Learning From Pairwise Preferences: An Introduction to the Bradley Terry Model

The Bradley-Terry model transforms simple head-to-head comparisons, such as "A is better than B," into a coherent probabilistic ranking. It operates by assigning each item a latent "strength" score, where the probability of one item winning depends on the relative difference between the two strengths. These hidden scores are learned from observed win-loss data by finding the values that make the actual outcomes most likely, aligning the model's predictions with reality. This approach is powerful because humans are often more consistent when making relative judgments than when assigning absolute scores, which avoids issues with inconsistent rating scales.