Skip to Main Content
Graph ranking algorithms have been successfully used in multi-document summarization. Among them, the basic link analysis model has drawn much attention due to its' mutual reinforcement principle which appears to be sound for the generic summarization task. In this paper, we explore effective strategies for extending the basic link analysis model to question-oriented multi-document summarization. Three kinds of strategies, namely link re-weighting, baseset downsizing and projection, are proposed to introduce question-dependent similarity metric, adjust the node number and refine the ranking process respectively. Experimental results evaluated on the DUC data sets demonstrate that these three strategies can achieve better results.