Complex Query Likelihood Retrieval Model

Abstract

In Chapter 3, we restricted the discussion to the family of query likelihood retrieval models that use simple smoothing methods based on a background language model. As a result of using simple smoothing methods, their efficiency is comparable to any traditional TF-IDF model. In this chapter, we review some extensions to these simple query likelihood retrieval models. These extensions often outperform, but also tend to be computationally more expensive than the simple models. All these improvements remain in the family of query-likelihood scoring, which distinguishes them from the other models to be reviewed in the next chapter; the latter uses language modeling in a different way than the query likelihood