Definition
A quantitative measure of how well a probability model, particularly a language model, predicts a sample; a lower perplexity score indicates a better model.
In short: A quantitative measure of how well a probability model, particularly a language model, predicts a sample; a lower perplexity score indicates a better model.