Scholarlink

A crucial aspect of natural language processing is the detection of the viewpoint or perspective expressed in texts which is known as stance detection. The goal of this work is to perform multilingual stance detection on news and social media posts in English, Kannada and Code Mixed Kannada and English, in a question-answer format on two topics, associated with stances such as, “Favour”, “Against” and “Neutral”. The other goal of this work is to develop and evaluate machine learning models for stance detection across multiple targets in Kannada and English, exploring the challenges and opportunities of linguistic diversity. We aim to investigate the effectiveness of stance detection in a multilingual context using a combination of traditional machine-learning techniques. This work presents a Kannada-English stance detection dataset, trained on classifiers like Logistic Regression, Random Forest, SVM, and a 1D-CNN with TF-IDF and multilingual BERT (mBERT) embeddings. SVM achieved the best performance by maximizing class margins, demonstrating robustness against class imbalance.

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

Kannada Stance Detection: Comparative Analysis of TF-IDF and BERT Embeddings with Traditional Classifiers

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shikha Reji,
  • Rishika Angadi Girish,
  • Khushi Suresh Muddi,
  • Shreya Suresh Jindrali,
  • M. Anand Kumar,
  • Richard Saldanha

摘要

A crucial aspect of natural language processing is the detection of the viewpoint or perspective expressed in texts which is known as stance detection. The goal of this work is to perform multilingual stance detection on news and social media posts in English, Kannada and Code Mixed Kannada and English, in a question-answer format on two topics, associated with stances such as, “Favour”, “Against” and “Neutral”. The other goal of this work is to develop and evaluate machine learning models for stance detection across multiple targets in Kannada and English, exploring the challenges and opportunities of linguistic diversity. We aim to investigate the effectiveness of stance detection in a multilingual context using a combination of traditional machine-learning techniques. This work presents a Kannada-English stance detection dataset, trained on classifiers like Logistic Regression, Random Forest, SVM, and a 1D-CNN with TF-IDF and multilingual BERT (mBERT) embeddings. SVM achieved the best performance by maximizing class margins, demonstrating robustness against class imbalance.