Bag of N-grams

An $n$-gram is a sequence of $n$ tokens (words). The classic Bag of Words (BoW) representation uses unigrams, meaning single words. The bag-of-$n$-grams is the generalization for different $n$s.

An example of bigrams ($n=2$)

The $n$ is an hyperparameter, so in order to get the best value we can just try different $n$s and see which is the best result. Typically the best $n$ is $2$ or $3$.

The $n$-gram splitting can also be used at the character level instead of word level, meaning the word is splitted into sequences of $n$ characters instead of words.

---

nlp