Add Semi-Supervised Learning Component

[Answerfxy]

In some scenarios, it has lots of unlabeled data. If we can use these data for model training, it would be great.

Adding semi-supervised learning component has two advantages:

1. Adding unlabeled data can increase the amount of the intersection. Intersection count can impact model evaluation.
2. Some cases only have positive labels, but without negative labels. We couldn't use supervised learning algorithms. So we need semi-supervised learning component to solve such problems.