Scholarlink

<p>With the rapid growth of social media, people progressively sharing their thoughts, emotions, and experiences online, offering valuable insights into their mental well-being. The proposed work emphasizes the importance of identifying depression at the early stage in people by mining the social media posts. The framework involves three main stages, beginning with data collection and cleaning, followed by data normalization, and finally detecting depression from the extracted features. A novel acronym replacement technique is employed to preserve the context of conversations, thereby enhancing the quality of the training data. Besides this, preprocessing techniques including removal of special symbols, handling stopwords, and contextual feature generation using back translation, further enhance data consistency. The processed text is modeled using a Recurrent Neural Network (RNN)+Long Short Term Memory (LSTM) architecture to classify the tweets as depressive or non-depressive accordingly using four different approaches. The efficacy of the work is assessed using accuracy, precision, and roc score. Experimental evaluation conducted via four approaches demonstrates a significant improvement in depression detection over existing models by achieving an accuracy of 99.78%. The proposed approach offers a reliable and effective solution for depression detection at an early stage from social media text.</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

Tuning LSTM network for early depression detection in Twitter texts: a grid search approach

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Shyam Sunder Jannu Soloman,
  • Nagaraju Baydeti

摘要

With the rapid growth of social media, people progressively sharing their thoughts, emotions, and experiences online, offering valuable insights into their mental well-being. The proposed work emphasizes the importance of identifying depression at the early stage in people by mining the social media posts. The framework involves three main stages, beginning with data collection and cleaning, followed by data normalization, and finally detecting depression from the extracted features. A novel acronym replacement technique is employed to preserve the context of conversations, thereby enhancing the quality of the training data. Besides this, preprocessing techniques including removal of special symbols, handling stopwords, and contextual feature generation using back translation, further enhance data consistency. The processed text is modeled using a Recurrent Neural Network (RNN)+Long Short Term Memory (LSTM) architecture to classify the tweets as depressive or non-depressive accordingly using four different approaches. The efficacy of the work is assessed using accuracy, precision, and roc score. Experimental evaluation conducted via four approaches demonstrates a significant improvement in depression detection over existing models by achieving an accuracy of 99.78%. The proposed approach offers a reliable and effective solution for depression detection at an early stage from social media text.