Deteksi Tren Topik Penelitian Bidang Kecerdasan Buatan Menggunakan LDA dan BERTopic

Research that is growing from year to year makes it difficult for research to understand trends in research topics without reading all scientific publications. This study aims to automatically determine trend of research topics in ICLR (International Conference on Learning Representation) from 2019 to 2023 using LDA (Latent Dirichlet Allocation) and BERTopic. Data is collected by crawling, a total of 14.613 data. Data preprocessing in this research,through several stages including preprocessing, corpus & dictionary for LDA, applying LDA and BERTopic modeling, evaluating modelling result using coherence. The result of this research show that both modelling can identify trends in research topics every year, and BERTopic modelling has a higher coherence value that LDA, this indicating that BERTopic has better quality in describing each word that has a relationship in a topic.