Figure 12.

Confidence scores of the FastText-LID neural network predictions for the month before and after the shift to 280 characters. We categorize messages into four classes based on the confidence scores we get from FastText-LID’s neural network. Predictions with confidence scores below 0.25 are labeled as Undefined (und). Messages with scores greater or equal to 0.25 but less than 0.5 are flagged as predictions with low confidence (low). Predictions that have scores in the range $[0.5,0.75)$ are considered moderate (mid), and messages with higher scores are labeled as predictions with high confidence (high). We note a symmetry indicating that the shift did not have a large impact on the network’s predictions across organic and retweeted messages