Rumor Detection Platform

Abstract

The rapid spread of rumors across digital platforms poses serious threats to public health, political stability, and social trust. Existing detection approaches, whether man ual fact-checking, automated AI models, or community moderation, each address the problem only partially, suffering respectively from lack of scalability, limited explainabil ity, and vulnerability to manipulation. This thesis presents the design and implementa tion of Rumify, a hybrid rumor verification platform that integrates three complementary layers: an AI-based analysis module combining a transformer text encoder and a GRU based source credibility model to generate preliminary credibility and certainty scores, a community-based verification module where user contributions are weighted by a dynamic reputation system, and an expert validation module that issues final authoritative ver dicts in a transparent and traceable manner while automatically recalibrating community reputation scores based on the alignment of user evaluations with the final decision.. A key contribution of this work is an original trust-weighted ranking algorithm extending the IARR method with Bayesian averaging to robustly combine machine learning pre dictions with community credibility signals. A multi-model feedback quality estimation architecture is also proposed to evaluate user commentary based on text correctness, thor oughness, and reasoning quality. The developed prototype demonstrates the feasibility of integrating AI analysis, participatory verification, and expert oversight within a unified verification pipeline.