Center for Artificial Intelligence Research Nepal (CAIR)
Misinformation across academic, healthcare, and political domains represents one of the most pressing challenges to information integrity in the digital age. This research develops a unified detection framework that combines knowledge graph construction with deep neural network methodologies to identify, categorize, and explain misinformation at both national and international scale.
A central commitment of this work is interpretability — every detection decision is accompanied by a transparent, evidence-backed explanation accessible to non-technical stakeholders including journalists, fact-checkers, and policymakers.
Domains
- Academic misinformation
- Healthcare misinformation
- Political misinformation
Geographic Coverage
- National: Nepal and the South Asian region (English and Nepali)
- International: English-language content
Data Sources
- Twitter and Reddit
Data is collected continuously through ethical, rate-limited pipelines and annotated using a hybrid strategy combining large language model labeling with expert validation. Annotated content is used to construct dynamically evolving knowledge graphs that capture entities, claims, relationships, and temporal propagation patterns.
A deep learning framework fuses graph-based and textual representations to perform multi-task learning across domain classification, misinformation categorization, and source credibility assessment. The system is designed for continuous refinement as new data accumulates.
Interpretability is a foundational design requirement, not an afterthought. The framework provides multi-level explanations — from attention mechanisms and feature attribution to natural language summaries — ensuring detection decisions can be understood, validated, and challenged by diverse stakeholders.
Active research — ongoing data collection, annotation, and model development.
Samriddha Pathak Center for Artificial Intelligence Research Nepal (CAIR) samriddha.pathak@cair-nepal.org