英文特定领域词组切分-借鉴中文的前向，后向，双向最大匹配
2020-01-18 本文已影响0人
IntoTheVoid
由于在不同的领域中，存在一些专有词汇，例如化学领域中，IUPAC的化学式命名，电商领域的商品词，这些词如果按照常规的英文空格切分，往往会给后续的任务带来困扰，因为如果词组被错误的切分会导致其所蕴含的意思发生变化，因此，往往需要借助行业词典，给非结构化文本中的词进行词组切分。
考虑到中文中为了分词的目的，往往会使用最大匹配法进行短语的切分。值的借鉴的是，可以将英文按照空格切分，把每个词当成中文中的一个字，借助相关词典，就可以实现最大匹配分词， 一下是代码的具体实施：
- 词典格式
machine learning
is
the
science
of
getting
computers
to
act
without
being
explicitly programmed
a
method
data analysis
that
automates analytical model building
an application
the ability
- 前向最大匹配, 需要将
./words.dic替换成你自己的词典，词典格式是一行一个词的txt
import re
class ForwardMaxMatch:
def __init__(self, dic='./words.dic', max_len=5):
self.max_len = max_len
self.dic_file = dic
self.dic = self.load_dic()
self.tokens = []
def load_dic(self):
with open(self.dic_file, 'r', encoding='utf-8') as f:
data = f.readlines()
dic = [i.strip() for i in data]
return dic
def segment(self, word_obj):
if isinstance(word_obj, str):
character_map = {".": " . ",
",": ' , '}
for origin, new in character_map.items():
word_obj = word_obj.replace(origin, new)
self.words_list =word_obj.lower().split()
elif isinstance(word_obj, list):
self.words_list = word_obj
else:
print("Not support object for segmentation!")
i = 0
tokens = []
while i < len(self.words_list):
maxWords = []
reverse = self.max_len + i
while reverse > i:
grams = self.words_list[i: reverse]
reverse -= 1
tempWords = ' '.join(grams)
if tempWords in self.dic:
maxWords = grams
break
if maxWords:
i += len(maxWords)
tokens.append(' '.join(maxWords))
else:
tokens.append(' '.join(self.words_list[i: i+1]))
i += 1
self.tokens = tokens
return tokens
def __call__(self, word_obj):
return self.segment(word_obj)
def __repr__(self):
return self.tokens
- 后向最大匹配, 需要将
./words.dic替换成你自己的词典，词典格式是一行一个词的txt
import re
class BackwardMaxMatch:
def __init__(self, dic='./words.dic', max_len=5):
self.max_len = max_len
self.dic_file = dic
self.dic = self.load_dic()
self.tokens = []
def load_dic(self):
with open(self.dic_file, 'r', encoding='utf-8') as f:
data = f.readlines()
dic = [i.strip() for i in data]
return dic
def segment(self, word_obj):
if isinstance(word_obj, str):
character_map = {".": " . ",
",": ' , '}
for origin, new in character_map.items():
word_obj = word_obj.replace(origin, new)
self.words_list =word_obj.lower().split()
elif isinstance(word_obj, list):
self.words_list = word_obj
else:
print("Not support object for segmentation!")
i = -1
tokens = []
while i > -len(self.words_list):
maxWords = []
reverse = -self.max_len + i
while reverse < i:
grams = self.words_list[reverse: i]
reverse += 1
tempWords = ' '.join(grams)
if tempWords in self.dic:
maxWords = grams
break
if maxWords:
i -= len(maxWords)
tokens.append(' '.join(maxWords))
else:
tokens.append(' '.join(self.words_list[i-1: i]))
i -= 1
self.tokens = list(reversed(tokens))
if self.words_list[-1] == '.': self.tokens.append('.')
return self.tokens
def __call__(self, word_obj):
return self.segment(word_obj)
def __repr__(self):
return self.tokens
- 双向最大匹配
————————————————
分词目标：
将正向最大匹配算法和逆向最大匹配算法进行比较，从而确定正确的分词方法。
算法流程：
比较正向最大匹配和逆向最大匹配结果
如果分词数量结果不同，那么取分词数量较少的那个
如果分词数量结果相同
分词结果相同，可以返回任何一个
分词结果不同，返回单字数比较少的那个
原文链接：https://blog.csdn.net/selinda001/article/details/79345072
————————————————
def BidirectMaxMatch(string):
back_tokenizer = BackwardMaxMatch()
back_tokens = back_tokenizer.segment(string)
forward_tokenizer = ForwardMaxMatch()
forward_tokens = forward_tokenizer.segment(string)
if len(back_tokens) == len(forward_tokens):
if back_tokens == forward_tokens:
return back_tokens
else:
back_single_w_cnt = sum([0 if len(bt.split(' ')) > 1 else 1 for bt in back_tokens])
forward_single_w_cnt = sum([0 if len(bt.split(' ')) > 1 else 1 for bt in forward_tokens])
if back_single_w_cnt < forward_single_w_cnt:
return back_tokens
else:
return forward_tokens
else:
if len(back_tokens) < len(forward_tokens):
return back_tokens
else:
return forward_tokens
