The Decay of Impact with Network Distance in Linear Diffusion Processes

This document is an arXiv CS/Social Informatics RSS batch (multiple new preprints) covering network science, social influence, NLP for regulatory documents, misinformation, and applied methods/datasets. Key items: (1) a theoretical and empirical analysis showing total impact in linear diffusion/social-influence models decays approximately exponentially with graph distance and can be approximated by eigenvector-centrality–weighted spectral expressions; (2) DetailBench and DetailDPO, a benchmark and targeted preference-optimization method that reduces LLM "detail hallucinations" in long-regtext/