使用网络钩子创建 fulfillment

Dialogflow 中的网络钩子 fulfilment 使我们能够更好地控制代理流。在本教程中，您需要一个网络钩子来验证在“序列”意图中收集的字母数字序列。该网络钩子会反复轮询该意图，以便在更易于管理的迭代中收集长序列。

使用内嵌编辑器创建网络钩子

Dialogflow 在控制台中有一个内嵌编辑器，使您可以直接写入 NodeJS 代码，然后将其部署作为网络钩子在 Cloud Functions 上运行。

如需使用 Dialogflow 的内嵌编辑器创建网络钩子，请按以下步骤操作：

  1. 点击导航栏上的 Fulfillment 标签页，转到“Fulfillment”页面。
  2. 将内嵌编辑器的按钮切换为“已启用”(ENABLED)。
  3. 删除内嵌编辑器的 package.json 标签页中的现有内容。

  4. 将以下 JSON 内容复制并粘贴到内嵌编辑器的 package.json 标签页中：

    {
  "name": "DialogflowFirebaseWebhook",
  "description": "Firebase Webhook dependencies for a Dialogflow agent.",
  "version": "0.0.1",
  "private": true,
  "license": "Apache Version 2.0",
  "author": "Google Inc.",
  "engines": {
    "node": "10"
  },
  "scripts": {
    "lint": "semistandard --fix \"**/*.js\"",
    "start": "firebase deploy --only functions",
    "deploy": "firebase deploy --only functions"
  },
  "dependencies": {
    "firebase-functions": "^2.0.2",
    "firebase-admin": "^5.13.1"
  }
}

  5. 删除内嵌编辑器的 index.js 标签页中的现有代码。

  6. 将以下代码复制并粘贴到内嵌编辑器的 index.js 页标签中：

    /**
 * Copyright 2020 Google Inc. All Rights Reserved.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

'use strict';

const functions = require('firebase-functions');

// TODO: set this to the minimum valid length for your sequence.
// There's no logic in here to enforce this length, but once the
// user has said this many digits, the slot-filling prompt will
// also instruct the user to say "that's all" to end the slot-filling.
const MIN_SEQUENCE_LENGTH = 10;

exports.dialogflowFirebaseFulfillment = functions.https.onRequest((request, response) => {
  let dfRequest = request.body;
  let action = dfRequest.queryResult.action;
  switch (action) {
    case 'handle-sequence':
      handleSequence(dfRequest, response);
      break;
    case 'validate-sequence':
      validateSequence(dfRequest, response);
      break;
    default:
      response.json({
        fulfillmentText: `Webhook for action "${action}" not implemented.`
      });
  }
});

////
// Helper functions

/* Send an SSML response.
 * @param request: Dialogflow WebhookRequest JSON with camelCase keys.
 *     See https://cloud.google.com/dialogflow/es/docs/reference/common-types#webhookrequest
 * @param response: Express JS response object
 * @param ssml: SSML string.
 * @example: sendSSML(request, response, 'hello')
 *     Will call response.json() with SSML payload '<speak>hello</speak>'
 */
function sendSSML(request, response, ssml) {
  ssml = `<speak>${ssml}</speak>`;

  if (request.originalDetectIntentRequest.source == 'GOOGLE_TELEPHONY') {
    // Dialogflow Phone Gateway Response
    // see https://cloud.google.com/dialogflow/es/docs/reference/rpc/google.cloud.dialogflow.v2beta1#google.cloud.dialogflow.v2beta1.Intent.Message.TelephonySynthesizeSpeech
    response.json({
      fulfillmentMessages: [{
        platform: 'TELEPHONY',
        telephonySynthesizeSpeech: {ssml: ssml}
      }]
    });
  }
  else {
    // Some CCAI telephony partners accept SSML in a plain text response.
    // Check your specific integration and customize the payload here.
    response.json({
      fulfillmentText: ssml
    });
  }
}

/* Extract an output context from the incoming WebhookRequest.
 * @param request: Dialogflow WebhookRequest JSON with camelCase keys.
 *     See https://cloud.google.com/dialogflow/es/docs/reference/common-types#webhookrequest
 * @param name: A string
 * @return: The context object if found, or undefined
 * @see: https://cloud.google.com/dialogflow/es/docs/reference/rpc/google.cloud.dialogflow.v2#google.cloud.dialogflow.v2.Context
 *     and note this webhook uses JSON camelCase instead of RPC snake_case.
 * @example:
 *     // Modify an existing output content
 *     let context = getOutputContext(request, 'some-context');
 *     context.lifespanCount = 5;
 *     context.parameters.some_parameter = 'new value';
 *     response.json({
 *       fulfillmentText: 'new value set',
 *       outputContexts: [context]
 *     });
 */
function getOutputContext(request, name) {
  return request.queryResult.outputContexts.find(
      context => context.name.endsWith(`/contexts/${name}`)
  );
}

