Web-based Learning Solutions for Communities of Practice
Web-based Learning Solutions for Communities of Practice
Web-based Learning Solutions for Communities of Practice
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
context within the target template. The result is a<br />
set <strong>of</strong> mappings (correspondences) relating facts<br />
from the source documents with the target template<br />
by encapsulating all necessary in<strong>for</strong>mation<br />
to trans<strong>for</strong>m source documents into instances <strong>of</strong><br />
the target template. In order to achieve this goal,<br />
we adopt a multi-criteria matching process. Each<br />
criterion will be represented by a service. These<br />
services are extensible. As new criterion becomes<br />
available, a new service is created. Examples <strong>of</strong><br />
services include: (1) Label matching: measures the<br />
similarities between entities <strong>based</strong> on the meaning<br />
inferred from their names (e.g. author and writer<br />
are synonyms). This service uses CoPs specific<br />
vocabularies (thesauri), local dictionaries (that<br />
define the common set <strong>of</strong> abbreviations, acronyms,<br />
and commonly used substring/short hand notations<br />
<strong>for</strong> a given CoP) as well as CoPs ontologies; (2)<br />
Constraint matching: relates entities <strong>based</strong> on their<br />
respective constraints. Such constraints include<br />
the use <strong>of</strong> Datatypes, occurrence constraints, etc.<br />
and (3) Structural matching: relates entities <strong>based</strong><br />
on the similarity <strong>of</strong> the structural context in which<br />
they appear. For this templates are modelled as<br />
trees (treated as a collection <strong>of</strong> paths). This has<br />
two advantages: (1) a match between two paths<br />
can be defined as long as there are some matching<br />
nodes along the paths; and (2) a match between<br />
two nodes can be defined irrespective <strong>of</strong> where<br />
they occur in the tree as long as there is at least a<br />
partial match between their paths.<br />
We have designed an algorithm to combine all<br />
the above criteria and produce a mapping result that<br />
clearly defines source and target mapped entities,<br />
required trans<strong>for</strong>mation operations, and conditions<br />
under which the mapping can be executed. Boukottaya<br />
(2004) summarizes an evaluation study<br />
<strong>of</strong> the proposed algorithm. The algorithm was<br />
evaluated in term <strong>of</strong> precision, recall, F-measure<br />
and overall using real world application: bibliographic<br />
date description. Globally it gives good<br />
result and only presents some limitations in some<br />
cases where complex matches are required. The<br />
proposed structural matching technique is <strong>based</strong> on<br />
20<br />
A Document Reuse Tool <strong>for</strong> <strong>Communities</strong> <strong>of</strong> <strong>Practice</strong><br />
the notion <strong>of</strong> node context. We define three kinds<br />
<strong>of</strong> contexts <strong>for</strong> a given node: the ancestor context,<br />
the child context and the leaf context. We show<br />
through a comparative study with other matching<br />
algorithms that the combination <strong>of</strong> these contexts<br />
highly improves the structural matching.<br />
Experience suggests that fully automated<br />
schema matching is infeasible, especially <strong>for</strong><br />
complex matches that involve complex trans<strong>for</strong>mation<br />
operations. For this, we designed an efficient<br />
user interface where the templates (source<br />
and target) are represented as trees and a set <strong>of</strong><br />
mapping operations have been made available to<br />
the user to modify the system mappings. Figure<br />
3 depicts the user interface used <strong>for</strong> mapping<br />
validation. Once the user validates the mapping,<br />
the system generates automatically the appropriate<br />
trans<strong>for</strong>mation scripts (XSLT programs). For this,<br />
we first have defined a model <strong>for</strong> structuring mapping<br />
results according to a mapping schema that<br />
describes the five dimensions <strong>of</strong> a mapping result:<br />
entity, cardinality, structural, trans<strong>for</strong>mation and<br />
constraint dimensions. Second we have designed<br />
an algorithm that generates automatically XSLT<br />
scripts <strong>based</strong> on the above mapping structure. For<br />
each matching node pair, the algorithm traverses<br />
the target template tree in a depth-first manner<br />
and generates progressively trans<strong>for</strong>mation rules.<br />
Templates, produced mapping results as well as<br />
the trans<strong>for</strong>mation scripts are stored. When new<br />
instances are available, the trans<strong>for</strong>mation process<br />
is done automatically without re-applying the<br />
entire matching process.<br />
Template-Driven Evolution<br />
In a community <strong>of</strong> practice context, templates<br />
continuously evolve to reflect a change in the<br />
practices, to adhere to new users’ requirements,<br />
to correct initial design errors, to allow expansion<br />
<strong>of</strong> the template scope over time or to simply allow<br />
<strong>for</strong> incremental maintenance. However, templates<br />
updates have a major consequence: documents<br />
being valid <strong>for</strong> the original template are no more