You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
293 lines
12 KiB
293 lines
12 KiB
5 months ago
|
import uuid
|
||
|
|
from langchain_chroma import Chroma
|
||
|
|
from langchain_community.embeddings import DashScopeEmbeddings
|
||
|
|
from langchain_community.document_loaders import TextLoader
|
||
|
|
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
||
|
|
from qwen_agent.agents import Assistant
|
||
|
|
import json_repair
|
||
|
|
from paddlenlp import Taskflow
|
||
|
|
embeddings = DashScopeEmbeddings(dashscope_api_key="sk-ea89cf04431645b185990b8af8c9bb13")
|
||
|
|
device_id=0
|
||
|
|
import re
|
||
|
|
import time
|
||
|
|
from docx import Document
|
||
|
|
import shutil
|
||
|
|
from docx.opc.pkgreader import _SerializedRelationships, _SerializedRelationship
|
||
|
|
from docx.opc.oxml import parse_xml
|
||
|
|
import logging
|
||
|
|
import logging.config
|
||
|
|
|
||
|
|
log_config = {
|
||
|
|
'version': 1,
|
||
|
|
'disable_existing_loggers': False,
|
||
|
|
'formatters': {
|
||
|
|
'standard': {
|
||
|
|
'format': '%(asctime)s - %(name)s - %(levelname)s - %(message)s',
|
||
|
|
},
|
||
|
|
},
|
||
|
|
'handlers': {
|
||
|
|
'console': {
|
||
|
|
'class': 'logging.StreamHandler',
|
||
|
|
'formatter': 'standard',
|
||
|
|
'level': logging.INFO,
|
||
|
|
},
|
||
|
|
'file': {
|
||
|
|
'class': 'logging.FileHandler',
|
||
|
|
'filename': 'Logger.log',
|
||
|
|
'formatter': 'standard',
|
||
|
|
'level': logging.INFO,
|
||
|
|
},
|
||
|
|
},
|
||
|
|
'loggers': {
|
||
|
|
'': {
|
||
|
|
'handlers': ['console', 'file'],
|
||
|
|
'level': logging.INFO,
|
||
|
|
'propagate': True,
|
||
|
|
},
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
logging.config.dictConfig(log_config)
|
||
|
|
|
||
|
|
logger = logging.getLogger("checkRepeatText")
|
||
|
|
|
||
|
|
def load_from_xml_v2(baseURI, rels_item_xml):
|
||
|
|
"""
|
||
|
|
Return |_SerializedRelationships| instance loaded with the
|
||
|
|
relationships contained in *rels_item_xml*. Returns an empty
|
||
|
|
collection if *rels_item_xml* is |None|.
|
||
|
|
"""
|
||
|
|
srels = _SerializedRelationships()
|
||
|
|
if rels_item_xml is not None:
|
||
|
|
rels_elm = parse_xml(rels_item_xml)
|
||
|
|
for rel_elm in rels_elm.Relationship_lst:
|
||
|
|
if rel_elm.target_ref in ('../NULL', 'NULL'):
|
||
|
|
continue
|
||
|
|
srels._srels.append(_SerializedRelationship(baseURI, rel_elm))
|
||
|
|
return srels
|
||
|
|
|
||
|
|
|
||
|
|
_SerializedRelationships.load_from_xml = load_from_xml_v2
|
||
|
|
# 记录程序开始的时间戳
|
||
|
|
def getOutlineLevel(inputXml):
|
||
|
|
"""
|
||
|
|
功能 从xml字段中提取出<w:outlineLvl w:val="number"/>中的数字number
|
||
|
|
参数 inputXml
|
||
|
|
返回 number
|
||
|
|
"""
|
||
|
|
start_index = inputXml.find('<w:outlineLvl')
|
||
|
|
end_index = inputXml.find('>', start_index)
|
||
|
|
