Analysis of Performance Labelling Sentiment Between K-Means Indobert And Inset Lexicon-Based

Sentiment analysis, a natural language processing technique, plays a key role in identifying opinions or sentiments from textual data. Accurate sentiment labelling within a dataset significantly impacts the performance of sentiment analysis models. However, manual labelling can be time-consuming. Many researchers utilize lexicon-based methods for sentiment labelling, but lexicons are often limited in reflecting topic-specific nuances, potentially leading to inaccurate sentiment representation. This inaccuracy can negatively affect classification models. Inset Lexicon (Indonesia Sentiment Lexicon) provides a pre-weighted list of sentiment words for sentiment analysis in Indonesian. This study aims to explore the use of K-means clustering as an automatic sentiment labelling technique and compare it to the performance of Inset Lexicon. For K-means clustering, IndoBERT is employed as the embedding model. The objective of this research is to evaluate the accuracy of automatic sentiment labelling by comparing it with actual data to assess the performance of both methods. The experiment accuracy shows that K-means with IndoBert achieves 74.79%, higher than Inset Lexicon that achieves only 59.82%