Sir M Visvesvaraya Institute of Technology: Kannada Sentiment Analysis Using Language Models

Introduction

Sir M Visvesvaraya Institute of Technology is advancing 2026 Natural Language Processing (NLP) by developing specialised sentiment analysis models for the Kannada language. As regional language content explodes on social media, the ability to accurately interpret public opinion in Kannada is vital for governance, marketing, and social research.

Breaking the Language Barrier in AI

Low-resource languages like Kannada often struggle with standard NLP tools designed primarily for English. By fine-tuning Transformer-based models on curated Kannada datasets, researchers can capture the nuances, slang, and cultural context necessary for high-accuracy emotion detection and opinion mining.

Fine-Tuning BERT for Morphologically Rich Languages

Kannada’s complex morphology requires sophisticated tokenisation strategies to handle its agglutinative nature. Sir M Visvesvaraya Institute of Technology utilises multilingual BERT (mBERT) and IndicBERT, fine-tuned on thousands of labelled Kannada movie reviews and news comments.

Implementation of Byte Pair Encoding (BPE) to handle out-of-vocabulary Kannada words.

Use of attention maps to identify key emotional triggers in long-form Kannada text.

Balancing of datasets to ensure neutral, positive, and negative sentiments are equally represented.

Code-Mixing and Transliteration

Sir M Visvesvaraya Institute of Technology research addresses the challenge of "Manglish" (Kannada written in English script) and code-mixed content prevalent on platforms like Twitter and Instagram. Developing models that understand mixed-language syntax is essential for accurate real-world sentiment tracking.

Creation of synthetic code-mixed datasets for robust model training.

Transliteration layers that convert Roman-script Kannada back to its native phonetics.

Handling of emojis and regional internet slang unique to the Karnataka digital ecosystem.

Applications in Public Governance and Branding

The output of these sentiment models allows government agencies to monitor public reaction to new policies in real-time. Similarly, businesses can use these tools to understand consumer feedback in regional markets, allowing for more personalised and culturally relevant communication strategies.

Real-time dashboarding of public sentiment during regional elections or festivals.

Automated filtering of hate speech and offensive content in Kannada forums.

Sentiment-driven recommendation systems for Kannada-language OTT platforms.

Conclusion

Sir M Visvesvaraya Institute of Technology provides a powerful value proposition by bridging the digital divide for regional languages through advanced AI. This research ensures that Kannada speakers are better understood by technology, fostering more inclusive digital services and data-driven insights for the state of Karnataka.