Scholarlink

<p>Sentiment analysis of online news is improved through a framework leveraging Backpropagation Neural Networks (BNNs) optimized by three algorithms: Levenberg-Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG). The model is trained and tested on the Online News Popularity dataset with different neuron setting to evaluate performance. Among the three algorithms, SCG showed the best performance, reaching a high classification accuracy of 99.83% and the lowest Mean Squared Error (MSE). TextBlob is used for feature extraction, specifically to obtain polarity and subjectivity scores from the text data, which serve as inputs to the neural network. Evaluation through metrics using cross-entropy (CE), ROC curve, precision, recall, and F-1 score confirms the strength of the SCG optimized BNNs. Comparative analysis with conventional machine learning and ANFIS models further highlights the usefulness of SCG-optimized BNNs. Future work will explore hybrid models to enhance scalability in large-scale, real-time news analytics.</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

Optimizing Sentiment Analysis Performance with Backpropagation Neural Networks: A Comparative Evaluation of LM, BR, and SCG

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Amit Kumar Srivasatava,
  • Pooja,
  • T. J. Siddiqui

摘要

Sentiment analysis of online news is improved through a framework leveraging Backpropagation Neural Networks (BNNs) optimized by three algorithms: Levenberg-Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG). The model is trained and tested on the Online News Popularity dataset with different neuron setting to evaluate performance. Among the three algorithms, SCG showed the best performance, reaching a high classification accuracy of 99.83% and the lowest Mean Squared Error (MSE). TextBlob is used for feature extraction, specifically to obtain polarity and subjectivity scores from the text data, which serve as inputs to the neural network. Evaluation through metrics using cross-entropy (CE), ROC curve, precision, recall, and F-1 score confirms the strength of the SCG optimized BNNs. Comparative analysis with conventional machine learning and ANFIS models further highlights the usefulness of SCG-optimized BNNs. Future work will explore hybrid models to enhance scalability in large-scale, real-time news analytics.