Relevance Models in Information Retrieval

We develop a simple statistical model, called a relevance model, for capturing the notion of topical relevance in information retrieval. Estimating probabilities of relevance has been an important part of many previous retrieval models, but we show how this estimation can be done in a more principled way based on a generative or language model approach. In particular, we focus on estimating relevance models when training examples (examples of relevant documents) are not available. We describe extensive evaluations of the relevance model approach on the TREC ad-hoc retrieval and cross-language tasks. In both cases, rankings based on relevance models significantly outperform strong baseline approaches.