posted on 2023-09-01, 15:10authored byVysakh Prasannan, Shahin Shemshian, Arinc Gurkan, Lakshmi Babu Saheer, Mahdi Maktab Dar Oghaz
Text-based Internet content is increasing at a very rapid rate day by day. As a result, even the best search engines are struggling to retrieve the exact expected results of users’ queries. On many occasions, the users’ expected result is embedded and scattered in a number of different documents and conventional search engines are unable to pinpoint it. To address this shortcoming, this study proposes a two-phased question answering system that utilizes a K-means clustering algorithm alongside the T5 deep encoder-decoder model to formulate a concise short answer to users’ queries. The proposed system has been trained using the Kaggle QA and SQuAD datasets and achieved the maximum F1-score of 0.564 and a minimum loss of 8.56.