同会话下聊天机器人无法检索历史数据的技术求助

自研基于OpenAI的医疗聊天机器人，同会话内无法正确检索历史数据。比如告知机器人姓名后，后续询问姓名时机器人无法识别。以下是相关代码和测试记录：

后端Flask代码 #

@app.route("/chat", methods=["POST"])
def chat():
    try:
        user_input = request.json.get("message", "")
        session_id = request.json.get("session_id", str(uuid.uuid4()))  # Generate a new session ID if not provided
        user_name = request.json.get("name", "User")  # Extract user name if provided

        # Store or update the user's name in the conversation model
        conversation = Conversation.query.filter_by(session_id=session_id).first()
        if conversation:
            conversation.user_name = user_name
            conversation.user_message = user_input
        else:
          conversation = Conversation(session_id=session_id, user_name=user_name, user_message=user_input, bot_reply="")
            db.session.add(conversation)
        db.session.commit()

        # Retrieve conversation history
        conversations = Conversation.query.filter_by(session_id=session_id).all()
        conversation_history = [c.user_message for c in conversations]

        # Retrieve user name from session
        user_name = session.get("name", "User")

        # Generate response using OpenAI's API with a system message
        app.logger.info("Generating response with ChatGPT...")
        try:
            response = openai.ChatCompletion.create(
                model="gpt-3.5-turbo",
                messages=[
                    {"role": "system", "content": f"You are a medical assistant. Your name is {user_name}. Provide accurate medical information and advice. If you are unsure about a medical issue, advise the user to consult a healthcare professional."},
                    {"role": "system", "content": "Conversation history: " + ", ".join(conversation_history)},
                    {"role": "user", "content": user_input}
                ],
                max_tokens=300,
                temperature=0.7
            )
            bot_reply = response['choices'][0]['message']['content'].strip()
            app.logger.info(f"ChatGPT reply: {bot_reply}")
        except Exception as e:
            app.logger.error(f"Error generating response with ChatGPT: {e}")
            bot_reply = "Error: Unable to generate response."

        # Update and store the conversation in the database
        app.logger.info("Updating conversation in database...")
        conversation.bot_reply = bot_reply
        db.session.commit()
        app.logger.info("Conversation updated successfully.")

        return jsonify({"response": bot_reply, "session_id": session_id, "name": user_name})

    except Exception as e:
        app.logger.error(f"Error processing request: {e}")
        return jsonify({"response": "Sorry, there was an error processing your request."})

测试对话 #

我：你好，我叫Osama
机器人：你好，Osama！今天我能为你提供哪些医疗相关的帮助？
我：我叫什么名字？
机器人：抱歉，我无法访问你的个人信息比如姓名。今天我能为你提供什么帮助？
我：你好，我叫osamah
机器人：你好，Osamah！今天我能为你提供什么帮助？
我：我叫什么名字？
机器人：抱歉，我无法访问该信息。今天我能为你提供什么帮助？

前端React app.js代码 #

function App() {
  const [chatHistory, setChatHistory] = useState([]);
  const [message, setMessage] = useState('');
  const [loading, setLoading] = useState(false);
  const [sessionId, setSessionId] = useState('');
  const [userName, setUserName] = useState('');

  // Initialize a new session ID when the component mounts
  useEffect(() => {
    const initSession = async () => {
      try {
        const response = await axios.post('http://localhost:5000/chat', { message: '' });
        setSessionId(response.data.session_id);
        console.log('Session initialized:', response.data);
      } catch (error) {
        console.error('Error initializing session:', error);
      }
    };
    initSession();
  }, []);

  const handleSendMessage = async () => {
    if (message.trim() === '' || !sessionId) return;

    setChatHistory([...chatHistory, { role: 'user', content: message }]);

    try {
      setLoading(true);
      const response = await axios.post('http://localhost:5000/chat', {
        message: message,
        session_id: sessionId,
        name: userName
      });
      console.log('Response from server:', response.data);
      setChatHistory(prev => [
        ...prev,
        { role: 'assistant', content: response.data.response }
      ]);
    } catch (error) {
      console.error('Error fetching data:', error);
      setChatHistory(prev => [
        ...prev,
        { role: 'assistant', content: 'Error: Unable to get a response.' }
      ]);
    } finally {
      setLoading(false);
      setMessage('');
    }
  };

  const handleUserNameChange = (event) => {
    setUserName(event.target.value);
  };

核心问题分析 #

1. 姓名变量被错误覆盖：后端存储用户姓名到数据库后，又用session.get("name", "User")覆盖了user_name变量。这里的session是Flask服务器端会话，和前端传入的会话ID不是同一概念，导致传给OpenAI的始终是默认值"User"。
2. 历史记录格式错误：用逗号拼接消息的方式无法让OpenAI模型正确理解对话上下文，且只传递了用户消息，缺少机器人的回复记录。
3. 会话记录被覆盖：更新会话时直接替换了user_message字段，导致数据库中仅保留最后一条用户消息，无法存储完整对话历史。

修复步骤 #

1. 修正姓名传递逻辑 #

删除user_name = session.get("name", "User")这一行，直接使用从前端传入并存储到数据库的user_name值。

2. 修复会话历史存储与传递 #

• 修改数据库操作，每次新增一条会话记录而非覆盖：

# 移除原有的单条记录更新逻辑，改为每次新增
conversation = Conversation(session_id=session_id, user_name=user_name, user_message=user_input, bot_reply="")
db.session.add(conversation)
db.session.commit()

• 调整历史记录格式，按照OpenAI要求的角色结构传递完整对话：

# 构建符合OpenAI要求的对话历史
conversations = Conversation.query.filter_by(session_id=session_id).all()
conversation_history = []
for c in conversations:
    conversation_history.append({"role": "user", "content": c.user_message})
    if c.bot_reply:
        conversation_history.append({"role": "assistant", "content": c.bot_reply})

• 更新OpenAI请求的消息结构：

response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[
        {"role": "system", "content": f"You are a medical assistant. Remember the user's name is {user_name}. Provide accurate medical information and advice. If unsure, advise consulting a healthcare professional."},
        *conversation_history,  # 展开完整对话历史
        {"role": "user", "content": user_input}
    ],
    max_tokens=300,
    temperature=0.7
)

3. 前端补充姓名输入绑定 #

在JSX中添加姓名输入框，确保用户输入的姓名能正确绑定到userName状态：

<input
  type="text"
  value={userName}
  onChange={handleUserNameChange}
  placeholder="请输入你的姓名"
/>

验证效果 #

修复后，后端会存储完整的对话历史（包括用户消息和机器人回复），并正确传递给OpenAI模型。用户告知姓名后，后续询问姓名时，模型能从历史上下文提取并回复正确姓名。

内容的提问来源于stack exchange，提问作者osamah alini