VerifyLK

BSc (Hons) Computer Science | Final Year Project

Harini De Mel

“Misinformation spreads faster than fact-checkers can respond. Existing automated fact-checking systems fail where it matters most: in politically polarized countries where media outlets exhibit systematic ideological biases. When citizens encounter contradictory claims from pro-government and opposition sources, current systems provide only a final verdict—not the reasoning needed for informed judgment. VerifyLK addresses this gap by making source bias visible and measurable. The system detects when media outlets apply double standards (skeptical of opposition claims, credulous of government ones) and adjusts evidence weighting proportionally. Rather than dismissing biased sources entirely, VerifyLK treats bias as context-dependent: the same outlet might be reliable on economic data but biased on political statements. How it works: VerifyLK profiles news outlet biases using 2,247 Sri Lankan articles, then verifies claims by retrieving evidence from 300k+ government documents and live web search. Six factors determine source weights: credibility, verification confidence, bias-claim alignment, source authority, temporal relevance, and time sensitivity. The result: transparent, explainable verdicts that citizens understand. Results: On real-world Sri Lankan political claims, VerifyLK achieved 87.3% accuracy with perfect precision on confident verdicts. Expert validation (4.5/5 domain experts, 5/5 technical experts) confirms framework appropriateness for polarized contexts. Impact: Citizens in democracies like Sri Lanka can finally verify claims while understanding source bias patterns. Framework applicable to any polarized media environment: Brazil, India, Philippines, USA.”