PKUTM participation at TAC 2011 Summarization Track

In this paper, we present details of the participation of PKUTM in the Guided Summarization and AESOP tracks at TAC 2011. For the Guided Summarization task, we develop two extractive summarization systems based on manifold ranking and Integer Linear Programming (ILP) respectively. In our first system, we score the sentences by linearly combining manifold ranking scores and scores based on other surface features. In our second system, we extend the traditional ILP approach to Tolerated ILP (T-ILP), where certain degree of redundancy is tolerated in the optimization process. We also introduce Wikipedia as domain knowledge into the concept weighting step of the T-ILP framework. For post-processing, we revise our sentence ordering scheme proposed in our last year’s participation. In the AESOP track, we submit for the no model task. We combine the Rouge scores by training regression models on the TAC 2010’s summarization evaluation data and apply the regression model to predict the TAC 2011’s system performance. Part I Guided Summarization Track