python_ai/checkRepeatText.py

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
import json
embeddings = DashScopeEmbeddings(dashscope_api_key="sk-ea89cf04431645b185990b8af8c9bb13")
# embeddings = HuggingFaceEmbeddings(model_name="shibing624/text2vec-base-chinese",model_kwargs={"device":"npu:5"})
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
import requests
# from myLogger import outLog

# outLog.logger = logging.getLogger("checkRepeatText")
userLog=None
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):
    loopCount = 0
    while True:
        loopCount+=1
        if(loopCount>=60):
            raise Exception("文档读取超时，或文档存在问题无法读取")
            break
        try:
            document = Document(docxPath)
            break
        except Exception as e:
            time.sleep(1)
            pass
    # 逐段读取docx文档的内容
    titleWords=[]
    firstTitle = 0
    firstTitleName=""
    secondTitle = 0
    sanjiTitle = 0
    levelText=""
    count = 0
    numid =0
    wordContent={}
    total = len(document.paragraphs)
    addStart = False#是否重新添加
    yield "文档相似性检查----文档内容解析中",str(count),str(total)
    for paragraph in document.paragraphs:
        count+=1
        yield "文档相似性检查----文档内容解析中",str(count),str(total)
        # 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)
                addStart=True
                firstTitleName=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)
            ##先判断是不是一级标题
            if addStart:
                wordContent[firstTitleName]=[]
                addStart=False
            if level:
                levelText=f"{int(level)+1}级标题-"+text
            else:
                if(text.startswith("图") or text.startswith("注：")):
                    continue
                if (len(text)>30 and firstTitleName):
                    numid+=1
                    wordContent[firstTitleName].append("{}：".format(levelText)+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',
    }
    yield '文档相似性检查----检查是否存在详细设计方案'
    findTitleName_bot = Assistant(llm=findTitleName_llm_cfg,
                                    name='Assistant',
                                    system_message='按照要求选择最合适的，是唯一的'
                                )
    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('`', ''))
    try:
        if(parsed_data["answer"]=="存在"):
            yield parsed_data["name"],wordContent
        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"
    except Exception as e:
        userLog.warning(e)
        userLog.warning(data)
        userLog.warning(parsed_data)
        yield "文档相似性检查----检查遇到问题，请联系管理员"
#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):
def getDocxToText(titleName,wordContent,vector_store_path):

    # loopCount = 0
    # while True:
    #     loopCount+=1
    #     if(loopCount>=15):
    #         raise Exception("文档读取超时，或文档存在问题无法读取")
    #         break
    #     try:
    #         document = Document(docxPath)
    #         break
    #     except Exception as e:
    #         time.sleep(1)
    #         pass
    # # 逐段读取docx文档的内容
    # levelList=[]
    words=[]
    # addStart = False
    # levelText=""
    # i = 0
    # count = 0
    # total = len(document.paragraphs)
    # yield "文档相似性检查----文档内容解析中",count,total
    # for paragraph in document.paragraphs:
    #     count+=1
    #     yield "文档相似性检查----文档内容解析中",count,total
    #     # 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
    # 遍历字典，查找包含 "标题的" 的键
    for key, value in wordContent.items():
        if (titleName.find(key)>=0 or key.find(titleName)>=0):
            words.extend(value)  # 将对应的值添加
    if len(words)==0:
        raise Exception("checkRepeatText，获取长度为0")
    text = '\n'.join(words)
    userLog.info(f"文档相似性检查----需要处理的总数是{len(words)}")
    # 将文本写入txt文件
    with open("checkRepeatText.txt", 'w', ) as txt_file:
        txt_file.write(text)
    time.sleep(1)
    yield "文档相似性检查----文档内容转换中",".","."
    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 = []
    yield "文档相似性检查----文档保存中",".","."
    global embeddings
    vectorstore = Chroma(persist_directory=vector_store_path, embedding_function=embeddings)
    for i in range(len(splits)):
        uuidStr=str(uuid.uuid4())
        uuids.append(uuidStr)
    logging.info(f"checkRepeatTextuuidLen{len(uuids)}")

    vectorstore.add_documents(documents=splits, ids=uuids)
    yield "文档相似性检查----校验文档是否已经完成保存",".","."
    while True:
        time.sleep(0.3)
        ress = vectorstore.similarity_search(words[0])
        if (len(ress) > 0):
            break
    yield words,uuids,vectorstore


# @app.route('/checkRepeatText/<filename>', methods=['GET'])
def checkRepeatText(filename,user_id,outLog):
    global userLog
    userLog=outLog.get_queue(user_id,"checkRepeatText")
    yield "文档相似性检查---启动中...."
    userLog.info("文档相似性检查---任务开始")
    vector_store_path="vector_store"+str(uuid.uuid4())
    for titleName in findTitleName(filename):
        if(isinstance(titleName ,tuple)):
            if(len(titleName)==3):
                yield titleName[0]+titleName[1]+"/"+titleName[2]
        else:
            yield titleName
    if(isinstance(titleName ,tuple)):
        # try:
        yield "文档相似性检查----文档内容转换中"
        try:
            for words,uuids,vectorstore in getDocxToText(titleName[0],titleName[1],vector_store_path):
                if isinstance(words, str):
                    yield words+uuids+vectorstore
        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}**相关内容或文件无法正常打开。可以尝试用WORD或WPS打开文件，进行修复并另存，用另存的文件再做一次尝试。"
            userLog.warning(e)
            userLog.warning(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")
            outLog.mark_done(user_id, "checkRepeatText")
            return
        # 记录程序开始的时间戳‘
        reslist = []
        count = 0
        for i in words:
            count += 1
            yield f"文档相似性检查--对{titleName[0]}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:
                    url = "http://0.0.0.0:8192/taskflow/checkRepeatText"
                    headers = {"Content-Type": "application/json"}
                    data = {
                        "data": {
                            "text": [[i[i.find('：') + 1:], text[text.find('：') + 1:]]],
                        }
                    }
                    r = requests.post(url=url, headers=headers, data=json.dumps(data))
                    res = json.loads(r.text)
                    res=res["data"]
                    # res = similarity([[i[i.find('：') + 1:], text[text.find('：') + 1:]]])
                except Exception as e:
                    userLog.warning("文档相似性检查--发生异常:")
                    userLog.warning(e)
                    userLog.warning(i)
                    userLog.warning(text)
                    continue
                if (res[0]["similarity"] >= 0.96):
                    # 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"]})
                            userLog.info("【在"+i[:i.find('：')].replace("\n","")+"下包含："+i[i.find('：') + 1:].replace("\n","")+"<br>在"+text[:text.find('：')].replace("\n","")+"**下包含："+text[text.find('：') + 1:].replace("\n","")+"<br>以上两段内容相似度："+'{:.2f}'.format(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])
                        userLog.info("【在"+i[:i.find('：')].replace("\n","")+"下包含："+i[i.find('：') + 1:].replace("\n","")+"<br>在"+text[:text.find('：')].replace("\n","")+"**下包含："+text[text.find('：') + 1:].replace("\n","")+"<br>以上两段内容相似度："+'{:.2f}'.format(res[0]["similarity"])+"】")
        # vectorstore.delete(ids=uuids)
        shutil.rmtree(vector_store_path)
        resInfo=f"对{titleName[0]}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
        else:
            yield "**未发现相似内容**"
            userLog.info("文档相似性检查----未发现相似内容**")
        outLog.mark_done(user_id, "checkRepeatText")