Classifying Telugu Sentiments Using Transformed-Based Methods

In resource-rich languages like English, sentiment analysis is a crucial part. However, because of a number of issues, like an insufficiency of training corpora, poor-resource languages like Telugu have seen little effort. NLP models and the lack of benchmark datasets for assessment. The recent explosion of transformed-based models in NLP makes it possible to achieve remarkable results in these tasks. Though they are giving good results, especially for languages where very less data is available. researchers are becoming more and more involved in investigating the possibilities of transformed based models that have already been pre-trained in multiple languages. The study evaluates the Telugu sentence-level sentiment classification performance of four multilingual pre-trained transformer models:IndicBERT, RoBERTa, DeBERTa, and XLM-RoBERTa. assessed the effectiveness of each of the four models with our datasets, which includes multiple Telugu language domain datasets. We evaluated these models' performance on three distinct datasets and found encouraging results. XLM-RoBERTa classifies binary sentiment with a decent accuracy of$\mathbf{8 0. 0 9} {\%}$. For Telugu language sentiment analysis.