Scholarlink

News has an important role to play in society. It helps keep the public informed and educated about what is happening around them. That is why, due to the immense responsibility that news has in influencing public thoughts, spreading fake news is an extremely relevant issue in today’s world, where it has become quite common. In this paper, we compare models-BERT, IndicBERT, roBERTa and Muril and an ensemble model. This paper also evaluates traditional models like support vector machines, linear regression, and random tree and aims to analyze and conclude which among them is best for fake news detection. This comparison is done based on various parameters like sentiment analysis, performance analysis, and accuracy analysis. The findings of the paper conclude the need for specific models solely designed for Dravidian languages and models that make comparisons not just based on pattern recognition and sentiment evaluation, but also through fact checking.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fake News Detection in Dravidian Languages: Comparative Analysis of Transformer Models, Ensemble Techniques, Traditional Classifiers, and Sentiment Influence on Prediction Performance

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rajalakshmi Sivanaiah,
  • S. Angel Deborah,
  • T. T. Mirnalinee,
  • Vishal Subramoniam,
  • Sudarshan Olirvel,
  • Sathvika K. Shankar,
  • Bhavana Balasubramanian

摘要

News has an important role to play in society. It helps keep the public informed and educated about what is happening around them. That is why, due to the immense responsibility that news has in influencing public thoughts, spreading fake news is an extremely relevant issue in today’s world, where it has become quite common. In this paper, we compare models-BERT, IndicBERT, roBERTa and Muril and an ensemble model. This paper also evaluates traditional models like support vector machines, linear regression, and random tree and aims to analyze and conclude which among them is best for fake news detection. This comparison is done based on various parameters like sentiment analysis, performance analysis, and accuracy analysis. The findings of the paper conclude the need for specific models solely designed for Dravidian languages and models that make comparisons not just based on pattern recognition and sentiment evaluation, but also through fact checking.