Scholarlink

<p>Emotion detection (ED) in natural language processing (NLP) is a critical task with applications in sentiment analysis, chatbots, and user feedback analysis. This study explored the impact of custom-trained embeddings on ED, specifically in Malayalam text. We developed and evaluated novel embedding models, <b>malayalam</b>_<b>word2vec.model</b> and <b>fasttext</b>_<b>model.bin</b>, trained specifically on a Malayalam ED dataset named MalEmo, created from Malayalam novels. We utilized IndicNLP for tokenization and optimized the embeddings to suit the linguistic characteristics of Malayalam. In this work, EmoVec, we developed custom embeddings for ED in Malayalam and compared their performance to pretrained FastText embeddings (cc.ml.300.bin). Additionally, we fine-tuned transformer models (mBERT and MuRIL) for ED tasks and demonstrated that the transformer-based models outperform the state-of-the-art results. Moreover, this work highlighted the importance of IndicNLP tokenization in the training of the proposed embedding models for ED in Malayalam.</p>

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

EmoVec: emotion detection in Malayalam using custom word embeddings

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • K. Anuja,
  • P. C. Reghu Raj,
  • K. R. Remesh Babu

摘要

Emotion detection (ED) in natural language processing (NLP) is a critical task with applications in sentiment analysis, chatbots, and user feedback analysis. This study explored the impact of custom-trained embeddings on ED, specifically in Malayalam text. We developed and evaluated novel embedding models, malayalam_word2vec.model and fasttext_model.bin, trained specifically on a Malayalam ED dataset named MalEmo, created from Malayalam novels. We utilized IndicNLP for tokenization and optimized the embeddings to suit the linguistic characteristics of Malayalam. In this work, EmoVec, we developed custom embeddings for ED in Malayalam and compared their performance to pretrained FastText embeddings (cc.ml.300.bin). Additionally, we fine-tuned transformer models (mBERT and MuRIL) for ED tasks and demonstrated that the transformer-based models outperform the state-of-the-art results. Moreover, this work highlighted the importance of IndicNLP tokenization in the training of the proposed embedding models for ED in Malayalam.