Multimodal Sentiment Analysis in Indian Regional Languages: A Comprehensive Literature Review and Unified Framework

Abstract

Sentiment analysis of Indian regional language has been of growing significance as a result of the rising development of social media content in languages other than English that are linguistically diverse, code mixed and multimodal in nature. Although there has been significant improvement in high- resource languages, sentiment analysis in Indian settings is not an easy task because of the lack of annotated information, the prevalence of code-mixing, inconsistency in transliteration, and increased popularity of multimodal content like memes and short videos. Available literature tends to solve these issues one at a time and therefore, either unimodal text processing, code-mixed processing, or multimodal sentiment detection, which leaves the solutions isolated and task-specific. The current paper provides a syntactic summary of the sentiment analysis studies on the Indian regional and code-mixed languages, which critically examines the methodological underpinnings, empirical results and the new trends in unimodal, code-mixed and multimodal contexts. The review reveals some important insights into the effectiveness of classical machine learning in the case of data deficiency, the use of multilingual and language-specific transformers, and the increased relevance of large language models with parameter-efficient fine-tuning. The paper suggests a single multimodal sentiment analysis system based on the gaps in the research that can be used to jointly model the textual, visual, and acoustic information with a common semantic structure. In the proposed architecture, the multilingual transformers, parameter-efficient adaptation and cross-modal attention-related fusion are proposed to outstandingly capture the sentiment articulated in different modalities of heterogeneity. This work summarizes the existing literature and offers a generalizable and scalable architectural framework that can guide the development of sentiment analysis in the Indian regional and code-mixed setting, as well as offer theoretical knowledge on the topic. The suggested methodology provides a solid basis of the future empirical assessment and practical implementation of multi-modal sentiment analysis systems.