Leveraging BERTopic for the Analysis of Scientific Papers on Seaweed

Abstract

Seaweed has many benefits, serving as a source of highly nutritious food, raw material for the cosmetic industry, pharmaceuticals, and animal feed. Seaweed also plays a crucial role in maintaining environmental ecosystems. Therefore, it is important to understand research trends concerning seaweed. BERTopic is a commonly used method in topic modeling for analyzing scientific publication data. This study aims to find the best model using BERTopic for analyzing scientific publications related to seaweed. This research tests and evaluates the performance of methods such as stopword removal, embedding, dimensionality reduction, and clustering based on topic coherence values to obtain the optimal model. The test results indicate that removing stopwords using CountVectorizer, employing SCIBERT as the transformer, utilizing UMAP as the dimensionality reduction algorithm, and applying k-means as the clustering algorithm yield the best coherence values for analyzing scientific publications related to seaweed.