Scholarlink

<p>Healthcare professionals are using sentiment analysis tools to detect emotions from patient evaluation textual data based on medical records and social media. Standard text handling methods using one-hot encoding and Bag of Words (BoW) are unable to match medical writing patterns. This research creates a new approach to sentiment classification, which joins two processes: Word Embedding Informed Finite Topic Modeling (WEI-FTM) to model semantics and Local Feature Weighted (LFW) for contextual embeddings. Heterogeneous data is fed into a multi-scale CNN (convolutional neural network) extractor that yields important features for XGBoost (extreme gradient boosting) to perform scalable classification effectively. The system achieves superior results on all four emotion datasets, with F1 scores ranging from 0.895 to 0.932, and over 90% classification accuracy across the BREC (Breast Cancer Reddit Emotional Corpus), CARES (Cancer Emotional Twitter Dataset), CLPsych (Computational Linguistics and Clinical Psychology subset), and EMO-CAN (Emotion-Cancer Corpus) datasets. Moreover, hyperparameter tuning shows that the highest performance is obtained with a Dirichlet concentration parameter of α = 0.1 for WEI-FTM and convolution filter sizes, which represent the topic sparsity and best detection of local features.</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

Hybrid semantic-contextual embedding framework for emotion-aware sentiment analysis

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Geeta Kondal,
  • Hari Singh,
  • Saurabh Srivastava

摘要

Healthcare professionals are using sentiment analysis tools to detect emotions from patient evaluation textual data based on medical records and social media. Standard text handling methods using one-hot encoding and Bag of Words (BoW) are unable to match medical writing patterns. This research creates a new approach to sentiment classification, which joins two processes: Word Embedding Informed Finite Topic Modeling (WEI-FTM) to model semantics and Local Feature Weighted (LFW) for contextual embeddings. Heterogeneous data is fed into a multi-scale CNN (convolutional neural network) extractor that yields important features for XGBoost (extreme gradient boosting) to perform scalable classification effectively. The system achieves superior results on all four emotion datasets, with F1 scores ranging from 0.895 to 0.932, and over 90% classification accuracy across the BREC (Breast Cancer Reddit Emotional Corpus), CARES (Cancer Emotional Twitter Dataset), CLPsych (Computational Linguistics and Clinical Psychology subset), and EMO-CAN (Emotion-Cancer Corpus) datasets. Moreover, hyperparameter tuning shows that the highest performance is obtained with a Dirichlet concentration parameter of α = 0.1 for WEI-FTM and convolution filter sizes, which represent the topic sparsity and best detection of local features.