Evaluaion mark in natural language processing field//updating

1.Bleu

Bleu[1] yes IBM stay 2002 Proposed , For the evaluation of machine translation tasks , Published in ACL, Number of citations 10000+, The original title is “BLEU: a Method for Automatic Evaluation of Machine Translation”.

Its overall idea is accuracy , If a standard translation is given reference, The sentences generated by neural network are candidate, The length of the sentence is n,candidate There is m A word appears in reference,m/n Namely bleu Of 1-gram Calculation formula .

BLEU There are many variants . according to n-gram Can be divided into a variety of evaluation indicators , Common indicators are BLEU-1、BLEU-2、BLEU-3、BLEU-4 Four kinds of , among n-gram It refers to the number of consecutive words n.

BLEU-1 It measures the accuracy of the word level , Higher level bleu It can measure the fluency of sentences .

BLEU Detailed explanation - You know

2.n-gram

N-gram Algorithm _Rnan-prince The blog of -CSDN Blog _n-gram Calculation

Natural language processing N-gram_ Zhang gongtiandao's blog -CSDN Blog _n-gram