Context-Aware Web Search

1
Context-Aware Web Search Using
Dynamically-Weighted Information Fusion Nicole
Anderson Advisor Bob Kessler
Introduction
Information Fusion Techniques
As search has become a cornerstone tool for
nearly everyone with Internet access, users are
often frustrated with not being able to find what
they need as quickly as possible. Our thesis is
that if we can determine what a user is doing at
the time of the search request, we can customize
the results based on their context and will
provide that user with more accurate answers.
Our innovation is in three main areas type of
context, query enhancement, and information
fusion. For context, we are investigating the
use of personal calendar/PIM information to help
determine what a user is currently doing their
physical location personal preferences personal
vocabulary and recommendations gathered from
co-workers. Query enhancement is taking the
users intended query and enhancing it with
relevant contextual information to create a more
focused query. Lastly, information fusion is how
we manipulate the results returned from the
search engine again with the contextual data.
Choosing how to fuse context data is as important
as selecting which context to utilize and is a
key element of our research. Our hypothesis is
that the marriage of these three techniques will
result in a search experience that is more
valuable than current tools.  

• It must be determined how best to combine the
  data to use it effectively. We call the process
  of combining context data from many sources
  dynamically weighted information fusion.
• We are considering several data fusion methods
• For the initial prototype system, a sum of
  products equation was used for fusing context
  data Pages Rank ? Pi Wi
• Another technique we would like to evaluate is
  using Bayesian analysis for utilizing context
  data. With this approach, we expect search
  results to improve over time.
• Other models will also be researched and
  considered.

Analysis
We will conduct a user study in which users
evaluate the search performance of USearch versus
commercial search engines. Evaluation will be
based on whether each of the top 10 results for a
search are classified as relevant or non-relevant
by the user. ROC curves will be used to
determine if the difference between the two sets
of results is statistically significant.
USearch
USearch
Query Enhancement
Web Query
GPS
Context
Reordered Results
Data Fusion
Context
Sports R Computers R Cycling
For addition information nanderso_at_cs.utah.edu