Skip to main content
. 2009 May 8;37(Web Server issue):W141–W146. doi: 10.1093/nar/gkp353

Figure 2.

Figure 2.

Parameter estimation. The number of abstracts in the background and training sets has an impact on the ROC area for various biomedical topics. The y-axis shows the mean ROC area after leave-one-out cross validations over 10 random background sets using 1000 training set abstracts (left column), or 10 bootstrapped training sets using the rest of Medline as background set (right column).