Graph-Based Fuzzy Logic for Extractive Text Summarization (GFLES)

Abstract

High accuracy models for automatic text summarization are increasingly required due to the overwhelming growth of available text on the internet. In this paper, a new unsupervised approach for Automatic Text Summarization (ATS) employing Fuzzy Logic (FL) to optimize summary generation is proposed. The approach is extractive, graph-based and is applicable for both single and multiple documents. Our Graph-based Fuzzy Logic Extractive Text Summarization (GFLES) approach consists of five main stages: text pre-processing, text representation using graph model, features extraction during graph construction, sentence clustering using our previously developed Graph-based Growing Self-Organizing Map (G-GSOM) and lastly, sentences are ranked using FL and a summary is generated. GFLES is distinctive from other approaches by its four advantages: 1) employment of the graph model for features extraction, 2) a better substitute of Vector Space Model (VSM), 3) appreciating the importance of sub-topics of text by taking them into account before generating the summary by clustering sentences and 4) applicable to both single and multiple documents. Moreover, the use of FL on each cluster at the same time provides a foundation for further development of GFLES to be applied on big data. Using the DUC 2004 dataset, our experimental results demonstrate that GFLES is able to generate comprehensive text summaries. These results establish the potential of GFLES as a high accuracy model for automatic text summarization.