Automatic Annotation and Opinion Mining of Gujlish Social Media Data: A Supervised Deep Learning Approach for an Under-Resourced Language

The exponential growth of multimedia content has led to an increasing presence of regional code-mixed data. However, the scarcity of language-specific tools and comprehensive corpora limits the effectiveness of opinion mining for under-resourced languages. This study addresses this gap by reviewing state-of-the-art research work and proposing a generic architecture that combines natural language processing, fuzzy logic, and deep learning techniques. An experimental framework was developed to classify opinions expressed in Gujarati–English code‑mixed comments from YouTube comedy videos using a convolutional neural network. To the best of our knowledge, this is the first exploration of Gujlish in the comedy YouTube domain for opinion mining. Key contributions include the introduction of an automatic annotation module and the creation of a Gujlish dataset. Despite a limited dataset size, the proposed model achieved superior results, attaining a maximum accuracy of 81%. The findings indicate that performance can be further improved as more labelled data becomes available. This work demonstrates the capability of AI‑driven systems to improve content development and user engagement on social media platforms. The research presents an innovative tool that bridges linguistic diversity with technological innovation. Its application has the potential to support inclusive strategies for multilingual populations, enabling governments and organizations to address diverse socio-linguistic needs effectively. The results provide novel insights that support the development of optimized approaches for resource-poor languages. This paper highlights the practical advantages of the proposed methodology and outlines potential directions for future enhancements.