TEXT SUMMARIZATION: USING CENTRALITY IN THE PATHFINDER NETWORK

Kaustubh R. Patil, Pavel Brazdil

2007 · 11 citations · 17 references

Concepts

Abstract

We present a graph theoretic technique for automatic text summarization aimed at producing extractive summaries of a single document. In our system, called as SumGraph, text is represented as a graph with sentences as nodes while weights on the links represent intra-sentence dissimilarity. Novelty of our approach lies in the use of Pathfinder Network Scaling (PFnet) technique representing conceptual organization of the text which in turn is used to compute importance of a sentence in the text. Importance of a sentence is defined using its centrality in the PFnet. Use of Latent Semantic Analysis (LSA) is also investigated. PFnet and LSA have been shown to model human aspects of semantic memory and linguistic acquisition respectively. We show that SumGraph performs better than other systems on DUC2001 and DUC2002 datasets for generating 100 word summaries, when compared using ROUGE-1 scores. We also show that SumGraph is different than other methods using a non-parametric statistical test.

References

17