Figure 9

Pearson’s r correlation between daily resolution Twitter sentiment time series for each sentiment dictionary. We see that there is strong agreement within dictionaries, with the biggest differences coming from the stop value of \(\Delta h = 0.5\) for labMT and WK. The labMT and OL dictionaries do not strongly disagree with any of the others, while LIWC is the least correlated overall with other dictionaries. labMT and OL correlate strongly with each other, and disagree most with the ANEW, LIWC, and MPQA dictionaries. The two least correlated dictionaries are the LIWC and MPQA dictionaries. Again, since there is no publicly accessible ground truth for Twitter sentiment, we compare dictionaries against the others, and look at the words. With these criteria, we find the labMT and OL dictionaries to be the most robust with Tweets.