Information Technology Reference
In-Depth Information
this section we will focus on some specific methodological challenges in learning
technology research and discuss emerging approaches and technologies helping to
address these issues.
When Data is Flowing in Streams
With the ever-increasing use of mobile and web technologies for learning in- and
outside of classrooms researchers gain potential access to quasi-continuous data
“streams.” This kind of data is invaluable given the current lack of process data that
is available on a level close to students' learning activities. However, to capitalize on
this potential a number of hurdles need to be overcome. Some of these are technical
(how to represent, store, and access the data), some legal (data privacy, etc.), and
some methodological: how to analyze process data, and along with that: how to the-
orize learning and development processes in pedagogical and psychological terms.
We trust engineers to overcome the technical hurdles, and law and policy makers
to deal with issues of privacy and governance. The methodological challenges we
cannot delegate to others, however.
The methods potentially useful to analyze process data of individual and group
learning as learners interact with technologies are too manifold to be surveyed here
(see Langley, 1999; Olson, Herbsleb, & Rueter, 1994; Reimann, 2009 for a recent
review specific to CSCL). Instead, we will focus on methods that are fully com-
putational (or at least for which automatization is feasible) and are hence prime
candidates for the (quasi-continuous) analysis of very large amounts of (quasi-
continuous) data that are increasingly available in form of log files, for instance,
from web-based learning environments and from “immersive” environments such
as Second Life. For such amounts of data, stemming from learning situations that
are rarely experimentally controlled, inductive data mining methods are helpful.
Data mining algorithms allow to analyze vast amounts of electronic traces, com-
bined or not with complementary information (such as exam marks), and discover
patterns that are not visible to the human eye (Ye, 2003). An important aspect of
learner data is the
temporal aspect
of its events. A particular interesting question for
understanding the quality of learning is whether there are some sequences of events
that are more frequent within successful learners than within less successful ones
(or vice versa). Sequential pattern mining (Srikant & Agrawal, 1996) is an example
of algorithm that considers this temporal aspect and detects frequent sequences of
items (here, events) appearing in a sequence data set a minimum number of times
(called
support
). For example in the example given in Table 12.1, the sequence
<a.d> has a support of 4 as it appears in all four sequences whereas <b,d> has a
support of 2.
Applied to collaborative data, items such as a, b, c, and d become a representation
of an atomic and multidimensional event (which typically would have the event's
author, the time, the type of intervention, and possibly more information). Perera,
Kay, Koprinska, Yacef, & Zaiane (2008), for instance, have built a variant of the
