Three datasets reporting unexpected events for everyday scenarios: Over 9000 events human-labelled for overall valence/sentiment, topic category, and relationship to the initial goal of the scenario.

The three datasets described in this paper were collected from online experiments distributed via Prolific.co participant system. Together, the three datasets comprise 9720 text responses of unexpected events participants predicted for everyday scenarios such as going shopping or preparing breakfast. Each event was labelled by at least two independent, human raters on their topic or category (relative to their initial scenario), the valence or sentiment of the event, and whether or not the event mentions words related to the goal stated in the initial scenario. We also include summary data from a pre- and post-test conducted in the course of these experiments, as well as the analysis code in the form of Jupyter Notebooks. We provide this data and relevant code for transparency and reproducibility alongside our paper. The dataset could be useful in training machine learning models on valence/sentiment of everyday unexpected events.