International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 02 | Feb 2025
p-ISSN: 2395-0072
www.irjet.net
Sentimental Analysis of Kannada Product Review Using BERT Prof. Sushma T M1, Sudeep T C2, Kiran Janardhan Marati3,N Sneha Reddy4, Rakesh V5 1,2,3,4,5Information Science and Engineering, Acharya Institute of Technology Bangaluru, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper addresses the underrepresentation of
Kannada in sentiment analysis tools and presents a novel finetuned BERT model for Kannada sentiment classification. The study highlights challenges such as data scarcity, linguistic nuances, and a lack of NLP resources. Utilizing the IndicSentiment dataset, the model achieved an accuracy of 89% with improvements across training epochs. Applications in e-commerce, social media monitoring, and policy feedback analysis demonstrate the versatility and impact of this work. Key Words: Sentiment Analysis, Kannada, BERT, Natural Language Processing, Regional NLP, IndicSentimentinsert.
1.INTRODUCTION
Fig-1: BERT model
Natural Language Processing (NLP) has experienced rapid advancements over the last decade, driven by innovations in machine learning and the development of transformer-based architectures like BERT. Despite these advancements, low-resource languages, such as Kannada, remain underserved, limiting the reach of NLP applications in regions where such languages are predominantly used.
The contributions of this work are multifold. First, it bridges the gap in sentiment analysis tools for Kannada by introducing a model fine-tuned specifically for the language. Second, it highlights the practical applications of such tools in areas like e-commerce, social media monitoring, and policy analysis. Finally, it sets a precedent for future research aimed at extending NLP technologies to other regional languages, promoting inclusivity in AI-driven solutions.
Sentiment analysis, a subset of NLP, plays a crucial role in extracting valuable insights from textual data. It enables businesses to understand customer feedback, policymakers to gauge public opinion, and researchers to analyze social trends. However, the scarcity of labeled data, unique linguistic characteristics, and limited computational resources have impeded the development of sentiment analysis tools for Kannada. Addressing these challenges requires innovative approaches that not only leverage modern architectures but also adapt them to the specific nuances of the Kannada language.
2. LITRATURE REVIEW Sentiment analysis in regional languages like Kannada has been an emerging area of research, but the field faces significant challenges due to the scarcity of annotated datasets, the complexity of the language, and the lack of domain-specific tools. This section highlights notable prior work that has shaped the foundation for Kannada sentiment analysis and NLP research.
2.1 Multi-Linguistic Sentimental Analysis Models
BERT (Bidirectional Encoder Representations from Transformers) has emerged as a transformative model in NLP, offering unparalleled capabilities in understanding the context and semantics of text. By pre-training on large corpora and fine-tuning for specific tasks, BERT provides a robust framework for addressing low-resource language challenges. This paper presents a fine-tuned BERT model tailored for Kannada sentiment analysis, utilizing the IndicSentiment dataset to achieve high accuracy and reliability.
© 2025, IRJET
|
Impact Factor value: 8.315
Proposed the Multi-Linguistic Sentiment Analyzer (MuLSA), which performs sentiment analysis for Kannada, Malayalam, and English using Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks. The system includes modules such as language detection, translation, and sentiment classification, achieving an accuracy of 82\% for Kannada sentiment classification tasks. However, its reliance on RNN-based architectures makes it computationally expensive and less effective for long-range dependencies. While effective for multilingual contexts, its Kannada-specific performance highlights the need for more optimized models.[1]
|
ISO 9001:2008 Certified Journal
|
Page 79