Analyzing Audience Sentiments in Digital Comedy: A Study of YouTube Comments Using LSTM Models

Abstract

The main objective of this paper is to analyze audience sentiment towards stand-up comedy content on the YouTube platform, specifically comments on stand-up comedy videos from Kompas TV, using the Long Short-Term Memory (LSTM) method. This research contributes significantly to a deeper understanding of how audiences engage with humorous content through a sentiment analysis approach that uses the LSTM model, which can capture complex nuances in humorous content, such as sarcasm, irony, and cultural references. The research methodology involves crawling data from YouTube, where user comments are extracted and processed through several stages of data cleaning, such as removing duplicate content, text normalization, and irrelevant comments. Once the data is prepared, the LSTM model is trained to analyze positive, negative, and neutral sentiments with varying accuracy rates of 85% for positive sentiment, 80% for negative sentiment, and 78% for neutral sentiment. The main results show that the LSTM model successfully classifies sentiments, although it needs help handling the more ambiguous neutral sentiments. The figures and tables included in this study illustrate the relationship between the number of views, likes, and the sentiment classification of the comments. One notable finding is a strong positive correlation between the number of views and video likes. The conclusions of this study underscore the need for model improvements to handle neutral sentiment better and capture the complexity of humor content. The implications of this research are useful for content creators and digital marketers in understanding and responding to audience preferences more effectively. They also pave the way for further research in sentiment analysis on more specific content genres on digital platforms.