|
The TIDE project's goal was to facilitate timely and selective information access by developing tools that automatically disseminate and organize documents based on relevance to models of collaborators'
tasks. These
tools were developed with two goals in mind: first, improved utility
of delivered documents, because they are selected according to their
value added to a user's current task(s); second, reduced burden on the
user during the information delivery process, supporting effective
dissemination of information in situations involving time pressure.
To support these practical objectives, the technical objectives of the
project were new representational formalisms for task modeling, a
technique for deriving document queries from higher-level descriptions
of information needs, and methods for adaptation of relevance metrics
based on relevance feedback from the user.
The TIDE information management system works by matching documents to user task
representations, as follows. Each user interacts with an Information
Delivery Module (IDM) that runs on his machine. The IDM allows the
user to select tasks from the system's task model and to instantiate
those tasks to reflect the specific current situation. The IDM then
computes a document query from the instantiated task model and sends
that query to an information retrieval engine, which uses its
similarity metric to compute a match between incoming documents and
the query. Documents matching the query are displayed to the user in
a document browser.
TIDE's method of deriving information needs (queries) from task
descriptions is based on a model of the probabilistic relationships
between tasks and their subtasks, and the relationships between tasks
and keywords that occur in documents relevant to them. A key element
of the method is an original approach to computing the relative degree
of importance of one task to another. The components of the overall
model that TIDE uses to match documents to task
descriptions---specifically, the probabilities that relate tasks to
one another---can be difficult to get exactly right the first time.
To address this issue, the TIDE investigators have developed a model
for adapting the probabilistic relationships between tasks, based on
relevance feedback---input provided by the user about the relevance
of documents delivered by the system. The adaptation model is
consistent with TIDE's probabilistic method of deriving information
needs. A user's relevance feedback causes the system to modify its
beliefs about tasks' importance to one another, and that in turn
causes the system to change the probabilities relating tasks. The
adaptation approach allows us to have a system that is constantly
improving the quality of its information delivery.
An important strength of TIDE's task representation and query
derivation methods is that they support changing the criterion for
document relevance to a task as a user learns and draws conclusions
about the domain. When the user describes to TIDE how his belief in
the answers to subquestions has changed based on the evidence seen so
far, the query derived for the current task is modified to reflect
these new beliefs. Viewed more broadly, TIDE's approach recognizes
that information gathering occurs in a context---a context in which
the user is continually digesting information and reaching conclusions
based on it---and allows the information gathering process to take
that context into account.
| |
