钉钉 AI 客服：对话状态管理

钉钉 AI 客服：对话状态管理

状态管理是多轮对话的基础。

---

一、状态类型

状态	说明
IDLE	空闲
COLLECTING	收集信息
CONFIRMING	确认中
PROCESSING	处理中
COMPLETED	完成

---

二、状态机设计

const STATES = {
  IDLE: {
    onMessage: (msg) => {
      if (isOrderQuery(msg)) return 'COLLECTING_ORDER';
      if (isRefund(msg)) return 'COLLECTING_REFUND';
      return 'IDLE';
    }
  },
  
  COLLECTING_ORDER: {
    onMessage: (msg, context) => {
      if (hasOrderId(msg)) {
        context.orderId = extractOrderId(msg);
        return 'CONFIRMING';
      }
      return 'COLLECTING_ORDER';
    }
  },
  
  CONFIRMING: {
    onMessage: (msg, context) => {
      if (isConfirm(msg)) return 'PROCESSING';
      if (isCancel(msg)) return 'IDLE';
      return 'CONFIRMING';
    }
  },
  
  PROCESSING: {
    onMessage: async (msg, context) => {
      await processOrder(context);
      return 'COMPLETED';
    }
  }
};

---

三、状态存储

// Redis 存储
async function saveState(sessionId, state, context) {
  await redis.hset('session', sessionId, JSON.stringify({ state, context }));
}

async function getState(sessionId) {
  const data = await redis.hget('session', sessionId);
  return data ? JSON.parse(data) : { state: 'IDLE', context: {} };
}

---

四、状态转换

async function handleMessage(sessionId, message) {
  // 获取当前状态
  const { state, context } = await getState(sessionId);
  
  // 执行状态转换
  const newState = await STATES[state].onMessage(message, context);
  
  // 保存新状态
  await saveState(sessionId, newState, context);
  
  // 返回响应
  return generateResponse(newState, context);
}

---

五、超时处理

// 设置超时
async function setTimeout(sessionId, seconds = 300) {
  await redis.expire(`session:${sessionId}`, seconds);
}

// 超时回调
redis.on('expired', async (key) => {
  const sessionId = key.split(':')[1];
  await handleTimeout(sessionId);
});

---

六、场景示例

订单查询流程

IDLE → 用户问订单 → COLLECTING_ORDER
COLLECTING_ORDER → 用户提供订单号 → CONFIRMING
CONFIRMING → 用户确认 → PROCESSING
PROCESSING → 查询完成 → COMPLETED
COMPLETED → → IDLE

---

七、最佳实践

• 状态清晰定义
• 转换逻辑简单
• 超时自动重置
• 日志记录转换

---

项目地址：GitHub - dingtalk-connector-pro 有问题欢迎 Issue 或评论区交流