Scholarlink

Aspect-based sentiment analysis (ABSA) is a fine-grained sentiment analysis which aims in extracting opinion terms, opinion targets, aspect category and the polarity associated with each aspect category. Most ABSA research is in English with minimal work in Nepali. We have added 1031 Nepali comments related to news and politics category from YouTube to the dataset by previous researcher under same annotation guideline to facilitate the research of ABSA in social media domain. This paper introduces target-oriented opinion term extraction (TOWE) in Nepali ABSA, alongside aspect category detection (ACD) and aspect category polarity (ACP) detection tasks. We present convincing results using NepaliBERT for TOWE with ROUGE-L score of 0.80. We experiment with various models including SVM, LSTM, BiLSTM, CNN, and multilingual BERT for ACD and aspect category polarity achieving F1-scores of 82.78% and 83.13% for ACD and ACP respectively. Our method has outperformed current State-of-the-art for classification tasks in Nepali.

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

Aspect-Based Sentiment Analysis of Nepali Text Using BERT Based Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Suprabha Regmi,
  • Aman Shakya,
  • Basanta Joshi

摘要

Aspect-based sentiment analysis (ABSA) is a fine-grained sentiment analysis which aims in extracting opinion terms, opinion targets, aspect category and the polarity associated with each aspect category. Most ABSA research is in English with minimal work in Nepali. We have added 1031 Nepali comments related to news and politics category from YouTube to the dataset by previous researcher under same annotation guideline to facilitate the research of ABSA in social media domain. This paper introduces target-oriented opinion term extraction (TOWE) in Nepali ABSA, alongside aspect category detection (ACD) and aspect category polarity (ACP) detection tasks. We present convincing results using NepaliBERT for TOWE with ROUGE-L score of 0.80. We experiment with various models including SVM, LSTM, BiLSTM, CNN, and multilingual BERT for ACD and aspect category polarity achieving F1-scores of 82.78% and 83.13% for ACD and ACP respectively. Our method has outperformed current State-of-the-art for classification tasks in Nepali.