Raw vs. Filter: Understanding the Difference in Entity Ranking

Raw and Filter results differ in how they handle training data during testing. Filter results exclude samples that appeared in the training set, providing a more accurate evaluation by eliminating bias from entities that were already familiar to the model. This is because the testing phase involves ranking all entities by replacing them, and while high-ranking entities from the training set are acceptable, removing them during testing with Filter offers a more precise view of the model's performance.