Anti-social Behavior Detection using Multi-lingual Model
Int'l Conf. on Advancements in Computational Sciences (ICACS), IEEE, pp. 1–9, Lahore, Pakistan
Natural Language Processing · Transformers
Sentiment analysis of Roman Urdu-English and Roman Hindi-English social media text using mBERT and XLM-R, applied to detecting anti-social behavior.
Abstract
Social media has become a vital communication platform between people and communities — but with billions of users, it has grown more challenging to stop hateful, abusive, or offensive content spread by extremists, various aspects of Anti-social Behavior (ASB). Users across South Asia commonly mix native, local, and other languages — Roman Urdu-English and Roman Hindi-English being the two most common on social media in the region — making multilingual ASB detection a significant area of interest for social platforms.
We perform sentiment analysis of Roman Urdu-English and Roman Hindi-English text using transformer-based mBERT and XLM-R models, then process the negatively classified sequences to detect anti-social behavior.
Proposed Framework
People
Related Publications