Page 206 -

Knowledge Application 189

Trace 2
Visitor 3
Visitor 4 Trace 3
Visitor 2 Trace 1
Visitor 5

Web server

Visitor 1 Visitor 6

Instead of Web-centric: profiling User-centric profiling

Figure 6.3
Alternative approach to personalization

organization. For example, push technologies are based on user models that look at
historical information requests in order to push or automatically send out similar new
content that becomes available.
We will need to be able to fi nd and use content based on an individual ’ s personal
model, that is, how they perceive the knowledge world around them. This is often
infl uenced by their particular background (e.g., IT vs. sociology), how long they
have been in the company, how expert they are in the topic as well as a whole
spectrum of preferences ranging from the linguistic to the format they prefer to receive
knowledge (e.g., visual types of people who prefer diagrams, or those who prefer to
read text). These are often represented as semantic networks (see fi gures 6.4 and 6.5 )
There are also systems that monitor users ’ tasks online and interpret them in
context, based on traces they leave behind. These systems work well for tasks that are
well identifi ed and where knowledge can be described in a clear ontology (e.g., a postal
address template). In general, this approach is based on a user interacting with a
computer system to perform a task that leads to changes in the system. An observer
agent (a software routine) observes these changes according to an observation model
to generate a log or trace of what the user has done. The trace is then analyzed to
identify and extract signifi cant episodes, and interpret them according to explained
task signatures. Each episode represents a pattern and each pattern can be mapped
onto a task, a subtask, or a more specifi c step that forms part of the subtask. For
example, if the user is trying to locate, open, and print out a particular fi le, there are
three distinct episodes that can be identifi ed: behaviors related to locating, opening,