Fig. 1

Overview of the Snorkel pipeline. First, users input a text corpus. Snorkel extracts relationships of interest by 1) detecting co-occurring entities (i.e., relationship candidates), 2) applying labeling functions to automatically label noisy training examples, and 3) training generative and discriminative machine learning models using the labeling functions and general features to predict which candidates are true relationships. The output is a binary prediction of a true relationship for each relationship candidate