Method for scoring correlated features in query expansion

Abstract

In this poster we describe experiments in information retrieval using a new method for scoring correlated features. This method uses information about word co-occurrences in the documents ranking high after the initial scoring to reduce combined scores of correlated words. We have experimented with this technique in conjunction with both simple Okapi scoring and a query expansion method using a probabilistic model, improving system performance in the context of TREC standardized tasks.