Graph embedding approaches for social media sentiment analysis with model explanation

Abstract

ChatGPT, the revolutionary chat agent launched in November 2022, is still an active topic of discussion among technology enthusiasts. This open-ended chatbot allows human-like conversations with it on almost all topics since it was trained on millions of documents and developed as a large language model. Since its inception, there have been several discussions and deliberations, especially on twitter and other social media handles, on the potential of ChatGPT and how the power of artificial intelligence is growing leaps and bounds in general. These platforms also witnessed several debates on the negative side of ChatGPT, such as adversely affecting the integrity and ethics and the biased training data. This work uses graph neural network embeddings with machine learning algorithms for classifying the user sentiments on ChatGPT. We have collected a total of 8202 tweets and manually labeled them into multiple classes such as positive, negative, and neutral. We make the models explainable using SHAP (SHapley Additive exPlanations), which is a game theoretical technique for explaining the output of any machine learning models. This paper also publishes our labeled dataset for other researchers to use and train advanced classification models. When our proposed approach was compared with some chosen baselines, the proposed graph embedding-based machine learning classifiers were found to be outperforming in terms of precision, recall, and accuracy.

Keywords

• Sentiment analysis
• Explainable artificial intelligence
• Natural language processing
• Social media
• Graph embedding
• Explainability