Fusion analysis of information retrieval models on biomedical collections

A variety of endeavors have been made to improve the performance of traditional information retrieval models in biomedical domain. However, majority of the studies have focused on improving the performance of individual information retrieval models, while few attempts have been made to the investigation of combining multiple information retrieval models and exploring their interactions in biomedical information retrieval area. In this study, a comprehensive performance evaluation of seven popular generic information retrieval models is conducted on a biomedical literature collection. In addition, an information fusion method called the Combinatorial Fusion Analysis is applied to perform extensive combinatorial experiments on these information retrieval models. Our experimental results have demonstrated that a combination of multiple information retrieval models can outperform a single model only if each of the individual models has different scoring and ranking behavior and relatively high performance.