文本分段的两个指标

Posted 2020-10-09 狂徒归来

1. Pk

   •  1 def pk(ref, hyp, k=None, boundary=‘1‘):
      2     """
      3     Compute the Pk metric for a pair of segmentations A segmentation
      4     is any sequence over a vocabulary of two items (e.g. "0", "1"),
      5     where the specified boundary value is used to mark the edge of a
      6     segmentation.
      7 
      8     >>> ‘%.2f‘ % pk(‘0100‘*100, ‘1‘*400, 2)
      9     ‘0.50‘
     10     >>> ‘%.2f‘ % pk(‘0100‘*100, ‘0‘*400, 2)
     11     ‘0.50‘
     12     >>> ‘%.2f‘ % pk(‘0100‘*100, ‘0100‘*100, 2)
     13     ‘0.00‘
     14 
     15     :param ref: the reference segmentation
     16     :type ref: str or list
     17     :param hyp: the segmentation to evaluate
     18     :type hyp: str or list
     19     :param k: window size, if None, set to half of the average reference segment length
     20     :type boundary: str or int or bool
     21     :param boundary: boundary value
     22     :type boundary: str or int or bool
     23     :rtype: float
     24     """
     25 
     26     if k is None:
     27         k = int(round(len(ref) / (ref.count(boundary) * 2.)))
     28 
     29     err = 0
     30     for i in range(len(ref)-k +1):
     31         r = ref[i:i+k].count(boundary) > 0
     32         h = hyp[i:i+k].count(boundary) > 0
     33         if r != h:
     34            err += 1
     35     return err / (len(ref)-k +1.)

      

2. WindowDiff
   • 　　

      1 def windowdiff(seg1, seg2, k, boundary="1", weighted=False):
      2     """
      3     Compute the windowdiff score for a pair of segmentations.  A
      4     segmentation is any sequence over a vocabulary of two items
      5     (e.g. "0", "1"), where the specified boundary value is used to
      6     mark the edge of a segmentation.
      7 
      8         >>> s1 = "000100000010"
      9         >>> s2 = "000010000100"
     10         >>> s3 = "100000010000"
     11         >>> ‘%.2f‘ % windowdiff(s1, s1, 3)
     12         ‘0.00‘
     13         >>> ‘%.2f‘ % windowdiff(s1, s2, 3)
     14         ‘0.30‘
     15         >>> ‘%.2f‘ % windowdiff(s2, s3, 3)
     16         ‘0.80‘
     17 
     18     :param seg1: a segmentation
     19     :type seg1: str or list
     20     :param seg2: a segmentation
     21     :type seg2: str or list
     22     :param k: window width
     23     :type k: int
     24     :param boundary: boundary value
     25     :type boundary: str or int or bool
     26     :param weighted: use the weighted variant of windowdiff
     27     :type weighted: boolean
     28     :rtype: float
     29     """
     30 
     31     if len(seg1) != len(seg2):
     32         raise ValueError("Segmentations have unequal length")
     33     if k > len(seg1):
     34         raise ValueError("Window width k should be smaller or equal than segmentation lengths")
     35     wd = 0
     36     for i in range(len(seg1) - k + 1):
     37         ndiff = abs(seg1[i:i + k].count(boundary) - seg2[i:i + k].count(boundary))
     38         if weighted:
     39             wd += ndiff
     40         else:
     41             wd += min(1, ndiff)
     42     return wd / (len(seg1) - k + 1.)

     这两个指标观看文献，还真是有点玄学！还好，找到了nltk中对应的实现，极其简单明了！