A Comparison of Topic Modeling Techniques for Sinhala

Topic modeling uncovers latent themes in expansive document collections, especially in travel, by interpreting patterns within vast narratives and reviews. Yet, for Sinhala, a low-resourced language, there's a notable research gap. Addressing this, our research conducts a comparative analysis of three leading topic models: LDA, LSA, and NMF, providing guidance on the optimal model for Sinhala datasets-a critical insight for professionals even outside computer science. While evaluating topic models presents inherent challenges, our study offers a comprehensive evaluation using multiple metrics under varied conditions, leveraging three datasets mined from social media. In light of our intensive evaluations, both quantitative and qualitative measures are in favor of NMF model across majority of datasets. However, it's essential to consider the contextual relevance of topics, the nature of the dataset, and the specific use-case when selecting a model. Thus, the determination of a singular ‘best model’ requires a holistic consideration of various factors.