Human annotators prepare pages and images for machine learning. GumGum gives annotators a Taxonomy (such as IAB v2) and the annotator uses a tool to pick the corresponding categories that best match each page (referred to as Ground-Truth). GumGum also leverages publicly available corpuses of labeled data for model training.
One of the challenges of machine learning is that machine learning algorithms are only as good as the training data they learn from. GumGum continues to make a significant investment in hand-annotations and the overall quality of labeled data sets.