Department of Computer Science and Engineering

Guest Speaker

Button An Axiomatic Approach to Information Retrieval


Hui Fang
Computer Science
University of Illinois at Urbana-Champaign

Apr 12 2007 3:30PM
480 Dreese Labs
All interested parties are invited to attend.
Refreshments will be serviced in the lecture room

Abstract:

With the explosive growth of online information, we have an urgent need for powerful search engines to help manage and make use of all the information. How effectively search engines, such as Google, help us find useful information directly affects our productivity and quality of life. The effectiveness of any search engine is mainly determined by the underlying information retrieval model. A common deficiency of existing retrieval models is that there is generally no guarantee of the optimality of performance, and heavy parameter tuning is always needed to achieve optimal performance on a particular data set, which not only is labor-intensive, but also provides no guarantee of optimality on future queries.

In this talk, I will present a novel axiomatic framework for information retrieval. This new framework is fundamentally different from all previous retrieval frameworks in that relevance of documents to a query is captured with formalized retrieval constraints defined at the level of terms. I will present specific research results that demonstrate three benefits of such an axiomatic framework:

    (1) it can predict the performance of a retrieval function analytically without needing empirical experimentation;
    (2) it serves as a roadmap and provides guidance for developing new effective retrieval functions; and
    (3) it suggests a novel evaluation methodology that can diagnose strengths and weaknesses of retrieval functions.
The axiomatic framework opens up a new promising direction in studying information retrieval models. Using the framework, we have derived several new etrieval functions that are more effective and robust than existing retrieval functions. Moreover, the derived functions can be used in any retrieval applications to improve search accuracy.

Host: Feng Qin

* Hui Fang is a CSE faculty candidate

OSU logo