Extractive Text Summarization for Wolaytta Language Using Recurrent Neural Network

Information overload is caused by an increasing number of overflowing information sources due to the availability of both online and offline information that doubles over time. Therefore, it is extremely challenging and time-consuming to filter out the information that is relevant from this abundant supply of information. Text summarization is a concept that can help ease this load, which has received a lot of attention in recent NLP studies. Wolaytta language has been written since the Sudan Interior Mission, in the 1940s. The language is utilized in teaching and learning in elementary, secondary, and higher education institutions, as well as processing information transfer in various mass media. But for the advancement of the language, to the best of the researcher’s knowledge, despite the fact that there are numerous deep learning-based extractive summarizer models for foreign languages, no deep learning-based text summarization research work has been done for Wolaytta. Therefore, the researchers planned, gathered, and produced 424 text documents and reference summaries for each document in the Wolaytta language. When the datasets were smoothed out, an RNN-based Wolaytta text summarizer was trained and then a relevant model summary was created. The outcome is evaluated using both subjective and objective evaluation techniques. In this research paper, the recall, precision, and f-measure values are 83.3%, 80.4%, and 82.7%, respectively. The research on the Wolaytta language conducted in the past used 92 paragraphs, and the findings revealed 61.16%, 60.69%, and 60.46% recall, precision, and f-measures, respectively. But by using more datasets, doing more preprocessing tasks, using a better strategy, and with better-recorded performance, this work achieves better results than the earlier work. Also, we made a ready-built Wolaytta text dataset available on Kaggle as a pointer for future investigations.