Abstract
Postgraduate supervisors are inundated with the routine task of
reviewing numerous students’ proposals, and students are often
fatigued from performing proofreading by themselves. Such a key
task in postgraduate supervision requires careful error-checking.
The study aimed at postgraduate supervision efficiency and sought
to improve the process for proofreading research proposals, a
typically "large document". The focus on understanding and
vetting large documents led to designing research artefacts using
Natural Language Processing (NLP) based Artificial Intelligence (AI)
techniques, particularly formal grammar. An AutoProofreader tool
that integrates the designed artefacts into a single software solution
for students and supervisors was developed during this research.
This study follows a Design Science Research (DSR) methodology
of five (5) methodological steps of problem diagnosis, objectives
definition, artefact(s) design, demonstration and evaluation of
designed artefacts, and communication with stakeholders for
feedback learning and reflection over designed artefacts. A
mixed-methods approach, combining qualitative and quantitative
investigations, was used to gain deep insights into postgraduate
supervision challenges in the problem diagnosis phase of DSR.
Rigorous checks on the tool’s accuracy, computed using a confusion
matrix at each design iteration, were conducted. Additionally, student
evaluation surveys and post-evaluation interviews with supervisors
were utilized to assess whether the developed artefacts improved
pedagogical experiences during the design, demonstration, and
evaluation phases.
The evaluation phase measured the tool’s performance against
defined criteria in meaningful extraction, recognition and parsing
of research proposal document sections and elements such as
title, author, supervisor, headings, paragraphs, research questions,
research objective, and page numbers. The AutoProofreader tool
was used to vet 80 research proposals of varying lengths, achieving
86% overall accuracy in assessing grammar and academic format
structure, including the recognition and placement of tables
and figures. The results confirmed significant improvements in
pedagogical processes experienced by students and supervisors. The
developed tool functions as an automatic proofreader for proposals
in the information systems domain and related disciplines. The
vii
implementation of this artefact has improved the turnaround time for
proofreading proposal manuscripts by students and their supervisors.
It benefits non-native English-speaking students and others who
would otherwise require the services of a paid human proofreader.
This study presents both a theoretical and practical approach
to solving complex real-world problems through an innovative
pedagogical tool. The tool provides structured (automated feedback),
domain-specific (designed for higher education), and AI-driven
(automatic proofreading) processes for vetting research proposals.
The responsibility for eliminating errors ultimately rests with the
human proofreader, however, an automated process can enhance the
effectiveness of a thorough scan of the research proposal to ensure
timely reviews. It contributes to ongoing research developments
in postgraduate supervision, promoting best practices in increasing
efficiency throughout the research journey. The research fits into
the Fourth Industrial Revolution (4IR) context providing such tools
helping to boost productivity, essential for assisting human experts
in the 21st century.