Multi-Document Summarization for Afaan Oromo using RNN

Abstract

Text Summary is a technique of producing short, fluent, and most significantly correct precis of a respectively longer textual content. There are two broad categories of methods to textual content summary: extraction method and abstraction method. The MDS is a system through which the content of many documents is summarized. We investigated the way of designing and developing multi-document summarization for AO using RNN and the way of tackling the problem of MDS for AO. The researcher used supervised machine learning, to train on manually created summaries by humans. We present the extractive MDS for AO using DL for which we used long short-term memory encoder-decoder with ROUGE evaluation metrics. For experimentation, 1800 datasets were used. During experiment, seven different scenarios were considered primarily based on the different epochs and batch size. The ROUGE consists of measures to robotically decide the best summary through evaluating it to reference summaries created by humans. It measures the quantity of overlapping units including both the n-gram and word pairs among system generated. This research achieved a performance of ROUGE-1 R = 0.27, P = 0.24 and f1-score= 0.26 and the ROUGE-2 R = 0.037, P = 0.034 and f1-score = 0.035. This work indicated that a LSTM algorithm was good and applicable to Afaan Oromo.