A comparison of text retrieval models

Many retrieval models have been proposed as the basis of text retrieval systems. The three main classes that have been investigated are the exact-match, vector space and probabilistic models. The retrieval effectiveness of strategies based on these models has been evaluated experimentally, but there has been little in the way of comparison in terms of their formal properties. In this paper we introduce a recent form of the probabilistic model based on inference networks, and show how the vector space and exact-match can be described in this framework. Differences between these models can be explained as differences in the estimation of probabilities, both in the initial search and during relevance feedback.