Nithya T, Ramkumar M Siva, Thavasimuthu Rajendran, Anitha T, Munimathan Arunkumar, Khan Anwar, Naveed Quadri Noorulhasan, Khan Shafat, Zewdie Amanuel
Department of Computer Science and Engineering, Rajalakshmi Institute of Technology, Chennai, Tamil Nadu, India.
Department of ECE, SNS College of Technology, Coimbatore, India.
Sci Rep. 2025 Aug 29;15(1):31845. doi: 10.1038/s41598-025-16927-0.
Opinion mining is more challenging than it was before because of all the user-generated material on social media. People use Twitter (X) to gather opinions on products, advancements, and laws. Sentiment Analysis (SA) examines people's thoughts, feelings, and views on numerous topics. Tweets can be analyzed to determine public opinion on news, regulations, society, and personalities. The existing SA system has poor prediction performance and needs improvements for instantaneous commercial applications. The insufficient data and complexity of model configuration, which make deep learning (DL) difficult, are the main causes of low accuracy and prediction rates. Convolutional Neural Long Short-Term Memory (OTCNLSTM) optimal tiered blocks with classification learning are proposed in this research to recognize emotions. The objective is to classify tweets as happy or sad. The TCNLSTM model consists of four training blocks for local features. These blocks are designed to extract local emotions hierarchically. The Boosted Killer Whale Predation (BKWOP) strategy is implemented to find the appropriate hyperparameter and its solution set and build a stable neural network model. This research reviews textual emotional classification experiments utilizing various sentiment models and methods. To further analyze this research, a comparative experimental study using the Kaggle Twitter dataset is conducted. The results indicate that the OTCNLSTM model had superior performance compared to the other models.
由于社交媒体上所有用户生成的内容，意见挖掘比以前更具挑战性。人们使用推特（X）来收集关于产品、进步和法律的意见。情感分析（SA）研究人们对众多主题的想法、感受和观点。可以分析推文以确定公众对新闻、法规、社会和人物的看法。现有的SA系统预测性能较差，需要改进以用于即时商业应用。数据不足和模型配置的复杂性使得深度学习（DL）变得困难，这是准确率和预测率低的主要原因。本研究提出了具有分类学习的卷积神经长短期记忆（OTCNLSTM）最优分层块来识别情绪。目标是将推文分类为开心或悲伤。TCNLSTM模型由四个用于局部特征的训练块组成。这些块旨在分层提取局部情绪。实施了增强型虎鲸捕食（BKWOP）策略来找到合适的超参数及其解集，并构建一个稳定的神经网络模型。本研究回顾了利用各种情感模型和方法进行的文本情感分类实验。为了进一步分析本研究，使用Kaggle推特数据集进行了对比实验研究。结果表明，OTCNLSTM模型的性能优于其他模型。
