Personalized Approaches to Supporting the Learning Needs of Lifelong Professional Learners
Date
2019-03-10
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0002-8245-4887
Type
Thesis
Degree Level
Doctoral
Abstract
Advanced learning technology research has begun to take on a complex challenge: supporting
lifelong learning. Professional learning is an essential subset of lifelong learning that is more
tractable than the full lifelong learning challenge. Professionals do not always have access to
professional teachers to provide input to the problems they encounter, so they rely on their
peers in an online learning community (OLC) to help meet their learning needs. Supporting
professional learners within an OLC is a difficult problem as the learning needs of each
learner continuously evolve, often in different ways from other learners. Hence, there is a
need to provide personalized support to learners adapted to their individual learning needs.
This thesis explores personalized approaches for detecting the unperceived learning needs
and meeting the expressed learning needs of learners in an OLC. The experimental test bed
for this research is Stack Overflow (SO), an OLC used by software professionals. To date,
seven experiments have been carried out mining SO peer-peer interaction data. Knowing that
question-answerers play a huge role in meeting the learning needs of the question-askers, the
first experiment aimed to detect the learning needs of the answerers. Results from experiment
1 show that reputable answerers themselves demonstrate unperceived learning needs as
revealed by a decline in quality answers in SO. Of course, a decline in quality answers could
impact the help-seeking experience of question-askers; hence experiment 2 sought to
understand the effects of the help-seeking experience of question-askers on their enthusiasm
to continuously participate within the OLC. As expected, negative help-seeking experiences
of question-askers had a large impact on their propensity to seek further help within the OLC.
To improve the help-seeking experience of question-askers, it is important to proactively
detect the learning needs of the question-answerers before they provide poor quality answers.
Thus, in experiment 3 the goal was to predict whether a question-answerer would give a poor
answer to a question based on their past peer-peer interactions. Under various assumptions,
accuracies ranging from 84.57% to 94.54% were achieved. Next, experiment 4 attempted to
detect the unperceived learning needs of question-askers even before they are aware of such
needs. Using information about a learner’s interactions over a 5-month period, a prediction
was made as to what they would be asking about during the next month, achieving recall and
precision values of 0.93 and 0.81. Knowing the learning needs of question-askers early
creates an opportunity to predict prospective answerers who could provide timely and quality
answers to their question. The goal of experiment 5 was thus to predict the actual answerers
for questions based only on information known at the time the question was asked. The
iv
success rate was at best 63.15%, which would only be marginally useful to inform a real-life
peer recommender system. Thus, experiment 6 explored new measures in predicting the
answerers, boosting the success rate to 89.64%. Of course, a peer recommender system
would be deemed to be especially useful if it can provide prompt interventions, especially to
get answers to questions that would otherwise not be answered quickly. To this end,
experiment 7 attempted to predict the question-askers whose questions would be answered
late or even remain unanswered, and a success rate of 68.4% was achieved.
Results from these experiments suggest that modelling the activities of learners in an OLC is
key in providing support to them to meet their learning needs. Perhaps, the most important
lesson learned in this research is that lightweight approaches can be developed to help meet
the evolving learning needs of professionals, even as knowledge changes within a profession.
Metrics based on the experiments above are exactly such lightweight methodologies and
could be the basis for useful tools to support professional learners.
Description
Keywords
Lifelong learning, education data mining, professional learning, personalization, user modelling, online learning community
Citation
Degree
Doctor of Philosophy (Ph.D.)
Department
Computer Science
Program
Computer Science