A framework for extractive text summarization using semantic graph based approach

Abstract

Automatic extractive text summarization finds the subset of the data, which represent the most salient information of the entire document. Now-a-days in the era of Internet, it is a demand of the users to understand the huge amount of texts within a very short time, as there are lots of redundant unnecessary textual information in the webpages. Here comes the value of automatic extractive text summarization. Internet users get the idea of entire content of the document by reading summary generated by the summarizer within a very short time and also can decide whether to read the entire content or not. There are several methods for generating extractive summary such as clustering based, optimization based, information and itemset based, and word embedding based methods. These methods fail to determine the redundancy generated from the summary sentences. Therefore, in this paper, we propose a novel approach for generating extractive summary by considering semantic relationships among the sentences of the document.Here, we extract the Predicate Argument Structure (PAS) from each sentence and measure semantic distance among the sentences is measured from the PAS to PAS semantic relationship of the sentences. Then, we apply modified page rank algorithm to rank the sentences. Subsequently, we employ Maximum Marginal Relevance (MMR) method to rerank the sentences. Accordingly, we propose top n sentence selection algorithm is proposed to select summary sentences. To evaluate the performance of our proposed method, we adopt Rouge-1, Rouge-2 evaluation metrics for computing precision, recall, and f-score of the summary. Furthermore, we also employ cosine simillarity measurement based approach to determine redundancy in the summary sentences. Our evaluation results demonstrate that the proposed approach achieves improved performance compared to the other existing methods available for generating an the extractive summary.