theories, and research confirms many engineering students rely on their social networks in coursework, for persistence in engineering, and even as a part of their career choices. When decomposed, these networks are comprised of a variety of individuals, including peers, family members, etc. In addition, these networks vary significantly in size and traits (e.g., interaction frequency, reason for development, communication type). Because understanding should precede application, educators must develop a broad understanding of students’ interactions before they can effectively foster the growth of positive networks.
Social Network Analysis (SNA) is a well-suited research method that quantitatively maps traditionally qualitative social networks, enabling the numerical representation of ties (interactions) between actors (individuals in a network). Further, researchers may plot these networks in graphs called sociograms. Together, these methods allow researchers to conduct statistical and visual analysis of relationships between networks and traits—like sub-network homophily vs. student grades. However, the difficulties associated with gathering accurate interaction data have veered those using SNA for engineering education toward oversimplified social environments. For example, researchers often observe students’ online interactions, where they can collect participant information concurrent with interaction data. Similarly, studies in face-to-face contexts are typically bounded to single classrooms, which confines the number of participants’ observed ties. Examining these easily monitored environments is a critical first step towards deploying SNA in educational fields but bars a complete depiction of students’ actual social networks. Hence, educators hoping to analyze authentic social networks will find existing research insufficient.
To better understand these networks, our research group is currently conducting an SNA study comparing all participating freshmen and sophomore engineering students’ peer interactions to academic outcomes at a large, public land-grant university. This study produced over a thousand survey responses, and because the interaction data is open response, we found difficulty attributing many names to their correct nodes. Therefore, this paper presents our recommendations for best practices and ongoing efforts in disambiguating large scale educational network data. To begin, we organized the overarching network development task into discrete stages to filter responses according to unique name-ambiguity circumstances. These stages begin with simple spelling checks and end with a sub-network comparison process. To complete these stages, our first iteration relied on manual substitution, which yielded a best estimate of the complete network. This strategy proved effective in matching many ambiguous names with the correct node, but also highlighted procedures that we can refine through modern clustering methods. Therefore, we present a faster, automated approach using agglomerative hierarchical clustering. The results of this study demonstrate first steps toward more efficient and repeatable disambiguation methods for educational network data.
AU - Adam Steven Weaver
AU - Jack Elliott
CY - Vancouver
DA - 2022/05/12
PB - ASEE Conferences
TI - Work in Progress: Developing Disambiguation Methods for Large-Scale Educational Network Data
UR - https://peer.asee.org/44760
DO - 10.18260/1-2--44760
ER -