Comparative Analysis of Topic Modeling Algorithms Based on Arabic News Documents

Topic modeling is a text mining technique that revolves around extracting latent topics from a collection of documents. Although the majority of research within the field of topic modeling has been conducted in the English language. Nonetheless, in recent years, there has been an interest in employing the topic modeling methodology within the Arabic language, although its utilization remains somewhat restricted in this language. In this paper, we propose a comparison among various techniques commonly utilized in topic modeling. These techniques include a Probabilistic model, specifically Latent Dirichlet Allocation (LDA), as well as matrix factorization methods like Non-Negative Matrix Factorization (NMF) and Latent Semantic Indexing (LSI). Additionally, we incorporate a transformer-based model known as BERTopic. The implementation was applied to the Arabic language, and the algorithms were trained using the TF-IDF text representation. This choice aimed to ensure a fair comparison between the algorithms. The evaluation of each model is conducted using topic coherence as the metric. The results indicate that both NMF and Bertopic give an excellent performance.