Two term-layers: an alternative topology for representing term relationships in the Bayesian Network Retrieval Model

The Bayesian Network Retrieval Model presents the advantage that may capture the main relationships among the terms from a collection by means of a polytree, network that allows the design and application of efficient learning and propagation algorithms. But in some situations where the number of nodes in the graph is very high (the collection represented in the Bayesian network is very large), these propagation methods could be not so fast as needed by an interactive retrieval system. In this paper we present an alternative topology for representing term relationships that avoids the propagation with exact algorithms, very suitable to manage large document collections. This new topology is based on two layers of terms. We compare the retrieval effectiveness of the new model with the old one using several document collections.