Information Technology Reference
In-Depth Information
groups. Each user group represents a significant and large enough group of users
sharing several properties together. Based on these user groups should be controlled
the functions of the e-shop. The identification of the user groups can be done based
on the users navigation behavior. The user's action while browsing a web site should
be observed and should be used to learn the user profiles.
The users interest may change over time. Therefore the user model should adapt to
this concept drift. A recent trend separates the user model into a short-term and a long
term user model
. The short-term user model is based on highly specific
information, whereas the long-term user model is based on more general information.
The short-term model is learned from the most recent observations only. It represents
user models, which can adjust more rapidly to the user´s changing interests. If the
short-term model can not classify the actual user at all, it is passed on to the long-term
model which represents stereotypical user groups
13
14
. The purpose of the long-term
model is to model the user's general preferences for certain products that could not be
classified by the short-term model.
19
5.2.3 Mining for the User Model
Webb et al.
18
summarize four major issue in learning user models:
The need for large data sets;
The need for labeled data;
Concept drift, and
Computational complexity.
The problem of the limited data set and the problem of concept drift has lead to
hybrid user models separated into a short-term and a long-term user model.
Most applications use the nearest neighbor method to model the short-term user
model. This method searches for similar cases in a data base and applies the action
associated to the nearest case to the actual problem. A specific problem of this method
is the selection of the right attributes that describe the user profile and/or the set up of
the feature weights
13
as well as the complexity. Bayesian classifiers are used for
the long-term model
.
We intend to use incrementally decision tree induction to learn both user models;
the short-term and the long-term user model. It allows us to use the same development
strategy for learning the models in both cases. This can be an important system
feature. To overcome the limited data set problem we use boosting for building the
short-term model. Decision tree induction can be used to learn the classification
model as well as to cluster data. In contrast to nearest neighbor methods, decision
tree induction generalizes over the data. This will give us a good understanding of the
user modeling process
13
.
Decision tree induction allows one to learn a set of rules and basic features
necessary for the user modeling. The induction process does not only act as a
knowledge discovery process, it also works as a feature selector, discovering a subset
of features from the whole set of features in the sample set that is the most relevant to
the problem solution. A decision tree partitions the decision space recursively into
5
