Figure 4. Effect of changing PBDB training database size on PDD quality.

Spearman rho is correlation between human- and machine- generated time series of diversity, as in Fig. 2c. Tuples refers to the number of human-constructed relationships between entities (i.e., the Waldron Formation is Silurian in age) that were used for distant supervision. References refers to the number of individual published papers that were read and processed by human and then used for distant supervision. Each curve was generated by subsampling to the specified number of human-processed references and tuples used for distant supervision.