Investigating ChatGPT on Reddit by Using Lexicon-Based Sentiment Analysis

Abstract

ChatGPT is one of the most favorable AI-driven applications, gaining widespread popularity and usage among users. Despite it provides numerous advantages, ChatGPT also introduces security vulnerabilities and ethical concerns, such as plagiarism, privacy issues, data security risks, and the potential for biased data generation. Hence, identifying and addressing the sentiment of ChatGPT is essential, as it helps understand public perception and highlights areas requiring improvement. The research investigates the efficacy of various lexicon-based sentiment analysis tools, namely VADER, Liu-Hu, SentiWordNet, and AFINN, in capturing user sentiment from Reddit comments related to ChatGPT. The study emphasizes the importance of rigorous text preprocessing steps, including case folding, normalization, cleaning, tokenizing, stop-words removal, and stemming, to enhance the accuracy and reliability of sentiment analysis. The result highlights the presence of some consistency in identifying neutral sentiment across the tools, but also underscores the notable variability in detecting positive and negative sentiments. AFINN tends to identify a greater number of positive sentiments and a lesser number of negative ones, while VADER and SentiWordNet offer a more balanced and nuanced detection across all sentiment categories. Liu-Hu, meanwhile, occupies a middle ground with higher positive sentiment counts and lower negative sentiment counts. The evaluation of the sentiment clustering quality is conducted using the Silhouette Index, which measures that VADER and SentiWordNet exhibit higher scores (0.5326 and 0.5173, respectively) than other tools. The results of the sentiment analysis indicate that ChatGPT garners significantly more positive sentiment compared to negative and neutral sentiments.

Ähnliche Arbeiten