Predicting Customer Sentiment in Social Media Interactions: Analysing Amazon Help Twitter Conversations Using Machine Learning

• Abey Litty

  Texas University

  Author

Social media platforms have transformed customer service interactions, creating vast repositories of unstructured data that contain valuable insights into customer satisfaction and brand perception. Despite the proliferation of machine learning techniques for sentiment analysis, limited research has systematically examined sentiment evolution within individual customer service conversations and developed predictive models capable of forecasting sentiment outcomes based on initial interaction patterns. This study addresses this gap by analyzing over 6,500 customer conversations with Amazon's official support account (@AmazonHelp) on Twitter, employing multiple machine learning algorithms to classify sentiment changes and predict overall sentiment trajectories. Using a quantitative, design-based research methodology, we extracted English-language tweets through the Twitter API, applied comprehensive preprocessing procedures, and evaluated seven machine learning classifiers including K-Nearest Neighbor, Naive Bayes, Artificial Neural Network, Support Vector Machine, Logistic Regression, and Bagging with RepTree. Results demonstrate that K-Nearest Neighbor and Support Vector Machine offer the optimal balance between accuracy and F-measure, while Bagging with RepTree achieves the highest predictive accuracy of 89.4% albeit with lower F-measure performance. These findings establish a replicable framework for real-time customer sentiment prediction, offering practitioners actionable metrics for proactive customer engagement and providing researchers with a validated methodological approach for conversational sentiment analysis in social media contexts.

• pdf

Copyright (c) 2026 Abey Litty (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.