Scholarlink

<p>This study presents a comprehensive Sentiment Analysis (SA) framework applied to a novel, real-time collected YouTube comment dataset categorized into five distinct emotional classes: happiness, sadness, fear, anger, and surprise. Unlike conventional studies that rely on pre-existing or simplistic binary datasets, we employ dynamic data acquisition and rigorously evaluate the performance of eleven distinct classifiers, including traditional Machine Learning (ML) algorithms (K-Nearest Neighbors, Naïve Bayes (NB), Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM)) and advanced Deep Learning (DL) models (Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Gated Recurrent Unit (GRU), Bidirectional Gated Recurrent Unit (BiGRU), and a CNN-LSTM hybrid). A central contribution involves the systematic comparison of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) as feature selection techniques across all models. The findings demonstrate that dimensionality reduction techniques significantly impact model efficacy, with the LSTM model achieving the highest performance (88% accuracy) on the PCA-processed dataset, matched only by LR on the LDA-processed dataset. This work provides critical insights into optimizing classifier choice based on feature processing methods for multi-class sentiment analysis using dynamic social media data.</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

Sentiment Analysis on Youtube Comments Using Machine Learning and Deep Learning with PCA- and LDA-Based Feature Selection

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Gulay Cicek,
  • Nazlı Buldağ,
  • Elif Aydın

摘要

This study presents a comprehensive Sentiment Analysis (SA) framework applied to a novel, real-time collected YouTube comment dataset categorized into five distinct emotional classes: happiness, sadness, fear, anger, and surprise. Unlike conventional studies that rely on pre-existing or simplistic binary datasets, we employ dynamic data acquisition and rigorously evaluate the performance of eleven distinct classifiers, including traditional Machine Learning (ML) algorithms (K-Nearest Neighbors, Naïve Bayes (NB), Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM)) and advanced Deep Learning (DL) models (Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Gated Recurrent Unit (GRU), Bidirectional Gated Recurrent Unit (BiGRU), and a CNN-LSTM hybrid). A central contribution involves the systematic comparison of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) as feature selection techniques across all models. The findings demonstrate that dimensionality reduction techniques significantly impact model efficacy, with the LSTM model achieving the highest performance (88% accuracy) on the PCA-processed dataset, matched only by LR on the LDA-processed dataset. This work provides critical insights into optimizing classifier choice based on feature processing methods for multi-class sentiment analysis using dynamic social media data.