Information Technology Reference
In-Depth Information
Table 12.1
Example of data
used in sequential pattern
mining
Sequence 1
<b,a,d,b>
Sequence 2
<a,b,b,d,c>
Sequence 3
<a,c,d>
Sequence 4
<a,c,c,d>
Generalized Sequential Pattern algorithm and applied it to collaborative learner data.
The context was a senior software development project where students worked in
teams of five to seven students and interacted over a collaboration tool comprising of
a wiki, a task allocation system and a subversion repository. Perera and colleagues
searched for sequential patterns within groups as well as within individuals and
classified the results according to the “quality” of the group, based on marks, and on
teachers' assessment of the group. Results pointed to the importance of leadership
and group interaction, and they were able to identify patterns indicative of good
and poor individual practice. An example of interesting patterns found was that
well-performing groups used more heavily the task allocation system than the wiki,
whereas weaker groups displayed the opposite characteristic.
Before forms of data mining such as just described can be conducted without
human intervention, numerous decisions pertaining to issues such as data granular-
ity (what should be considered to be a “event”?), model granularity (on which level
to express learning and change: time series, event sequences, holistic structures such
as textual summaries?), and regarding the tuning of the many parameters data min-
ing methods come with need to be decided. Many of these decisions will depend on
the
purpose
of the data analysis, of course constraint by the kind of data and the pro-
cessing resources available. For instance, one might make different decisions when
the purpose is monitoring of students' learning instead of finding a theoretically
elegant description of regularities in the data.
Viewing individual or collaborative learning in terms of sequences of events
makes few assumptions about the nature of the learning situation. For instance, we
can look for sequences in classroom interactions as well as in informal learning situ-
ations. However, in situations where students' learning activities can be expected to
follow at least partially a predefined structure, such as a group collaboration script
(Kollar, Fischer, & Hesse, 2006), we can take into account this information by using
process models representations of higher granularity, such as Finite State Machines
or Petri Nets. Reimann, Frerejean, & Thompson (2009) for instance, made use of
discreet event modeling techniques to represent group decision-making processes
from chat log data in form of dependency graphs as illustrated in Fig. 12.3. The
arcs on the right side of the boxes that point back at their own box indicate loops,
meaning that statements of this type often occurred multiple times in a row. The
numbers along the arcs show the temporal dependency of the relationship between
two events, with the arrow pointing from a to be meaning that a is followed by b.
The second number indicates the number of times this order of events occurred. The
numbers in the boxes indicate the frequency of the respective event type.
Generalizing from this example, process mining approaches as developed so far
mainly for enterprise computing could also play a role in educational e-research
