Global ranking, a new information retrieval (IR) technology, uses a ranking model for cases in which there exist relationships between the objects to be ranked. In the ranking task, the ranking model is defined as a function of the properties of the objects as well as the relations between the objects. Existing global ranking approaches address the problem by "learning to rank". In this paper, we propose a global ranking framework that solves the problem via data fusion. The idea is to take each retrieved document as a pseudo-IR system. Each document generates a pseudo- ranked list by a global function. The data fusion algorithm is then adapted to generate the final ranked list. Taking a biomedical information extraction task, namely, interactor normalization task (INT), as an example, we explain how the problem can be formulated as a global ranking problem, and demonstrate how the proposed fusion-based framework outperforms baseline methods. By using the proposed framework, we improve the performance of the top 1 INT system by 3.2% using the official evaluation metric of the BioCreAtIvE challenge. In addition, by employing the standard ranking quality measure, NDCG, we demonstrate that the proposed framework can be cascaded with different local ranking models and improve their ranking results.
|出版狀態||已出版 - 2010|
|事件||23rd International Conference on Computational Linguistics, Coling 2010 - Beijing, China|
持續時間: 23 8月 2010 → 27 8月 2010
|???event.eventtypes.event.conference???||23rd International Conference on Computational Linguistics, Coling 2010|
|期間||23/08/10 → 27/08/10|