The current advancement of information technology has led to changes in various
aspects of human life, including the automotive sector. Many people use the
internet and social media to share opinions and information, with YouTube being
a popular platform for uploading videos and watching a wide range of content.
Based on this context, this study offers a solution to analyze the topic modeling of
content created by automotive YouTubers. The research employs the Latent
Dirichlet Allocation (LDA) method, one of the topic modeling techniques, to help
identify the topics that emerge on automotive YouTuber accounts. The LDA
method involves summarizing, clustering, linking, and processing data to
generate a weighted list of topics for each document. The data collection in this
study was conducted through observation and web scraping on YouTube using
the YouTube API V3. The data obtained in the initial stage were raw, so the next
step involved preprocessing to clean the data, making it easier for subsequent
processing. The topic modeling process using Latent Dirichlet Allocation aimed
to form the best possible topic model. The results showed a perplexity score of -
8.1616 and a coherence score of 0.4732 from six topics, with the most dominant
topic being the first topic, which had an accuracy rate of 27.7%, containing
keywords such as Honda, Yamaha, review, Kawasaki, engine, Ninja, Nmax, and
Vario. This study acknowledges certain limitations and weaknesses. Therefore,
the researchers recommend the following for future studies: 1. Adding a word
cloud to visualize the data, and 2. Incorporating additional machine learning
algorithms to compare and determine which algorithm performs better in future
research. Keywords: Clustering, YouTube, Automotive, Topic Modeling, Latent
Dirichlet Allocation.