////
// Action handler functions

/*
 * Fulfillment function for:
 *     actions: handle-sequence
 *     intents: "Sequence", "Sequence - Edit"
 * @param request: Dialogflow WebhookRequest JSON with camelCase keys.
 *     See https://cloud.google.com/dialogflow/es/docs/reference/common-types#webhookrequest
 * @param response: Express JS response object
 */
function handleSequence(request, response) {
  let parameters = request.queryResult.parameters;
  let isSlotFilling = !request.queryResult.allRequiredParamsPresent;
  let isEditing = getOutputContext(request, 'editing-sequence');
  console.log(request.queryResult.action + ': ' + JSON.stringify(parameters));

  if (isSlotFilling) {
    // Prompt the user for the sequence

    let verbatim = `<prosody rate="slow"><say-as interpret-as="verbatim">${parameters.existing_sequence}</say-as></prosody>`;

    if (!parameters.existing_sequence && !parameters.new_sequence) {
      // Initial prompt
      response.json({
        fulfillmentText: "What is your sequence? Please pause after a few characters so I can confirm as we go."
      });
    }
    else if (!isEditing) {
      // Confirm what the system heard with the user. We customize the response
      // according to how many sequences we've heard to make the prompts less
      // verbose.
      if (!parameters.previous_sequence) {
        // after the first input
        sendSSML(request, response,
            `Say "no" to correct me at any time. Otherwise, what comes after ${verbatim}`);
      }
      else if (parameters.existing_sequence.length < MIN_SEQUENCE_LENGTH) {
        // we know there are more characters to go
        sendSSML(request, response,
            `${verbatim} What's next?`);
      }
      else {
        // we might have all we need
        sendSSML(request, response,
            `${verbatim} What's next? Or say "that's all".`);
      }
    }
    else {
      // User just said "no"
      sendSSML(request, response,
          `Let's try again. What comes after ${verbatim}`);
    }
  }
  else {
    // Slot filling is complete.

    // Construct the full sequence.
    let sequence = (parameters.existing_sequence || '') + (parameters.new_sequence || '');

    // Trigger the follow up event to get back into slot filling for the
    // next sequence.
    response.json({
      followupEventInput: {
        name: 'continue-sequence',
        parameters: {
          existing_sequence: sequence,
          previous_sequence: parameters.existing_sequence || ''
        }
      }
    });

    // TODO: CHALLENGE: consider validating the sequence here.
    // The user has already confirmed existing_sequence, so if you find a unique
    // record in your database with this existing_sequence prefix, you could send
    // a followUpEventInput like 'validated-sequence' to skip to the next part
    // of the flow. You could either create a new intent for this event, or
    // reuse the "Sequence - done" intent. If you reuse the "done" intent, you
    // could add another parameter "assumed_sequence" with value
    // "#validated-sequence.sequence", then modify the validateSequence function
    // below to customize the response for this case.
  }
}

/*
 * Fulfillment function for:
 *     action: validate-sequence
 *     intents: "Sequence - Done"
 * @param request: Dialogflow WebhookRequest JSON with camelCase keys.
 *     See https://cloud.google.com/dialogflow/es/docs/reference/common-types#webhookrequest
 * @param response: Express JS response object
 */
function validateSequence(request, response) {
  let parameters = request.queryResult.parameters;
  // TODO: add logic to validate the sequence and customize your response
  let verbatim = `<say-as interpret-as="verbatim">${parameters.sequence}</say-as>`;
  sendSSML(request, response, `Thank you. Your sequence is ${verbatim}`);
}

  7. 点击部署 (DEPLOY)。

现在，您应该可以通过调用代理来测试集成了。如果您尚未执行此操作，建议您立即设置我们的合作伙伴提供的一键式电话集成，或设置 Dialogflow Phone Gateway 来通过电话测试代理。

了解代码

作为网络钩子的入口点，每次触发网络钩子时，都会调用此处的 dialogflowFirebaseFulfillment 函数。对于每个请求，Dialogflow 都会发送您在 Dialogflow 控制台中为意图指定的“操作”名称。代码会使用此操作名称来确定要调用的网络钩子函数（handleSequence 或 validateSequence）。

处理序列

handleSequence 是本教程的核心功能。它负责序列槽位填充的所有方面，包括：

  • 当会话首次输入意图时读出初始说明。
  • 先重复该序列，然后再提示下一组。
  • 告知最终用户如何更正聊天机器人。
  • 识别有效序列何时具有足够的数位，并告知最终用户如何最后确定输入（请参阅代码中的“MIN_SEQUENCE_LENGTH”）。
  • 循环槽位填充，以收集多个部分序列。
  • 将部分序列连接成一个长序列。

验证序列

使用 validateSequence 时，您需要添加与数据存储区的连接，以验证最终序列并根据该数据返回自定义消息。例如，如果您要构建订单查询代理，可以在此处自定义响应，例如：

Thank you. Your order ${verbatim} will arrive on ${lookup.date} and will ${lookup.require_signature ? '' : 'not'} require a signature.

其中，lookup 是您在数据存储区中找到的与此订单相关的某个对象。

辅助函数

此示例不使用任何特定于 Dialogflow 的依赖项。请改为遵循 WebhookRequest 参考文档了解 request.body 中的预期内容，并遵循 WebhookResponse 参考文档来了解要使用 response.json({...}) 响应的内容。

此代码包含两个辅助函数，更易于：

  • 通过将字符串传递给 sendSSML，针对当前平台发送正确的响应 JSON。
  • 通过将上下文名称传递给 getOutputContext，从请求中查找有效的 Dialogflow 上下文。

进一步改进

这可以帮助您开始使用网络钩子来实现高级使用场景。您设计了一个代理，该代理可在最终用户说出其序列时循环显示序列提示，从而确保虚拟代理能够正确地听到它们。

以下是一些进一步改进体验的方法：

  • 更改一些网络钩子响应，以与您的品牌相符。例如，您可以修改代码，指定“您的订单号是什么？”，而不是宽泛的“您的序列是什么？”您可以改为在...”上查找。
  • 考虑向“序列 - 完成”意图添加其他输出上下文，然后在该输入上下文下创建一些新意图，以便用户可以就其订单继续提问。
  • 如果您想深入了解此使用场景，请查看上面示例代码中的 TODO: CHALLENGE，了解如何进一步改进用户体验。
上一页
创建序列流程意图