Topic Modeling for Nepali Text

Abstract

Topic Modeling identifies latent themes or topics within individual documents or collections of documents. By analyzing text, Topic Modeling detects patterns and relationships that reveal these underlying themes, offering valuable insights for decision-making in large datasets. Because most research has focused on Topic Modeling based on traditional techniques, and limited studies have explored its application in Nepali text, this research explores an innovative approach to topic modeling for Nepali text using BERTopic. The coherence score and Topic Diversity have been used as the evaluation criteria to explore the quality and interpretability of topics generated by BERTopic with the embedding model paraphrase-multilingual-MiniLM-L12-v2. The coherence Score and topic diversity were recorded as 0.58 and 0.85, respectively, using paraphrase-multilingual MiniLM-L12-v2. This study concludes that BERTopic can effectively model Nepali text and extract coherent topics despite the unique challenges posed by the Nepali language.