[Colloq] PhD Research Proposal - Alan Feuer, Friday 10:00am

[Rachel Kalweit]

PhD Research Proposal

Alan Feuer
Towards increasing conversation in search
Friday, June 23rd, 2006
10:00 AM
366 West Village H

Committee members:
Jay Aslam (advisor), Carole Hafner, Betty Salzberg, and Susan Dumais 
(Microsoft Research)

Abstract
Many people have observed, and controlled studies shown, that searching 
is inherently an iterative process. Yet log studies of Web-based search 
engines reveal that most search sessions consist of just one or two 
queries, and most queries consist of three or fewer words. We argue that 
lack of search engine transparency and poor search engine guidance 
partially explain the discrepancy.

In this talk we propose a model for increasing conversation in search 
based on three principles:
•	The search process should be transparent;
•	When a query fails to retrieve a satisfactory result, the search 
engine should suggest modified queries based on the nature of the failure;
•	Each suggestion should be accompanied by an indication of how 
successful taking the suggestion would be.

Queries commonly fail because they are too narrow or too broad, 
resulting in too few or too many hits. To enlarge the result set we will 
pre-search for all sub-phrases of the original query. To shrink the 
result set we will pre-search for all super-phrases of the query. These 
query sets can be very large. The number of sub-phrases is on the order 
of 2n for an n-term query and there is effectively no bound on the 
number of super-phrases. Our research will include development of 
algorithms to compute these queries in parallel.

The model will be tested by inserting the query modification algorithms 
into a search engine currently in active use on about 200 websites. Log 
files for the search engine will be analyzed before and after the 
changes are inserted to ascertain whether the algorithmic changes alter 
searcher behavior.