Human–AI Coevolution Dynamics dataset

Personality characteristics were derived from the H3P personality modeling framework (Zhou et al., 2025), producing MBTI-oriented personality distributions. Relational and conversational variables were automatically inferred using the Chinese semantic encoder BGE-Large-ZH-v1.5 and Qwen2.5-7B, allowing the extraction of interpersonal attributes such as trust, intimacy, engagement, politeness, humor, warmth, and supportiveness.All variables were normalized into bounded continuous ranges to ensure numerical stability and comparability across analyses.Table 1 summarizes the major categories of social cognitive variables included in the dataset.Table 1.

Categories of social cognitive variables included in the socially annotated interaction dataset. Category Representative Variables HACD-H Construct Emotion angry, fear, happy, neutral, sad, surprise Emotional Dynamics Personality MBTI distributions, latent personality vectors Personality State Relationship trust, intimacy, familiarity, affection, engagement Relational Dynamics Style politeness, warmth, humor, formality, verbosity Social Behavior Interaction Dynamics emotion_shift, emotion_delta_magnitude State Transition Memory memory_decay_score Social Memory Events event_type, event_importance Environmental Context Growth Variables personality_growth, attachment_growth Coevolution Indicators Latent Representation latent personality embeddings State-Space Encoding The resulting dataset transforms conversational interaction into a collection of socially evolving trajectories and provides a suitable empirical foundation for testing the theoretical predictions of HACD-H.