Langchain-Chatchat/textsplitter/chinese_text_splitter.py

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:
            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:
            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