【浅学】星火知识库文档检索生成问答Demo实测

前置准备

用讯飞大模型3.5搭建好应用，具体操作可以看我的这篇：讯飞星火大模型API，实名认证免费领一年有效期的200万Token，在控制台的左侧有星火知识库，实名认证过就可以开通免费的部分。用这个纯粹是因为免费，关于这个大模型的使用体验啥的不做评价，大家可以也选择自己喜欢的其他模型，如使用其他模型则下文代码中的API接口调用部分需要自行根据所选模型的文档说明进行调整

文档准备

爬虫demo爬取的百度百科——藜麦数据，这个的demo有bug，百度百科网页的class_隔一段时间会变，如果执行报错了就打开百度百科——藜麦数据的网页源码找到这一段内容的类名替换即可，不会操作的话可以留言我出教程。(因为我还没学过这块所以只会替换类名这样的傻瓜式操作，如果有更好的方法感谢各位大佬的指点)

#文档准备
import requests
from bs4 import BeautifulSoup

url = "https://baike.baidu.com/item/%E8%97%9C%E9%BA%A6/5843874"
response = requests.get(url)
soup = BeautifulSoup(response.text, "html.parser")

# 获取词条名称
title = soup.find("div", class_="lemmaTitleBox_dR4Nr").find("h1").text

summary_list = soup.find('div', class_='J-lemma-content').find_all('div',class_='para_VW7X7 content_V8j6o MARK_MODULE')

print('词条：'+ title)
print('简介：')

with open("./藜.txt","w") as file:    ##打开读写文件，逐行将列表读入文件内
    for summary in summary_list:
        file.write(summary.text+"\n\n")
        print(summary.text)

文档上传

# -*- coding:utf-8 -*-
import hashlib
import base64
import hmac
import time
import random
from urllib.parse import urlencode
import json
import requests
from requests_toolbelt.multipart.encoder import MultipartEncoder

class Document_Upload:
    def __init__(self, APPId, APISecret, timestamp):
        self.APPId = APPId
        self.APISecret = APISecret
        self.Timestamp = timestamp

    def get_origin_signature(self):
        m2 = hashlib.md5()
        data = bytes(self.APPId + self.Timestamp, encoding="utf-8")
        m2.update(data)
        checkSum = m2.hexdigest()
        return checkSum


    def get_signature(self):
        # 获取原始签名
        signature_origin = self.get_origin_signature()
        # 使用加密键加密文本
        signature = hmac.new(self.APISecret.encode('utf-8'), signature_origin.encode('utf-8'),
                                 digestmod=hashlib.sha1).digest()
        # base64密文编码
        signature = base64.b64encode(signature).decode(encoding='utf-8')
        return signature

    def get_header(self):
        signature = self.get_signature()
        header = {
            "appId": self.APPId,
            "timestamp": self.Timestamp,
            "signature": signature,
        }
        return header

    # 提交网络文件
    def get_body(self):
        body = {
            "file": "",
            "url": "文件网络地址 例如： https://xxx.xx.com/xxx.pdf",
            "fileName": "文件名, 例如：xxx.pdf",
            "fileType": "wiki",
            "callbackUrl": "your_callbackUrl"
        }
        form = MultipartEncoder(
            fields=body,
            boundary='------------------' + str(random.randint(1e28, 1e29 - 1))
        )
        return form

    # 提交本地文件，注意File是字符串比如这里的File为"./藜.txt"
    def get_files_and_body(self,File):
        body = {
            "url": "",
            "fileName": File,
            "fileType": "wiki",
            "needSummary": False,
            "stepByStep": False,
            "callbackUrl": "your_callbackUrl",
        }
        files = {'file': open(File, 'rb')}
        return files, body



def getFileId(Id,Secret,FileId):
    # 先去 开放平台控制台（https://console.xfyun.cn）创建应用，获取下列应用信息进行替换
    APPId = Id
    APISecret = Secret

    curTime = str(int(time.time()))
    request_url = "https://chatdoc.xfyun.cn/openapi/fileUpload"

    document_upload = Document_Upload(APPId, APISecret, curTime)
    headers = document_upload.get_header()

    # ******************提交网络文件
    # body = document_upload.get_body()
    # headers['Content-Type'] = body.content_type
    # response = requests.post(request_url, data=body, headers=headers)

