Deep Learning based Poet Attribution model for Punjabi Poetry

Poetry is a profound medium for expressing emotions and beliefs, with South Asian languages like Urdu, Hindi, and Punjabi boasting rich poetic traditions. The digital age and the rise of social media have amplified the visibility of these poetic works but also heightened instances of plagiarism and misattribution. In this paper, we present a Deep Learning model for Poet Attribution for Punjabi poetry using Shahmukhi, Gurmukhi, and Roman scripts. Our dataset consists of 830 poems from 11 different poets. We utilize Multilingual DistilBERT to generate the embedding of each poem in the 768-dimensional vector space. By conducting extensive experiments and utilizing advanced approaches, including Bi-LSTM and Bi-GRU we obtained a test accuracy of $91.57 \%$ on the Roman script. Additionally, we achieved accuracies of $90.36 \%$ and $87.95 \%$ on the Gurmukhi and Shahmukhi scripts, respectively.