An Improvised Extractive Approach to Hindi Text Summarization

Abstract

Text summarization is defined as a task of minimizing a text that is produced from one or more texts such that the actual significant information in the texts is not lost. A text summarization tool compresses the text and displays only the important content to the user. Using text summarization, decisions can be made in lesser time and the core of the document be understood. This paper emphasizes on an extractive approach and its implementation on Java. The extractive approach selects the significant sentences based on a thematic approach. Before selecting the thematic words the Hindi stop-words was removed and also the stemming process to retrieve the root words in the sentences under consideration. Stop-word elimination eliminates the semantically null words from the input document and stemming helps in clustering together words with the same radix term. The system is based on an algorithm for scoring the sentences based on occurrence of the radix of thematic words. The sentences with highest score are added to the summary. The generated summary is further processed based on removal of extraneous phrases from the previously selected summary sentences so as to bring the sentences closer to human generated summary. The testing of the accuracy of the system can be made by using a technique called The Expert Game. In expert game, experts underline and extract the most interesting or informative fragments of the text. The recall and precision of the system’s summary is measured against the human’s extract. Based on the testing, the system is found to be 85 % accurate.