Experiments in Applying Information Flow Analysis in Query Expansion

In recent literature there have been a number of promising approaches proposed for estimating query language models Lavrenko & Croft estimate the query language model in terms of a Relevance Model [6]. The query Q is considered to be a random sample from the unknown relevance model R. R can be envisaged as a unknown process from which words can be sampled, so if query terms m q q , , 1 K have been sampled, what is the probability of term w will be sampled next. Essentially this model approximates the probability of observing w in the relevant set by the probability of co-occurrence between w and Q. They stated that “from a traditional IR perspective, our method is a massive query expansion technique”.