Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{209705,
        author = {Swapnali Purushottam Kulthe and Prof. Shubhangi S. Mandwale and Prof. Abhijeet Khadke},
        title = {Emoji Aware Multilingual Sentiment Analysis for Marathi and Hindi},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {5},
        pages = {3012-3018},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209705},
        abstract = {This paper presents an emoji-aware multilingual sentiment-analysis system for Marathi and Hindi short text. The system preserves Unicode pictographs, enriches them with descriptive tokens, and compares three modelling families under a common data split and interface: TF-IDF with a linear support-vector machine, fastText representations with logistic regression, and the MuRIL Transformer. On the Marathi held-out set, the models obtained accuracy and macro-F1 scores of 0.61/0.61, 0.65/0.65, and 0.804/0.803, respectively. On the Hindi IITP data, MuRIL achieved 0.6419 accuracy and 0.5507 macro-F1, exceeding TF-IDF and fastText. The results show that contextual pre-training is most effective overall, while classical and subword baselines remain valuable when speed, storage, and interpretability matter. A Streamlit dashboard supports training, batch prediction, confidence inspection, confusion matrices, and artifact-isolated model comparison. The study also identifies the limits imposed by small samples, domain shift, class imbalance, code-mixing, and context-dependent emoji meaning.},
        keywords = {sentiment analysis, emoji, Marathi, Hindi, MuRIL, fastText, TF-IDF, multilingual NLP},
        month = {October},
        }