Combining opinion and structural similarity in link recommendations to counter extreme polarization

This feed contains five recent arXiv papers addressing recommender-system dynamics, inference-stability diagnostics, LLM-augmented crowd moderation for health misinformation, heterogeneous interaction network analytics for learning, and integrating behavioral experiments into diffusion models. Key findings: (1) link-recommendation based on opinion and structural similarity can each produce polarization but differ in network fragmentation and opinion diversity; modest structural-similarity bias can reduce fragmentation and foster moderation; (2) the Inference Headroom Ratio (IHR) is a dimension