FakeNewsChallenge/fnc-1-baseline

Implements stance classification between headlines and article bodies using k-fold cross-validation on the FNC-1 dataset, with utilities for data splitting, feature extraction, and evaluation via a custom scoring function that weights correct predictions differently across stance classes (agree, disagree, discuss, unrelated). The baseline achieves 75.20% on the competition leaderboard using simple hand-crafted features, providing a reference point for more advanced approaches.

139 stars. No commits in the last 6 months.