The notion of query performance prediction is straightforward. It refers to inferring the retrieval performance of a given query.
The query performance prediction methods can be classified into two categories: post-retrieval prediction and pre-retrieval prediction.
In a post-retrieval prediction process, the prediction happens after the first-pass retrieval. In particular, it involves the use of relevance scores. Following are some popular post-retrieval prediction methods:
SVR for Query Performance Prediction using post-retrieval query features
The pre-retrieval performance prediction uses the descriptive features of a given query, such as the idf factors and the composing query terms, to predict the query performance. The prediction process happens before the first-pass retrieval takes place, and does not have the need for relevance score. A list of pre-retrieval query performance predictors are introduced in:
B. He and I. Ounis. Inferring Query Performance Using Pre-retrieval Predictors. In Proceedings of the SPIRE 2004. Pages 43 - 54. Padova, Italy. October, 2004.
V. Plachouras, B. He and I. Ounis. University of Glasgow at TREC2004: Experiments in Web, Robust and Terabyte tracks with Terrier. In Proceedings of the 13thText REtrieval Conference (TREC2004). Gaithersburg, MD. November, 2004.
Moreover, the SVR for Query Performance Prediction using pre-retrieval query features is applied in:
K. Kwok, L. Grunfeld, H. Sun and P. Deng. REC 2004 Robust Track Experiments Using PIRCS. In Proceedings of the 13thText REtrieval Conference (TREC2004). Gaithersburg, MD. November, 2004.