A Deep Learning based Approach for Sindhi Poet Classification using Couplets

Sindhi poetry serves as a monument to the cultural richness of Sindhi heritage, encompassing themes of affection, dedication, and community harmony. However, the contemporary decline in Sindhi language usage calls for innovative preservation efforts. In this study, we present a Deep Learning-based model for the automatic classification of Sindhi poets based on couplets. The dataset consists of 3000 couplets authored by five well-known Sindhi poets, each representing distinct periods and subject styles. We utilize PySpark’s Word2Vec, a pre-trained BERT-based model called MuRIL, and Keras’s Embedding Layer to produce sentence-level embeddings for each couplet. By doing extensive experiments and utilizing advanced approaches including Stochastic Gradient Descent, Random Forest, and 1D Convolutional Neural Networks, we have achieved a test accuracy of 87.2%. Our study is the first to focus on Sindhi poetry, as to the best of our knowledge, no prior work has been done in this field.