number = inputXml[start_index:end_index + 1]
|
||
|
|
number = re.search("\d+", number).group()
|
||
|
|
return number
|
||
|
|
|
||
|
|
|
||
|
|
def isTitle(paragraph):
|
||
|
|
"""
|
||
|
|
功能 判断该段落是否设置了大纲等级
|
||
|
|
参数 paragraph:段落
|
||
|
|
返回 None:普通正文，没有大纲级别 0:一级标题 1:二级标题 2:三级标题
|
||
|
|
"""
|
||
|
|
# 如果是空行，直接返回None
|
||
|
|
if paragraph.text.strip() == '':
|
||
|
|
return None
|
||
|
|
|
||
|
|
# 如果该段落是直接在段落里设置大纲级别的，根据xml判断大纲级别
|
||
|
|
paragraphXml = paragraph._p.xml
|
||
|
|
if paragraphXml.find('<w:outlineLvl') >= 0:
|
||
|
|
return getOutlineLevel(paragraphXml)
|
||
|
|
# 如果该段落是通过样式设置大纲级别的，逐级检索样式及其父样式，判断大纲级别
|
||
|
|
targetStyle = paragraph.style
|
||
|
|
while targetStyle is not None:
|
||
|
|
# 如果在该级style中找到了大纲级别，返回
|
||
|
|
if targetStyle.element.xml.find('<w:outlineLvl') >= 0:
|
||
|
|
return getOutlineLevel(targetStyle.element.xml)
|
||
|
|
else:
|
||
|
|
targetStyle = targetStyle.base_style
|
||
|
|
# 如果在段落、样式里都没有找到大纲级别，返回None
|
||
|
|
return None
|
||
|
|
|
||
|
|
#寻找标题名称
|
||
|
|
def findTitleName(docxPath):
|
||
|
|
yield '文档相似性检查----检查是否存在详细设计方案'
|
||
|
|
document = Document(docxPath)
|
||
|
|
# 逐段读取docx文档的内容
|
||
|
|
titleWords=[]
|
||
|
|
firstTitle = 0
|
||
|
|
secondTitle = 0
|
||
|
|
sanjiTitle = 0
|
||
|
|
for paragraph in document.paragraphs:
|
||
|
|
# 判断该段落的标题级别
|
||
|
|
# 这里用isTitle()临时代表，具体见下文介绍的方法
|
||
|
|
text = paragraph.text
|
||
|
|
if text.strip():#非空判断
|
||
|
|
level = isTitle(paragraph)
|
||
|
|
if level=="0":
|
||
|
|
firstTitle+=1
|
||
|
|
secondTitle = 0
|
||
|
|
if(text.find("附件")>=0):
|
||
|
|
continue
|
||
|
|
titleWords.append("一级标题:".format(firstTitle)+text)
|
||
|
|
elif level=="1":
|
||
|
|
secondTitle+=1
|
||
|
|
sanjiTitle=0
|
||
|
|
# words.append("\t"+"{}.{}".format(firstTitle,secondTitle)+text)
|
||
|
|
# titleWords.append("第{}章的二级标题:".format(firstTitle,firstTitle,secondTitle)+text)
|
||
|
|
elif level=="2":
|
||
|
|
sanjiTitle += 1
|
||
|
|
# words.append("\t"+"{}.{}".format(firstTitle,secondTitle)+text)
|
||
|
|
# titleWords.append("第{}章的三级标题".format(firstTitle, secondTitle,firstTitle, secondTitle,sanjiTitle) + text)
|
||
|
|
findTitleName_llm_cfg = {
|
||
|
|
#'model': 'qwen1.5-72b-chat',
|
||
|
|
'model':"qwen2-72b",
|
||
|
|
'model_server': 'http://127.0.0.1:1025/v1', # base_url, also known as api_base
|
||
|
|
# 'api_key': 'sk-ea89cf04431645b185990b8af8c9bb13',
|
||
|
|
}
|
||
|
|
findTitleName_bot = Assistant(llm=findTitleName_llm_cfg,
|
||
|
|
name='Assistant',
|
||
|
|
# system_message='1：这样的是一级标题。1.1：这样的是二级标题。1.1.1：这样的是三级标题'
|
||
|
|
)
|
||
|
|
prompt='''\n是文档的大纲，一级标题组成，哪一章存在与方案相关的内容
|
||
|
|
类似详细设计方案,详细服务方案，详细建设方案为最相关的，优先选择
|
||
|
|
类似设计方案，服务方案，建设方案为次相关，次级选择
|
||
|
|
类似方案是最后选择
|
||
|
|
按照这样的顺序选择最合适的
|
||
|
|
你只能从这两个答案中选择一个：{"name":"一级标题名称","answer":"存在"}或{"name":"","answer":"不存在"}，不做过多的解释,严格按回答格式作答
|
||
|
|
'''
|
