LLM_文本生成评估指标

一、BLEU (precision-based metric)

• 评估准确率: 和准确率类似-生成的多少词出现在reference的词中
• 引起问题1：
  • 如果生成重复的词，并且该词在引用中出现，那么我们会得到较高的分数
  • 针对这点作者指出修正方法：一个单词只计算它在引用中出现的次数。
    • example:
      • ref-“the cat is on the mat”
      • g-“the the the the the the”
    • $P_{vanlilla}=\frac{6}{6}$, $P_{mod}=\frac{2}{6}$
• 修正问题1：clip
  • 这意味着一个n-gram的出现次数以它在参考句中出现的次数为上限

$$p_n=\frac{{\sum }_{geSnt\in C}{\sum }_{n-gram\in geSnt}Coun{t}_{clip}(n-gram)}{{\sum }_{geSnt\in C}{\sum }_{n-gram\in geSnt}Count(n-gram)}$$

• 引起问题2：
  • 因为这个准确率的评估，很显然会对较短的评估对有力，会低估较长生成的结果。
• 修正问题2：简短惩罚 brevity penalty
  • $BR=min(1,e^{1-\frac{l_{ref}}{l_{gen}}})$ : 生成长度大于原句子：1, 生成长度小于原句子：$(0,1)$,

最终公式:

$$BLEU-N=BR\ast (\prod _{n=1}^N{p}_n{)}^{1/N}$$

Example: 计算BLEU-4

• ref-“the cat sat on the mat”
• g-“the cat the cat is on the mat”
• BR: $BR=min(1,e^{1-6/8})=1$
• n=1
  • 1-gram: org:{"the", "cat", "sat", "on", "mat"} ge:{"the", "cat", "is", "on", "mat"}
  • clip: $coun{t}_{clip}("the")=2,coun{t}_{clip}("cat")=1,coun{t}_{clip}("is")=0,1-gram\in geSnt$
  • $p_1=\frac{5}{8}$
• n=2
  • 2-gram: org:{"the cat", "cat sat", "sat on", "on the", "the mat"} ge:{"the cat", "cat the", "cat is", "is on", "on the", "the mat"}
  • $p_2=\frac{3}{7}$
• n=3
  • 3-gram: org:{"the cat sat", "cat sat on", "sat on the", "on the mat"} ge:{"the cat the", "cat the cat", "the cat is", "cat is on", "is on the", "on the mat"}
  • $p_3=\frac{1}{6}$
• n=4
  • 3-gram: org:{"the cat sat on", "cat sat on the", "sat on the mat"} ge:{"the cat the cat", "cat the cat is", "the cat is on", "cat is on the", "is on the mat"}
  • $p_4=\frac{0}{5}$
• BLEU-4: $1\ast (\frac{5}{8}\ast \frac{3}{7}\ast \frac{1}{6}\ast \frac{0}{5}{)}^{1/4}=0.$

1.1 huggingface load_metric 调用sacrebleu

可以看出包内的计算原理同上述

from datasets import load_metric
!pip install sacrebleu
bleu_metric = load_metric("sacrebleu")
bleu_metric.add(prediction="the cat the cat is on the mat", reference=["the cat sat on the mat"])
results = bleu_metric.compute(smooth_method="floor", smooth_value=0)
results
"""
{'score': 0.0,
 'counts': [5, 3, 1, 0],
 'totals': [8, 7, 6, 5],
 'precisions': [62.5, 42.857142857142854, 16.666666666666668, 0.0],
 'bp': 1.0,
 'sys_len': 8,
 'ref_len': 6}
"""

二、 ROUGE (recall-based metric)

• 评估召回: 和召回率类似- reference的词中有多少出现在生成词中

$$ROUGE=\frac{{\sum }_{orgSnt\in C}{\sum }_{n-gram\in orgSnt}Coun{t}_{match}(n-gram)}{{\sum }_{orgSnt\in C}{\sum }_{n-gram\in orgSnt}Count(n-gram)}$$

• 对于最长公共子串longest common substring有个单独的分数ROUGE-L

$$R_{LCS}=\frac{LCS(X,Y)}{m};P_{LCS}=\frac{LCS(X,Y)}{n}$$
$$F_{LCS}=\frac{(1+{\beta }^2)R_{LCS}{P}_{LCS}}{R_{LCS}+\beta P_{LCS}},\beta =\frac{P_{LCS}}{R_{LCS}}$$

Example: 计算 ROUGE1

• ref-“the cat sat on the mat”
• g-“the cat the cat is on the mat”
• 1-gram: org:{“the”, “cat”, “sat”, “on”, “mat”} ge:{“the”, “cat”, “is”, “on”, “mat”}
• $ROUGE-1^r=\frac{2+1+0+1+1}{6}=\frac{5}{6}$
• $BLEU-1^p=\frac{min(3,2)+min(2,1)+0+1+1}{8}=\frac{5}{8}$

1.2 huggingface load_metric 调用sacrebleu

可以看出包内的计算原理同上述

from datasets import load_metric
!pip install rouge_score
rouge_metric = load_metric("rouge")
rouge_metric.add(prediction="the cat the cat is on the mat", reference=["the cat sat on the mat"])
results = rouge_metric.compute()
print(1/(0.5* 1/0.625 + 0.5* 1/0.8333333333333334))
results
"""
0.7142857142857143
{'rouge1': AggregateScore(low=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143), mid=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143), high=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143)),
 'rouge2': AggregateScore(low=Score(precision=0.42857142857142855, recall=0.6, fmeasure=0.5), mid=Score(precision=0.42857142857142855, recall=0.6, fmeasure=0.5), high=Score(precision=0.42857142857142855, recall=0.6, fmeasure=0.5)),
 'rougeL': AggregateScore(low=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143), mid=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143), high=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143)),
 'rougeLsum': AggregateScore(low=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143), mid=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143), high=Score(precision=0.625, recall=0.8333333333333334, fmeasure=0.7142857142857143))}

"""