EXPLORING WORD EMBEDDINGS FOR SENTIMENT ANALYSIS OF MARATHI POLITICAL TWEETS: A MACHINE LEARNING APPROACH

Sentiment analysis of textual data is becoming increasingly significant in research. Many researchers are developing new technologies to enhance the accuracy and performance of sentiment analysis. This process is particularly vital in analysing customer reviews across various domains. One of new domain which was explored by the researchers is Political domain. After the inception of Smartphones and Internet availability, various political parties are using the social media to influence the people. As every people has their own opinion related to political context, they always try to put it on various social media handles like Facebook, Twitter (changed to X), Instagram, YouTube etc. As there is lots of research and resources carried out for few languages such as English, Chinese, Arabic but still few languages still lag in it like Marathi, Gujrati, Telegu, Greek etc. In view of this we have done the sentiment analysis for Marathi Tweets related to political domain using various ML models and Word embedding techniques like FastText, IndicNLP, Bag of Words and TF-IDF. We employed the hyperparameter tuning to optimize each model’s performance. Among the tested embeddings, IndicNLP proved most effective, yielding superior accuracy and robustness across different machine learning models, likely due to its ability to capture linguistic intricacies specific to Indian languages. Our findings highlight the effectiveness of advanced word embeddings like IndicNLP in sentiment analysis tasks for under-resourced languages like Marathi, demonstrating their potential for broader applications in regional language processing.