Analyzing Public Sentiment and Discussion Topics on AI Conversational Agents: A Comprehensive Study Using Machine Learning and Topic Modeling Techniques

Abstract

Abstract—This paper explores sentiment analysis and topic modeling in the context of AI tools and conversational agents. The emerging popularity of these AI-powered technologies in everyday life, such as ChatGPT, Google Bard, Bing, and others, has led to an increasing need to understand public sentiment and discussion topics concerning these agents. Our approach involves data scraping from Twitter, preprocessing to clean the data, applying Latent Dirichlet Allocation (LDA) for topic modeling, and then conducting sentiment analysis on each topic using machine learning (Ml) models including Random Forest (RF), Support Vector Machines (SVM), Multinomial Naive Bayes (NB), Gradient Boosting (GB) Classifier, Logistic Regression (LR), and an Ensemble Learning (EL) method. The performance of the models was evaluated using metrics such as confusion matrix, ACC, Prec, Rec, and F1-s. Results demonstrated that EL and LR were particularly effective in sentiment classification. Furthermore, the LDA model successfully unveiled distinct topics in the discourse around AI tools and conversational agents. Future work includes implementing deep learning models for improved sentiment analysis and extending the scope to a larger, more diverse dataset for comprehensive insights. The study provides a useful methodology for understanding public sentiment and discourse about rapidly evolving AI technologies.