2026-10-14T23:42:54Z https://www.scidb.cn/oai

10.57760/sciencedb.j00133.00696 2026-06-29T08:58:15Z

2026-06-29 Supporting Dataset for Latent Collaboration States in User-LLM Multi-Turn Dialogues doi:10.57760/sciencedb.j00133.00696 en This dataset is the supporting data for the paper "Potential Collaborative State Recognition and Dynamic Evolution in User Large Language Model Multi round Dialogue: Empirical Study Based on Hidden Markov Model". The data is sourced from the third-party public dataset WildChat-4.8M, which was formed through sample screening, text feature extraction, anonymization, and de textualization processing. The dataset contains 4219 Chinese multi round user large language model dialogues, 34544 user round sub observation units, and provides tables of round level derived variables, five state decoding results, state labels, model selection results, state emission mean, baseline transition matrix, task heterogeneity, and state output quality required for Hidden Markov Model analysis. The main variables include instruction detail, instruction strength, context dependency, constraint clarity, current round target alignment, current round hard boundary violation, lag one period target alignment, lag one period hard boundary violation, and HMM decoding status. This dataset does not contain sensitive information such as raw user input text, AI reply text, raw conversation ID, message ID, hash value, timestamp, geographic location, or source file path. It only publicly supports de identified derived data and model results for paper conclusion verification and repeated validation.  Large language models; Multi-turn dialogue; Hidden Markov model bao qi kai PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank