This is a dataset from the paper "The Validity of Sentiment Analysis: Comparing Manual Annotation, Crowd-Coding, Dictionary Approaches, and Machine Learning Algorithms." The data frame contains four columns: id (identifier), headline (the actual text data), value (sentiment: 0 Neutral, +1 Positive, -1 Negative), gold (whether or not this row is "gold standard", i.e. test set). The data is available from Wouter van Atteveldt's Github. https://github.com/vanatteveldt/ecosent