Behavioral Coding for 1,454 AI Chat Turns on Cultural Heritage Topics

1,146 heritage-related conversations were screened from three public AI conversation corpora, with 1,454 user turns across 396 relevant conversations coded into eight behavioral categories. Two independent coders applied a scheme including Factual Inquiry and Creative Engagement, with their assignments provided for reliability verification. The dataset was created by Huining Guan and last updated in June 2026.

Use Cases

• Analyzing the distribution of user intent categories like Factual Inquiry and Guidance Seeking in heritage-related AI chats.
• Training or evaluating models to classify conversational behavior based on the eight-category coding scheme.
• Studying inter-coder reliability for behavioral annotation tasks using the provided independent coder assignments.
• Investigating the nature of user engagement (e.g., Creative Engagement, Personal Expression) with AI on cultural topics.

Strengths

• Includes independent coding assignments from two coders for 1,454 user turns, enabling reliability analysis.
• Provides a complete codebook with category definitions, decision rules, and examples for the eight behavioral categories.
• Screens a defined corpus of 1,146 heritage-related conversations from three distinct public sources.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the primary coded data is unknown, which may limit suitability assessment.
• The original conversation text is not included; researchers must obtain source datasets independently.

Provenance

Source

Conversations extracted from WildChat, LMArena, and ShareChat public corpora.

Collection Method

Heritage-related conversations were extracted using a keyword list, then manually coded using a deductive behavioral scheme.

Freshness

Last updated 2026-06-04 15:26:12; freshness should be verified.