Weak Supervision and Transformed-based Sentiment Analysis on Multi-lingual Data

Sentiment analysis on social media is an important research field with applications in many domains. This can be leveraged to predict unrest situations, mob violence, etc., on the occurrence of an event concerning national security. The paper aims to develop an automated intelligent framework to detect the sentiments of users in a multilingual and geospatial manner. This paper covers an important case study to obtain a comprehensive dataset in three languages, i.e., English, Hindi, and Urdu, from three regions. An automated approach for data annotation using weak supervision is presented to highlight the challenge of obtaining training data. We propose a BERT-based transformer model to effectively identify the sentiment considering the context. The study also presents an interactive visualization of results to obtain valuable insights.