Figure 4.

Coverage of risk vocabularies in news headlines. A. Proportion of news headlines ($n=15,031$) covered, where coverage is defined at different thresholds (at least one word, two words, or three words can be predicted by the model). Vocabularies include those of [7] (30 risks), [15] (428 risks), Basel Risk Norms (1004 risks), and GloVe & SWOW (11,895 words). B. UMAP visualization of each headline, with headlines colored according to the mean riskiness of the words contained according to the text and free association ensemble model (GloVe & SWOW). Headlines with mean riskiness greater than an arbitrary threshold are labeled with the riskiest noun or verb in the headline (75 headlines). One label was removed due to its being a trigger word