change in the light of new information. Over time,
we apply statistical methods to successive lists of que-
ry expansion terms and use the resultant evidence to
predict the degree of change (or development) in a
searcher's information need. Different degrees of per-
ceived change result in different interface responses.
The interface therefore offers two forms of sup-
port: the implicit selection of terms to expand the
query and an estimation of information need change.
Through implicitly monitoring interaction at the
results interface, searchers are no longer required to
assess the relevance of a number of documents, or
indeed consider entire documents for relevance.
Our approach makes inferences based on inter-
action and selects terms that approximate searcher
needs. It uses the evidence it gathers to track poten-
tial changes in information need and tailor the results
presentation to suit the degree of change in need.
Large changes in perceived information needs result
in new searches, but smaller changes result in less rad-
ical operations, such as re-ranking the list of retrieved
documents or re-ordering representations of the doc-
The main aim of our approach is to develop a
means of better representing searcher needs while
minimising the burden of explicitly reformulating
queries or directly providing relevance information.
Devising systems that adapt to the information needs
of those who use them is an important step in devel-
oping systems to help struggling searchers find the
information they seek.
We have explored a similar approach for image
retrieval. We have developed an interface (Fig-
ure 2) in which a user starts browsing with one
example image. Subsequently a new set of similar
images are presented to the user.
As a next step, the user through selecting one of
the returned documents updates the query, which
now consists of the original image and the select-
ed image from the set of returned candidates. After a
couple of iterations the query is based on the path of
In this approach, the emphasis is placed on the
user's activity and the context, rather than predefined
internal representation of the data. A path represents
a user's motion through information, and taken as
a whole is used to build up a representation of the
instantaneous information need.
In a nutshell, it supports both browse-based and
query-based approaches. It supports a query-less
interface, in which the user's indication of the rele-
vance of an image by selecting an image is inter-
preted as evidence of its being relevant to his current
information need. Therefore, it allows direct searching
without the need for formally describing the infor-
: A P
n the above applications, we were modelling
users' current information needs using interaction
between the system and the user. In the approach
below, we model users' long-term information needs
Figure 1: Interface of the Implicit Feedback based
Web search system
Figure 2: Image search interface