Finding topics in Urdu: A study of applicability of document clustering in Urdu language

In this research, we present the results of a study conducted to ascertain the applicability of document clustering techniques on Urdu Language corpus. This study, which is first of its kind, employs a fully probabilistic Bayesian method, Latent Dirichlet Allocation, for clustering Urdu language corpus by using the features collected from the documents. Results obtained are compared with those obtained from a simplistic classification technique. Analysis of the results shows that supervised and unsupervised techniques for grouping documents perform reasonably well on this corpus. Results further indicate that Urdu document clustering technique outperforms document classification technique in some cases with an accuracy of above 90%.