A Contextual Prediction Network With Dynamic Neural Alignment For Sentiment Analysis in Tamil and Malayalam Texts

Abstract

The rapid growthin the popularity of social media has generated an enormous user generated content in the Dravidian languages such as Tamil and Malayalam. Due to its informality, the regular switching of codes to the English language, and the fact that the text is very noisy, the analysis of the sentiment has become very hard. The end goal of this project is to develop a robust and contextual sentiment analysis system that is capable of withstanding the language richness, the contextual ambiguity, as well as the annotated data sparsity on the low resource languages. In order to achieve that, the researchproposed a Contextual Prediction Network with Dynamic Neural Alignment (CPN-DNA) algorithm, according to prediction to enforce dynamic features alignment, feature-level attention,sample-level attention and enhanced contextual consistency. The proposed model combines text representations and contextual feature injection as well as a context-sensitive loss to achieve better stability and perceptiveness.The proposed method is effective as it provides higher accuracy rates of 98.80 and 98.50 with Tamil and Malayalam benchmark data’s respectively, when compared to the traditional machine learning, deep learning, and transformer-based baselines. Other advancements in the recall and F1-score are also indicative of balanced and consistent forecasts of sentiments. In general, the proposed CPN-DNA system represents a plausible, coherent, and comprehensible sentiment analysis system in Tamil and Malayalam, which can be crudely extrapolated to the context of social media analytics of low-resource and code-mixed language settings.