||
|
|
# print("\n".join(titleWords)+prompt)
|
||
|
|
messages = [({'role': 'user', 'content': "\n".join(titleWords)+prompt})]
|
||
|
|
runList=[]
|
||
|
|
for rsp in findTitleName_bot.run(messages):
|
||
|
|
runList.append(rsp)
|
||
|
|
data = runList[len(runList) - 1][0]["content"]
|
||
|
|
parsed_data = json_repair.loads(data.replace('`', ''))
|
||
|
|
logger.info(parsed_data)
|
||
|
|
if(parsed_data["answer"]=="存在"):
|
||
|
|
yield parsed_data["name"]
|
||
|
|
else:
|
||
|
|
yield "文档相似性检查----未找到与详细设计方案相关内容，无法进行相似性比较"
|
||
|
|
#获取文档中 详细设计方案 章节的所有内容
|
||
|
|
def getDocxToText(docxPath,titleName,vector_store_path):
|
||
|
|
document = Document(docxPath)
|
||
|
|
# 逐段读取docx文档的内容
|
||
|
|
levelList=[]
|
||
|
|
words=[]
|
||
|
|
addStart = False
|
||
|
|
levelText=""
|
||
|
|
i = 0
|
||
|
|
for paragraph in document.paragraphs:
|
||
|
|
# 判断该段落的标题级别
|
||
|
|
# 这里用isTitle()临时代表，具体见下文介绍的方法
|
||
|
|
text = paragraph.text
|
||
|
|
if text.strip():#非空判断
|
||
|
|
if titleName:
|
||
|
|
level = isTitle(paragraph)
|
||
|
|
if(addStart and level=="0"):
|
||
|
|
addStart=False
|
||
|
|
if(level=="0" and (titleName.find(text)>=0 or text.find(titleName)>=0)):
|
||
|
|
addStart=True
|
||
|
|
if level:
|
||
|
|
levelList.append("{}：".format(level)+paragraph.text)
|
||
|
|
levelText=f"{int(level)+1}级标题-"+text
|
||
|
|
else:
|
||
|
|
if addStart:
|
||
|
|
if(text.startswith("图") or text.startswith("注：")):
|
||
|
|
continue
|
||
|
|
if(len(text)>30):
|
||
|
|
i=i+1
|
||
|
|
words.append("{}：".format(levelText)+text)
|
||
|
|
# 将所有段落文本拼接成一个字符串，并用换行符分隔
|
||
|
|
if len(words)==0:
|
||
|
|
raise Exception("checkRepeatText，获取长度为0")
|
||
|
|
text = '\n'.join(words)
|
||
|
|
|
||
|
|
# 将文本写入txt文件
|
||
|
|
with open("checkRepeatText.txt", 'w', ) as txt_file:
|
||
|
|
txt_file.write(text)
|
||
|
|
time.sleep(3)
|
||
|
|
loader = TextLoader(file_path='checkRepeatText.txt')
|
||
|
|
docs = loader.load()
|
||
|
|
# print(docs)
|
||
|
|
text_splitter = RecursiveCharacterTextSplitter(chunk_size=100, chunk_overlap=10, add_start_index=True,
|
||
|
|
separators=["\n\n", "\n"])
|
||
|
|
|
||
|
|
splits = text_splitter.split_documents(docs)
|
||
|
|
uuids = []
|
||
|
|
for i in range(len(splits)):
|
||
|
|
uuids.append(str(uuid.uuid4()))
|
||
|
|
logging.info(f"checkRepeatTextuuidLen{len(uuids)}")
|
||
|
|
|
||
|
|
vectorstore = Chroma(persist_directory=vector_store_path, embedding_function=embeddings)
|
||
|
|
vectorstore.add_documents(documents=splits, ids=uuids)
|
||
|
|
while True:
|
||
|
|
time.sleep(0.3)
|
||
|
|
ress = vectorstore.similarity_search(words[0])
|
||
|
|
if (len(ress) > 0):
|
||
|
|
break
|
||
|
|
return words,uuids,vectorstore
|
||
|
|
|
||
|
|
|
||
|
|
# @app.route('/checkRepeatText/<filename>', methods=['GET'])
|
||
|
|
def checkRepeatText(filename):
|
||
|
|
yield "文档相似性检查---启动中...."
