Annotation of Positive vs Negative User Interactions for Social Sign Prediction

Collection of recent arXiv CS (social/information) papers covering: zero-shot LLM annotation of interaction-level relational signals for sign prediction; RecLM, a unified framework that eliminates out-of-domain recommendations (includes Microsoft/RecAI GitHub link); reviews and methods for network model selection; RedditPersona, a modular Reddit-derived pipeline that trains QLoRA adapters per community (16M+ comments, user profiling); an audit framework showing LLM-based stance simulation is highly context-sensitive (risk of manipulated simulated stances); analysis of Mastodon governance at-s/