Single document extractive text summarization using Genetic Algorithms

Abstract

This paper presents an extraction based single document text summarization technique using Genetic Algorithms. A given document is represented as a weighted Directed Acyclic Graph. A fitness function is defined to mathematically express the quality of a summary in terms of some desired properties of a summary, such as, topic relation, cohesion and readability. Genetic Algorithm is designed to maximize this fitness function, and get the corresponding summary by extracting the most important sentences. Results are compared with a couple of other existing text summarization methods keeping the DUC2002 data as benchmark, and using the precision-recall evaluation technique. The initial results obtained seem promising and encouraging for future work in this area.