    # ******************提交本地文件
    files, body = document_upload.get_files_and_body(File)
    response = requests.post(request_url, files=files, data=body, headers=headers)
    
	file_id = response.json()['data']['fileId']
	return file_id

    # 文档上传成功

问答检索

# -*- coding:utf-8 -*-
import hashlib
import base64
import hmac
import time
from urllib.parse import urlencode
import json
import websocket
import _thread as thread
import ssl

class Document_Q_And_A:
    def __init__(self, APPId, APISecret, TimeStamp, OriginUrl):
        self.appId = APPId
        self.apiSecret = APISecret
        self.timeStamp = TimeStamp
        self.originUrl = OriginUrl

    def get_origin_signature(self):
        m2 = hashlib.md5()
        data = bytes(self.appId + self.timeStamp, encoding="utf-8")
        m2.update(data)
        checkSum = m2.hexdigest()
        return checkSum



    def get_signature(self):
        # 获取原始签名
        signature_origin = self.get_origin_signature()
        # print(signature_origin)
        # 使用加密键加密文本
        signature = hmac.new(self.apiSecret.encode('utf-8'), signature_origin.encode('utf-8'),
                             digestmod=hashlib.sha1).digest()
        # base64密文编码
        signature = base64.b64encode(signature).decode(encoding='utf-8')
        # print(signature)
        return signature



    def get_header(self):
        signature = self.get_signature()
        header = {
            "Content-Type": "application/json",
            "appId": self.appId,
            "timestamp": self.timeStamp,
            "signature": signature
        }
        return header

    def get_url(self):
        signature = self.get_signature()
        header = {
            "appId": self.appId,
            "timestamp": self.timeStamp,
            "signature": signature
        }
        return self.originUrl + "?" + f'appId={self.appId}&timestamp={self.timeStamp}&signature={signature}'
        # 使用urlencode会导致签名乱码
        # return self.originUrl + "?" + urlencode(header)



    def get_body(self,content,FileId):
        data = {
            "chatExtends": {
                "wikiPromptTpl": "请将以下内容作为已知信息：\n<wikicontent>\n请根据以上内容回答用户的问题。\n问题:<wikiquestion>\n回答:",
                "wikiFilterScore": 0.83,
                "temperature": 0.5
            },
            "fileIds": [
               	FileId
            ],
            "messages": [
                {
                    "role": "user",
                    "content": content
                }
            ]
        }
        return data

# 收到websocket错误的处理
def on_error(ws, error):
    print("### error:", error)


# 收到websocket关闭的处理
def on_close(ws, close_status_code, close_msg):
    print("### closed ###")
    print("关闭代码：", close_status_code)
    print("关闭原因：", close_msg)


# 收到websocket连接建立的处理
def on_open(ws):
    thread.start_new_thread(run, (ws,))


def run(ws, *args):
    data = json.dumps(ws.question)
    ws.send(data)

# 收到websocket消息的处理
def on_message(ws, message):
    # print(message)
    data = json.loads(message)
    code = data['code']
    if code != 0:
        print(f'请求错误: {code}, {data}')
        ws.close()
    else:
        content = data["content"]
        status = data["status"]
        # print(f'status = {status}')
        print(content, end='')
        if status == 2:
            ws.close()


def GPT(Id, Secret,FileId,content):
    # 先去 开放平台控制台（https://console.xfyun.cn）创建应用，获取下列应用信息进行替换
    APPId = Id
    APISecret = Secret 

    curTime = str(int(time.time()))
    OriginUrl = "wss://chatdoc.xfyun.cn/openapi/chat"
    document_Q_And_A = Document_Q_And_A(APPId, APISecret, curTime, OriginUrl)

    wsUrl = document_Q_And_A.get_url()
    print(wsUrl)
    headers = document_Q_And_A.get_header()
    body = document_Q_And_A.get_body(content,FileId)

    # 禁用WebSocket库的跟踪功能，使其不再输出详细的调试信息。
    websocket.enableTrace(False)
    ws = websocket.WebSocketApp(wsUrl, on_message=on_message, on_error=on_error, on_close=on_close, on_open=on_open)
    ws.appid = APPId
    ws.question = body
    ws.run_forever(sslopt={"cert_reqs": ssl.CERT_NONE})


    # 文档问答成功

Id和Secret在讯飞开放平台的控制台星火知识库页面找到

if __name__ == '__main__':
    Id = '*******'
    Secret = '*****'
    File = './藜.txt'
    FileId = getFileId(Id,Secret,File)
    content = '藜怎么防治虫害？'
    GPT(Id, Secret,FileId,content)

输出结果

### error: 'content'
### error: 'content'
藜麦的常见虫害有象甲虫、金针虫、蝼蛄、黄条跳甲、横纹菜蝽、萹蓄齿胫叶甲、潜叶蝇、蚜虫、夜蛾等。对于这些虫害的防治，可以采取以下方法：

1. 每亩用3%的辛硫磷颗粒剂2-2.5千克于耕地前均匀撒施，随耕地翻入土中。

2. 每亩用40%的辛硫磷乳油250毫升，加水1-2千克，拌细土20-25千克配成毒土，撒施地面翻入土中，这样可以有效防治地下害虫。

此外，在藜麦8叶龄时，需要将行中的杂草、病株及残株拔掉，提高整齐度和增加通风透光，同时进行根部培土，以防止后期倒伏。### closed ###
关闭代码： None
关闭原因： None