<![CDATA[The vast amount of textual data -particularly undergraduate theses abstracts produced in the digital age- makes it difficult for readers to identify the topics contained within them. Topic modeling facilitates readers in identifying topics within a collection of textual data. One method for topic modeling is BERTopic. BERTopic is a framework for topic modeling that utilizes the BERT model in embedding stage. This study use IndoSBERT and multilingual SBERT in the BERTopic embedding stage to determine which model performs better in generating topic for a dataset of Indonesian-language undergraduate theses abstract. The topic generated using these IndoSBERT and multilingual SBERT embedding are then evaluated using the topic coherence and topic diversity metrics. The research results show that topics generated by IndoSBERT have higher topic coherence and topic diversity scores, than those generated by multilingual SBERT. These results indicate that IndoSBERT is better in generating topics with topic coherence and topic diversity than multilingual SBERT. The contribution of this research lies in the modification of the BERTopic embedding stage using IndoSBERT for topic modelling in Bahasa. This is because the use of IndoSBERT has so far been limited for text classification task.Key words: Bahasa Indonesia, BERTopic, IndoSBERT, Information System, Topic modelling.]]>
Copyrights © 2026
