Figure 8

An intuitive example of how the unlabeled examples could be helpful. This figure illustrates the potential benefits of using unlabeled examples when the ranking function is run. When only positive examples are used (blue crosses), a machine learning method may define a region surrounding them as the region of positive examples (in dotted lines). Using in addition unlabeled examples may suggest a different region, such as a frontier between positive and negative examples in a region with few points.