Here we present an interactive version of labMT happiness scores versus ambient happiness scores given in Figure 1 of the main paper, and ambient happiness time series for every word in the labMT dataset.
Solicited public opinion surveys reach a limited subpopulation of willing participants and are expensive to conduct, leading to poor time resolution and a restricted pool of expert-chosen survey topics. In this study, we demonstrate that unsolicited public opinion polling through sentiment analysis applied to Twitter correlates well with a set of traditional survey measures. We also examine Twitter's potential to canvas topics seldom surveyed, including ideas, personal feelings, and perceptions of commercial enterprises. Two of our major observations are that appropriately filtered Twitter sentiment (1) predicts President Obama's job approval three months in advance, and (2) correlates well with surveyed consumer sentiment. To make possible a full examination of our work and to enable others' research, black we make public a massive data set consisting of co-occurence counts associated with 10,000 terms, each a seven-year timeseries of daily word counts for tweets containing a frequently used search term.