Query-Oriented Summarization Based on Neighborhood Graph Model

In this paper, we investigate how to combine the link-aware and link-free information in sentence ranking for query-oriented summarization. Although the link structure has been emphasized in the existing graph-based summarization models, there is lack of pertinent analysis on how to use the links. By contrasting the text graph with the web graph. we propose to evaluate significance of sentences based on neighborhood graph model. Taking the advantage of the link information provided on the graph, each sentence is evaluated according to its own value as well as the cumulative impacts from its neighbors. For a task like query-oriented summarization, it is critical to explore how to reflect the influence of the query. To better incorporate query information into the model. we further design a query-sensitive similarity measure to estimate the association between a pair of sentences. When evaluated on DUC 2005 dataset. the results of the pro-posed approach are promising.