Information Technology Reference
In-Depth Information
transition probabilities could be obtained by several methods, including: subjective
evaluation based on
expert opinions
,
survey
of target customers, and
measurement
of actual usage. UMMs can be constructed based on existing access logs, using a
combination of existing tools and internally implemented utility programs. How-
ever, as with most model construction activities, fully automated support is neither
practical nor necessary. Human involvement is essential in making various modeling
decisions, such as to extract UMM hierarchies and to group pages or links:
Not every higher-level state needs to be expanded into lower-level models,
because testing using lower-level model are to be performed selectively. There-
fore, a threshold should be set up so that only the ones above it need to be
expanded with their corresponding lower-level UMMs constructed.
For traditional organizations, there is usually a natural hierarchy, such as
university-school-department-individual, which is also reflected in their official
web sites. There are generally closer interconnections as represented by more
frequent referrals within a unit than across units. This natural hierarchy is used
as the starting point for the hierarchies in these UMMs for web testing, which
are later adjusted based on other referral frequencies.
For links associated with very small link probability values, grouping them to-
gether to form a single link would significantly simplify the resulting model, and
highlight the frequently used navigation patterns. A simple lower-level model
for this group can be obtained by linking this single grouped node to all those it
represents to form a one-level tree.
Web pages related by contents or location in the overall site structure can also
be grouped together to simplify UMMs.
Although any web page can be a potential entry point or initial state in an FSM
or its corresponding Markov chain, one basic idea in statistical testing is to narrow
this down to a few entry points based on their usage frequencies. The destinations
of incoming links to a web site from external sources or start-ups are the entry
points for UMMs. These links include URL accesses from dialog boxes, user book-
marks, search engine results, explicit links from external pages, or other external
sources. All these accesses were recorded in the access log, and the analysis result
for SMU/SEAS is summarized in the entry page report in
Table III
. For this web site,
the root page “
/index.html
” outnumber other pages as the entry page by a large
margin. In addition, these top entry pages are not tightly connected. These facts lead
us to build a single set of UMMs
[20]
with this root page as the main entry node to
the top-level Markov chain.
The issue with exit points is more complicated. Potentially any page can be the
exit point, if the user decides to end accessing the web site. That is probably why
