An approach for summarizing hindi text using Restricted Boltzmann machine in deep learning

Abstract

Text summarization plays a crucial role nowadays due to large data available in day to day life. Reduced documents are useful and essential in the busy schedule of our lives. In this paper documents are summarized by four phases. They are preprocessing, feature vector generation, sentence score generation and summary generation. Sentence score is generated by Restricted Boltzmann machine (RBM) to improvise the result accuracy without losing the important information. Each of the sentence in the document undergoes all the phases and final summary generated is better as comparison among the existing (NN +fuzzy) and proposed method (Fuzzy + DL) with α value as 0.25 and β as 0.75. According to the analysis for precision rate at CR 30%, the Fuzzy+DL method is higher compared to the NN+fuzzy method (Fuzzy+DL)-0.875 and (NN+fuzzy)-0.256).The results shows the comparison graph for recall rate at 30%, the Fuzzy+DL method is higher compared to the NN+fuzzy method ((Fuzzy+DL)-0.777 and (NN+fuzzy)-0.6667) and the result also depicts the comparison graph for F-measure at CR 30%, the Fuzzy+DL method is higher compared to the NN+fuzzy method ((Fuzzy+DL)-0.8235 and (NN+fuzzy)-0.36363.