Eigen Analysis Based Document Summarization

This paper presents an effective method of summarizing a text document using the Eigen analysis of the sentence semantic similarity matrix, assuming the graph-based document model. The semantic similarity between sentences is identified using the similarity information extracted from Wordnet. Eigen analysis reveals the candidate sentences to be chosen for the extractive summary. The sentences that make up the document summary are extracted by prioritizing the Eigen values of the semantic similarity matrix of each paragraph. The proposed method was evaluated over sets of research papers, short stories, news articles, etc., The ROUGE score obtained shows that the proposed system generates summaries of acceptable quality.

I. Introduction

The objective of this work is to describe an algorithm to summarize text documents by constructing inter-sentence similarity/correlation matrix of each paragraph in order to identify the most important sentences in a document. The work mainly consists of the following steps: 1)

Construction of inter-sentence correlation matrix for each paragraph,

2)

Ranking the sentences of each paragraph,

3)

Extracting important sentences from each paragraph based on Eigen values, and

4)

Composing the summary of the document using the candidate sentences.