This data structure looks like a good solution for multi-gram vectors. Essentially you parse a text and create strings of n features. The first level of the trie is made up of the first feature in each string. The subsequent levels are are possible features that follow the features before it in a string, as well as a count for that n-gram.
This data structure looks like a good solution for multi-gram vectors. Essentially you parse a text and create strings of n features. The first level of the trie is made up of the first feature in each string. The subsequent levels are are possible features that follow the features before it in a string, as well as a count for that n-gram.