International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 12 | Dec 2024
p-ISSN: 2395-0072
www.irjet.net
Progress and Innovations in Question Answering Systems: An Extensive Literature Review Arti Karche1, Amruta Deokate2, Sejal Talekar3 Anushka Shah4 1Student, Dept. of Computer Engineering, VPKBIET Baramati, Maharashtra, India 2Student, Dept. of Computer Engineering, VPKBIET Baramati, Maharashtra, India 3Student, Dept. of Computer Engineering, VPKBIET Baramati, Maharashtra, India 4Student, Dept. of Computer Engineering, VPKBIET Baramati, Maharashtra, India
---------------------------------------------------------------------***------------------------------------------------------------------
Abstract - Traditional question-answering systems struggle
with interpreting semantically rich, context heavy texts, particularly those from ancient scriptures like the Atharv Ved. To address this, we propose a multi-layered architecture that begins with preprocessing steps such as text cleaning and segmentation into manageable units like verses or hymns. A pretrained T5 model is then used for text transformation and summarization to ensure proper structuring. The core of the system employs a BERT + BiLSTM architecture, where BERT enervates deep contextual embeddings by capturing bidirectional context, and BiLSTM models the sequential flow of information in both forward and backward directions. An attention mechanism prioritizes key phrases or sections, focusing on the most relevant parts of the text for accurate answer generation. This system is evaluated on a dataset derived from the Atharva Veda and the SQuAD dataset, highlighting its ability to answer complex, context-dependent queries. Our approach demonstrates the potential of AI in making ancient cultural texts more accessible and interpretable, offering a powerful tool for understanding rich, contextually complex content. Key Words: Atharv Ved, Word Embeddings, Deep Learning, AI, BERT, Attention Mechanism, BiLSTM, NLP, T5
1.INTRODUCTION The exponential increase in data volume and velocity in today's digital landscape poses persistent challenges for extracting relevant and accurate information. Conventional information retrieval (IR) platforms, including search engines, often require users to sift through extensive content, which results in inefficiency and information saturation. Question answering (QA). systems have been developed to address this issue, offering direct responses to queries in natural language, thus enhancing accessibility and userfriendliness [3, 4, 5]. These systems have evolved from initial rule-based approaches, which are limited in handling complex or ambiguous queries, to statistical models, which
© 2024, IRJET
|
Impact Factor value: 8.315
|
facilitate more adaptable and dynamic matching through language modelling [6, 7]. Although significant advancements have been made, earlier iterations of these systems still encounter difficulties in providing highly accurate and contextually appropriate answers. Recent advancements in AI powered question-answering (QA) systems have been marked by significant progress through the integration of machine learning and natural language processing techniques [8, 9, 10]. These improvements have enhanced the capacity of systems to comprehend the underlying meaning of user queries, resulting in more precise and contextually appropriate responses. This development is particularly crucial in fields such as healthcare, where information accuracy is paramount [28, 29]. Additionally, knowledge graphs have become integral components that boost the ability of systems to retrieve semantically relevant information [14, 15]. However, challenges persist, especially in specialized domains such as medical QA, where the reliability of information and fair access to resources are essential [28, 30]. This overview examines the progression of QA systems and underscores their potential to revolutionize information retrieval across various sectors, with particular emphasis on specialized fields [3, 4, 28].
1.1 Current Research Limitations: Question Answering Systems (QAS) face challenges in processing complex questions, interpreting semantic content, and addressing language-specific issues. Current systems are often limited to simple, objective questions and struggle with nuanced understanding due to model limitations (Ansari et al., 2016; Alanazi et al., 2021). Language-specific challenges, such as Arabic QAS dealing with complex morphology and lack of vowels, furthercomplicate their development (Albarghothi et al., 2017). To address these issues, researchers are employing deep neural networks, enhanced word encoding, and AI-driven approaches (Alanazi et al., 2021). Future work should focus on creating adaptable QAS capable of handling diverse question types and languages.
ISO 9001:2008 Certified Journal
|
Page 825