SALS-SIG Research Seminar

Experiments with probabilistic models in information retrieval

Abstract:

In this paper, I will give an overview of the probabilistic approach to text retrieval, and indicate some of the work undertaken in this area at City University in the last few years. Much of this work is concerned with relevance feedback, in which the user is invited to mark as relevant one or more items retrieved in a first search, and the system then attempts to find similar items. The probabilistic models involved are generally fairly simple, but are typically used with a very small amount of linguistic analysis, possibly some subject knowledge, a great deal of pragmatism, and a great deal of formal experimentation. I will attempt to convey the flavour of this mix, and will look at the implications of these ideas for retrieval involving other media than text.

The major experimental effort of recent years is the international TREC (Text REtrieval Conference) programme. This is a fascinating exercise to take part in, and is very useful for answering some types of questions, but is not so good for aspects which directly involve the user. I will describe some of the issues involved in experimental work on retrieval.

Enquiries: sals@mri.mq.edu.au