In her doctoral dissertation, MSc (Tech) Tanvi Sharma develops a new artificial intelligence framework for analysing misinformation by identifying and classifying relationships between people, organisations, events, diseases, treatments, products and public policies. Instead of treating information simply as true or false, the method distinguishes between multiple types of relationships and misinformation, providing a more detailed and interpretable picture of how misleading claims are constructed and spread. The research introduces new datasets for COVID-19 misinformation and relation extraction and shows that transformer-based deep learning models outperform conventional machine learning approaches. The framework is further extended to domains such as healthcare, politics, finance, crime, and entertainment using semi-supervised learning, topic modelling, and novel AI architectures, including emerging state-space models such as Mamba, as well as Hybrid Transformer-Mamba model. The results demonstrate that misinformation can be quantified more accurately, efficiently, and across different domains at scale.
The doctoral dissertation of MSc (Tech) Tanvi Sharma titled Multiclass Relation Extraction with Deep Learning Models will be publicly examined via Zoom on 7 October 2026. The dissertation is in the field of data science.
The Opponent will be Professor Kakoli Banerjee from JSS University, India. The Custos will be Professor Konstantinos Stefanidis from the Faculty of Information Technology and Communication Sciences.