|
||
|
|
vector_store_path="vector_store"+str(uuid.uuid4())
|
||
|
|
for titleName in findTitleName(filename):
|
||
|
|
yield titleName
|
||
|
|
if(titleName!="文档相似性检查----未找到与详细设计方案相关内容，无法进行相似性比较"):
|
||
|
|
try:
|
||
|
|
yield "文档相似性检查----文档内容解析中"
|
||
|
|
words,uuids,vectorstore=getDocxToText(filename,titleName,vector_store_path)
|
||
|
|
except Exception as e:
|
||
|
|
yield f"文档相似性检查----文档内容获取失败，未找到**{titleName}**相关内容或文档打开失败"
|
||
|
|
return
|
||
|
|
# 记录程序开始的时间戳‘
|
||
|
|
global device_id
|
||
|
|
similarity = Taskflow("text_similarity",device_id=3)
|
||
|
|
# device_id+=1
|
||
|
|
# if(device_id>1):
|
||
|
|
# device_id=0
|
||
|
|
reslist = []
|
||
|
|
count = 0
|
||
|
|
for i in words:
|
||
|
|
count += 1
|
||
|
|
yield f"文档相似性检查--对{titleName}章节，进行文档内容检查中{count}/{len(words)}"
|
||
|
|
result = vectorstore.similarity_search(i)
|
||
|
|
textTag = i.split("：")[0]
|
||
|
|
for content in result:
|
||
|
|
text = content.page_content
|
||
|
|
tag = text.split("：")[0].replace('\n', '')
|
||
|
|
if (textTag.find(tag) >= 0):
|
||
|
|
continue
|
||
|
|
try:
|
||
|
|
res = similarity([[i[i.find('：') + 1:], text[text.find('：') + 1:]]])
|
||
|
|
except Exception as e:
|
||
|
|
logger.warning("文档相似性检查--发生异常:",e)
|
||
|
|
logger.warning(i)
|
||
|
|
logger.warning(text)
|
||
|
|
if (res[0]["similarity"] > 0.90):
|
||
|
|
# 判断重复内容是否被放入
|
||
|
|
if (len(reslist) > 0):
|
||
|
|
isExist = False
|
||
|
|
for neirong in reslist:
|
||
|
|
if i in neirong.values():
|
||
|
|
isExist = True
|
||
|
|
break
|
||
|
|
if not isExist:
|
||
|
|
# reslist.append({"yuanwen1":i[i.find('：') + 1:],"yuanwen2":text[text.find('：') + 1:],"similarity":res[0]["similarity"]})
|
||
|
|
reslist.append({"yuanwen1":i.replace("\n",""),"yuanwen2":text.replace("\n",""),"similarity":res[0]["similarity"]})
|
||
|
|
else:
|
||
|
|
reslist.append({"yuanwen1":i.replace("\n",""),"yuanwen2":text.replace("\n",""),"similarity":res[0]["similarity"]})
|
||
|
|
# print(i.split("：")[1] + "\n" + text.split("：")[1])
|
||
|
|
# vectorstore.delete(ids=uuids)
|
||
|
|
shutil.rmtree(vector_store_path)
|
||
|
|
logger.info("已删除")
|
||
|
|
logger.info(reslist)
|
||
|
|
resInfo=f"对{titleName}章节，发现相似内容：<br>"
|
||
|
|
if(len(reslist)>0):
|
||
|
|
for res in reslist:
|
||
|
|
resInfo+="【在**"+res["yuanwen1"][:res["yuanwen1"].find('：')]+"**下包含："+res["yuanwen1"][res["yuanwen1"].find('：') + 1:]+"<br>在**"+res["yuanwen2"][:res["yuanwen2"].find('：')]+"**下包含："+res["yuanwen2"][res["yuanwen2"].find('：') + 1:]+"<br>以上两段内容***相似度***："+'{:.2f}'.format(res['similarity'])+"】<br>"
|
||
|
|
yield resInfo
|
||
|
|
logger.info(resInfo)
|
||
|
|
else:
|
||
|
|
yield "未发现相似内容"
|