Adaptive Information Filtering - Text Mining: Classification, Clustering, and Applications - page 191

does not know these or thinks he/she needs to know one thing but actually
needs something else. For example, a financial analyst may search for news
in order to check whether the earnings of a company matches the projected
earnings. However, also relevant to this task is the large number of customer
complaints about the company's product in the blog space. Another example
is a research scientist often wants to keep up-to-date with what is happening
within a research field, but not looking for a specific answer.
If the information need of a user is more or less stable over a long period
of time, a filtering system is a good environment to learn user profiles (also
called user models) from a fair amount of user feedback that can be accu-
mulated over time. In other words, the adaptive filtering system can serve
the user better by learning user profiles while interacting with the user, thus
information delivered to the user can be personalized to an individual user's
information needs automatically. Even if the user's interest drifts or changes,
the adaptive filtering system can still adapt to the user's new interest by
constantly updating the user profile from training data, creating new classes
automatically, or letting the user create/delete classes.
Adaptive filtering vs. collaborative filtering: Collaborative filter-
ing is an alternative approach used by push system to provide personalized
recommendations to users. Adaptive filtering, which is also called content
based filtering, assumes what a user will like is similar to what the user liked
before, and thus make recommendations for one user based on the user's feed-
back about past documents. Collaborative filtering assumes users have similar
tastes on some items may also have similar preferences on other items, and
thus make recommendations for one user based on the feedback from other
users that are similar to this user. Memory-based heuristics and model based
approaches have been used in collaborative filtering task (29) (22) (10). This
chapter does not intend to compare adaptive filtering with collaborative fil-
tering or claim which one is better. We think each complements the other.
Adaptive filtering is extremely useful for handling new documents/items with
little or no user feedback, while collaborative filtering leverages information
from other users with similar tastes and preferences in the past. Researchers
have found that a recommendation system will be more effective when both
techniques are combined. However, this is beyond the scope of this chapter
and thus not discussed here.
Adaptive filtering vs. Topic Detection and Tracking: The super-
vised tracking task at the Topic Detection and Tracking (TDT) Workshops
is a forum closely related to information filtering (1). TDT research focuses
on discovering topically related material in streams of data. TDT is different
from adaptive filtering in several aspects. In TDT, a topic is user independent
and defined as an event or activity, along with all directly related events and
activities. In adaptive filtering, an information need is user specific and has a
broader definition. A user information needs may be a topic about a specific
subject, such as “2004 presidential election,” or not, such as “weird stories.”
However, TDT-style topic tracking and TREC-style adaptive filtering have