An Analysis on Transliterated Dravidian Text

The rapid growth and accessibility of social media platforms have led to an exponential increase in content generation across multilingual code-mixed languages. Sentiment analysis on multilingual inputs presents significant challenges due to the absence of well-developed corpora, standardized script systems, and the widespread use of transliterated text. In language processing, transliteration enables users to interact with original scripts through more familiar scripts, such as the Latin alphabet. However, it introduces complexities like out-of-vocabulary words, dialectal variations, and orthographic inconsistencies. To bridge this gap, we present a comprehensive study on sentiment analysis of transliterated Dravidian languages. Utilizing online discourse platforms, we developed a large-scale corpus consisting of 26,000 examples and constructed two models. The first model is trained to classify text into three categories, while the second model generates descriptive sentiment labels, offering deeper insights into user opinions. The presented work has acquired an accuracy of 89% and has implications for businesses, policymakers, and researchers seeking to analyze and understand public opinion in Kannada-speaking regions.