Analisis Pemodelan Topik Pada Teks Terjemahan Al-Qur’an Berbahasa Indonesia Dengan Metode Correlated Topic Modeling (CTM) Dan BERTopic

Abstract

This study aims to analyze the thematic structure of the Indonesian translation of the Quran using the Correlated Topic Model (CTM) and BERTopic methods. The research stages include text preprocessing, evaluation of preprocessing variations, model parameter optimization, topic modeling, and analysis of the results. The preprocessing process is carried out through case folding, tokenization, stopword removal, and stemming. The evaluation results show that the advanced preprocessed method provides the best balance with a vocabulary reduction of 83.75% and meaning preservation of 73.20%. In CTM modeling, optimization of the number of topics shows that the optimal value is at K = 14 based on a combination of log-likelihood, topic diversity, and composite score metrics. The CTM model is able to produce representative topics with low inter-topic correlation, although the internal coherence value is still at a moderate level. Meanwhile, in BERTopic, the optimal parameter is obtained at min_topic_size = 25 with a coherence value of 0.5735 (C_v > 0.5), which is quite good. The positive NPMI value = 0.0530 indicates that there is a statistically significant association between words in the topic. The UMass value = -5.3257, which is the best value (closest to zero) compared to other configurations, indicates that the model has good probabilistic consistency towards document distribution. The results of combining topics between CTM and BERTopic using the Jaccard similarity approach show that there is a relationship between topics even with a relatively low to moderate level of similarity (0.08–0.25).