Debunking decisions for dynamically emerging topic-associated rumors

In public emergency events, topic-associated rumors typically erupt in a streaming, multi-wave fashion, requiring authorities to continuously produce debunking decisions as new rumors keep emerging, all under unknown diffusion-model parameters. Inspired by Planck’s blackbody radiation law, we propose the Topic-associated Rumor Dynamic-emergence Competitive Cascade (TRDCC) model with a cognitive-fatigue effect, cast dynamic debunking as an online learning problem under the query–decision regression paradigm, and develop the Topic-associated Rumor Dynamic Control (TRDC) algorithm. Theoretically, we establish the NP-hardness of the problem and the submodularity of the objective, and derive a probabilistic approximation bound for the score function as well as a generalization-error upper bound that accounts for state distribution shift. On four real-world networks, TRDC attains time-averaged Performance Ratio (PR) values of 0.859, 0.891, 0.657 and 0.713 even under fully unknown diffusion parameters. Across all four networks, it outperforms the best-performing heuristic, learn-and-optimize and deep reinforcement learning baseline by margins of at least 0.23, 0.25 and 0.28 in PR, respectively, and remains stable throughout the entire decision stream.