AWARD: An Adaptive Web Based-Tool Prototype for Neurocognitive Individualized Assessment and Training - Foundations on Natural and Artificial Computation - page 543

3 Adaptive Web Tools
Traditionally, the concept of adaptive web tool has been associated more with
the domain of web personalization and computer education, using as a data
source the records server (logs), the set of metadata, databases, the structure of
the own tool and the actual domain knowledge, often expressed as ontology. The
rapid growth of Internet encompasses a growing range of activities, offering a
multitude of features to different and varied user profiles, each one with specific
needs that can not be satisfied on account of the static character of traditional
web applications.
From the point of view, both architectural and algorithms, the revision made
of different types of traditional approaches to development of the automatic
customization of systems reveals limitations [7]. As noted by this study, the
main disadvantage of these approaches is related to the concept of scalability
due to the huge amount of information to be stored of each user. For this reason,
the current review of the techniques used for the development of adaptive web
systems has been focused on data mining techniques.
Data mining is an essential step in the process of knowledge discovery in
databases ( KDD ) that consists of applying data analysis and discovery algo-
rithms that produce a particular enumeration of structures over the data [4].
In our case, by combining data mining techniques, classic statistics techniques
and knowledge gained directly from the domain expert, we have obtained a set
of temporal patterns representing the overall behavior of a user. During the ex-
ecution of a cognitive task in the tool, the detection of patient profile will take
place in a dynamic way, looking at the profiles database of patients previously
evaluated with a similar temporal evolution. In general, the concept of similar-
ity is essential in the area of data mining, and in our case, will be the basis on
which rests the adaptive nature of the proposed application. The idea of sim-
ilarity is the fact that it is not su cient to consider the equality or inequality
between objects (data). We need to consider how similar are two objects. The
significance of the similarity between two objects depends on the type of data
being treated. In one set of data you can see various types of similarities, there-
fore, different measures of similarity may reflect different aspects of the data set,
so depending on application of the similarity measure we have to analyze which
to choose.
One possible approach is in which the similarity between objects is defined
in terms of complementary concept of distance. A distance measure d ,must
be natural in some sense, and should describe the aspects of the data that we
are interesting to see. So, determining the most appropriate user profile can be
established by calculating the degree of similarity between sequences of events
using the edit distance algorithm. This distance is defined as the number of edit-
ing operations (insertion, deletion and/or modification) required to transform a
source sequence in a target sequence. To tailor the extent costs can be assigned
to each type of editing operation for each type of event, establishing a hierarchy
of types of events. To calculate the weighted edit distance between two sequences
of events, we used a version of the algorithm presented by Pirjo and Mannila [6].