Extractive Text Summarization: A Graph based Sentence Scoring Approach

Abstract

Text Summarization is an emerging field in Natural Language Process (NLP) domain. It is very full for summarizing full-length document automatically. We have proposed the statistical model for the text summarization process. Because of features like easy-to- use, less complex, low computation resources requirement, we have chosen the extractive text summarization model over the abstractive model. We have also used a graph-based model to establish relations between sentences, and mathematical operations to establish a relation between words and sentences. We have used documents from the “WiKiHow” dataset to test our model. We have also demonstrated the effect of the sentence clustering method by including the graph clustering stage in the post-processing phase. We have used the GraphFrame module from the Apache Spark environment as it has the ability to parallel the execution of graph-based operations. To evaluate our model we have used Recall-oriented Understudying Gisting Evaluation (ROUGE) parameters. Based on the obtained results, it is clear that graph clustering is dependent on the type of documents provided in the dataset.