123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475 |
- from langchain.text_splitter import CharacterTextSplitter
- import re
- from typing import List
- from configs.model_config import SENTENCE_SIZE
- class ChineseTextSplitter(CharacterTextSplitter):
- def __init__(self, pdf: bool = False, **kwargs):
- super().__init__(**kwargs)
- self.pdf = pdf
- def split_text1(self, text: str, use_document_segmentation: bool = False) -> List[str]:
- # use_document_segmentation参数指定是否用语义切分文档,此处采取的文档语义分割模型为达摩院开源的nlp_bert_document-segmentation_chinese-base,论文见https://arxiv.org/abs/2107.09278
- # 如果使用模型进行文档语义切分,那么需要安装modelscope[nlp]:pip install "modelscope[nlp]" -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html
- # 考虑到使用了三个模型,可能对于低配置gpu不太友好,因此这里将模型load进cpu计算,有需要的话可以替换device为自己的显卡id
- if self.pdf:
- text = re.sub(r"\n{3,}", "\n", text)
- text = re.sub('\s', ' ', text)
- text = text.replace("\n\n", "")
- if use_document_segmentation:
- result = p(documents=text)
- sent_list = [i for i in result["text"].split("\n\t") if i]
- else:
- sent_sep_pattern = re.compile('([﹒﹔﹖﹗.。!?]["’”」』]{0,2}|(?=["‘“「『]{1,2}|$))') # del :;
- sent_list = []
- for ele in sent_sep_pattern.split(text):
- if sent_sep_pattern.match(ele) and sent_list:
- sent_list[-1] += ele
- elif ele:
- sent_list.append(ele)
- return sent_list
- def split_text(self, text: str, use_document_segmentation: bool = False) -> List[str]:
- if self.pdf:
- text = re.sub(r"\n{3,}", r"\n", text)
- text = re.sub('\s', " ", text)
- text = re.sub("\n\n", "", text)
- if use_document_segmentation:
- from modelscope.pipelines import pipeline
- p = pipeline(
- task="document-segmentation",
- model='damo/nlp_bert_document-segmentation_chinese-base',
- device="cpu")
- result = p(documents=text)
- sent_list = [i for i in result["text"].split("\n\t") if i]
- return sent_list
- else:
- text = re.sub(r'([;;.!?。!?\?])([^”’])', r"\1\n\2", text) # 单字符断句符
- text = re.sub(r'(\.{6})([^"’”」』])', r"\1\n\2", text) # 英文省略号
- text = re.sub(r'(\…{2})([^"’”」』])', r"\1\n\2", text) # 中文省略号
- text = re.sub(r'([;;!?。!?\?]["’”」』]{0,2})([^;;!?,。!?\?])', r'\1\n\2', text)
- # 如果双引号前有终止符,那么双引号才是句子的终点,把分句符\n放到双引号后,注意前面的几句都小心保留了双引号
- text = text.rstrip() # 段尾如果有多余的\n就去掉它
- # 很多规则中会考虑分号;,但是这里我把它忽略不计,破折号、英文双引号等同样忽略,需要的再做些简单调整即可。
- ls = [i for i in text.split("\n") if i]
- for ele in ls:
- if len(ele) > SENTENCE_SIZE:
- ele1 = re.sub(r'([,,.]["’”」』]{0,2})([^,,.])', r'\1\n\2', ele)
- ele1_ls = ele1.split("\n")
- for ele_ele1 in ele1_ls:
- if len(ele_ele1) > SENTENCE_SIZE:
- ele_ele2 = re.sub(r'([\n]{1,}| {2,}["’”」』]{0,2})([^\s])', r'\1\n\2', ele_ele1)
- ele2_ls = ele_ele2.split("\n")
- for ele_ele2 in ele2_ls:
- if len(ele_ele2) > SENTENCE_SIZE:
- ele_ele3 = re.sub('( ["’”」』]{0,2})([^ ])', r'\1\n\2', ele_ele2)
- ele2_id = ele2_ls.index(ele_ele2)
- ele2_ls = ele2_ls[:ele2_id] + [i for i in ele_ele3.split("\n") if i] + ele2_ls[
- ele2_id + 1:]
- ele_id = ele1_ls.index(ele_ele1)
- ele1_ls = ele1_ls[:ele_id] + [i for i in ele2_ls if i] + ele1_ls[ele_id + 1:]
- id = ls.index(ele)
- ls = ls[:id] + [i for i in ele1_ls if i] + ls[id + 1:]
- return ls
|