Self voting classification model for online meeting app review sentiment analysis and topic modeling

Area of Science:

• Natural Language Processing
• Machine Learning
• Software Engineering

Background:

• Online meeting applications are crucial for communication, education, and business, especially post-COVID-19.
• User feedback via reviews is vital for improving these applications.
• Accurate sentiment analysis is essential for understanding user opinions from feedback.

Purpose of the Study:

• To propose and evaluate a novel ensemble method, self voting classification (SVC), for sentiment analysis of online meeting app reviews.
• To enhance the performance of traditional machine learning models through SVC.
• To perform topic-wise sentiment analysis using Latent Dirichlet Allocation (LDA).

Main Methods:

• Developed a self voting classification (SVC) approach, training multiple model variants with different feature extraction methods (bag of words, TF-IDF, hashing).
• Applied SVC with Support Vector Machine (SVM), Decision Tree, Logistic Regression, and K-Nearest Neighbor models.
• Utilized both hard voting (HV) and soft voting (SV) criteria for ensembling model predictions.
• Employed Latent Dirichlet Allocation (LDA) for topic modeling and topic-wise sentiment analysis.

Main Results:

• The proposed SVC approach substantially improved the performance of traditional machine learning models.
• Support Vector Machine (SVM) achieved perfect (1.00) and near-perfect (0.98) accuracy scores using hard voting and soft voting criteria, respectively, with SVC.
• Demonstrated the effectiveness of ensemble methods in sentiment analysis for online meeting applications.

Conclusions:

• Self voting classification (SVC) is an effective strategy for enhancing sentiment analysis accuracy in online meeting applications.
• Ensemble techniques, particularly SVC, offer significant performance gains over traditional single-model approaches.
• The study provides a robust method for analyzing user sentiment and topics within app reviews.