Livestreaming pitch- and minute-level dataset

Full description

This dataset accompanies the manuscript “Emotional Valence Transitions in Livestreaming: Elevate Engagement but Sabotage Sales” (In Press).  It contains two linked observational datasets derived from Douyin livestream commerce: a pitch-level dataset and a minute-level dataset, together with the Stata code files used for data preparation and empirical analyses. 

The datasets include measures of livestream performance outcomes, including unit sales, likes, and comments, as well as variables describing emotional valence transition rates derived from frame-level facial-expression recognition. Emotional expressions were extracted using DeepFace and mapped into three valence categories - positive, neutral, and negative - then aggregated from frames to seconds, minutes, and pitches. Based on these classifications, the data include both the total valence transition rate and six direction-specific transition rates: positive-to-neutral, neutral-to-positive, positive-to-negative, negative-to-positive, neutral-to-negative, and negative-to-neutral. Additional variables include product-related characteristics, contextual controls, and product-comment measures used in the main and moderation analyses.