A query-based pre-retrieval model selection approach to information retrieval
Ben He, Iadh Ounis · 2004
In this paper, we propose a query-based pre-retrieval approach to the model selection problem, which automatically selects the best-performing retrieval model before the retrieval process takes place. In this approach, the queries are clustered according to their statistics and the bestperforming retrieval model is associated to each cluster. For a given new query, we assign the closest cluster to the query, and then we apply the model associated to the cluster. We evaluate the model selection approach on the disk1&2 of the TREC collections. The results show that our model selection approach achieves stable performance, which could outperform the use of the optimal retrieval model indi#erently for each query. The results also show that, interestingly, a retrieval model provides consistent performance for queries belonging to the same cluster.