Login Search
FMDB Transactions on Sustainable Computer Letters Submit Paper
    Crossref logo

    Archiving

    Portico

    Fernando Martins De Bulhão is a member of

    Crossref logo

Automated Classification and Answer Extraction for Open-Ended and Closed-Ended Questions in Natural Language Texts

Authors:
S. Benila, Lekshmi Kalinathan, Athish Subba Reddi Ramanathan, K. Devi, Janaki Meena, Vidhula Sundhari Ganesh, Vijesh Varadharajan

Addresses:
School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, Tamil Nadu, India.

Abstract:

This work presents a hybrid question-answering (QA) system that can handle both open-ended and closed-ended forms of queries. Initially, the system utilises a BERT-based classification model to distinguish between these various types of questions. To establish a balanced training set, closed-ended questions are taken from the SQuAD dataset, and open-ended questions are taken from the QACoQA dataset. Both sets of questions are drawn from the same dataset.  Through the processing of closed-ended questions, a BERT model ensures strong performance on formal queries, which is necessary for the purpose of answer extraction. Open-ended questions, on the other hand, are processed by a hybrid BERT-BiLSTM model, which enables improved contextualization and the capture of longer-range effects of dependencies. A confidence-scoring technique that evaluates replies based on semantic relevance, logical coherence, and attention-based scoring is offered as a means of enhancing the reliability of the answers. Following an evaluation of the system's performance using the F1-score, Exact Match (EM), ROUGE, and the proposed confidence score, it is determined that the system demonstrates improved answer accuracy and validity. Although the model is extremely precise, additional optimisation is necessary for jobs that occur in real-time.  

Keywords: Question Answering; BERT and LSTM; Bidirectional LSTMs; Exact Match; Recurrent Neural Networks; Contextual Features; Natural Language Processing; Entity Features; Confidence Scoring.

Received on: 22/11/2024, Revised on: 29/01/2025, Accepted on: 10/03/2025, Published on: 05/09/2025

DOI: 10.69888/FTSCL.2025.000429

FMDB Transactions on Sustainable Computer Letters, 2025 Vol. 3 No. 3, Pages: 136-149

  • Views : 107
  • Downloads : 10
Download PDF