haveal / messenger-bot-nlp

A Facebook Messenger bot sample integrated with built-in NLP from wit.ai

Messenger Bot with built-in NLP

A Facebook Messenger bot sample integrated with built-in NLP from wit.ai fully deployable to Heroku

Before we start

• Install Node.js
• A github account
• Create a free account and install Heroku Command Line
• Make sure you have a Facebook Developer App that is not currently using webhooks
• And a free wit.ai account, it will do the NLP heavy lifting for free!

Set-up

• Fork this repository. Then when deploying to Heroku the code will stay synced with whatever changes you make on your copy
• Create a Facebook page that will be the face of your bot
• Create a wit.ai app that will be its brain
• Take note of your wit.ai Server Access Token. You can get it from the settings page
• Now we are ready to configure your Facebook App. Open it at https://developers.facebook.com/apps/<APP_ID> and go to the Messenger tab
  • In the Token Generation section select your page and generate a token. This will allow our server to listen and reply to conversations
  • Find the Webhooks section and subscribe to all events. Then find and select your page and subscribe
  • In the Built-In NLP section, select your page, choose the Custom model, add you wit.ai Server Access Token previously retrieved and Save
• You are ready to go! Tap deploy button and you will need to enter the App ID and App Secret from you Facebook App that can be found in the Dasboard tab. You will also need a Verification Token, but that can be any string you want.

• One last thing, now that you have a node.js webserver in production we need to tell Facebook where to forward all incoming messages to your bot. Go to the Webhooks section on your Facebook app and click on Edit Subscription. You can get the Callbacck URL from Heroku in the Open App button, it should be https://<APP_ID>.herokuapp.com/webhook. The Verify Token is the one you entered in the previous step
• Talk to your page, it should reply
• Now it's your turn to give it some personality.

Adding NLP

A quick overview of wit.ai NLU engine: it transforms a text snippet into a list of entities, both built-in or custom defined. Entities come in 4 flavours. You can check the docs for more info.

• Keywords are entities from a finite set of posibilites such as countries or vegetables
• Free-text are specific parts of the sentence that we want to extract, like the message content to send, or the text snippet to translate
• Free-text & keywords is of course a mixture of the last two, where we cannot anticipate all the possible values
• Traits are labels that affect the whole sentence, not just a single part, like the sentiment or the intent of the phrase

In your wit.ai console, create a trait entity called intent. We will use that trait to store the main intentions the bot is expected to recognize, like greetings, help and your bot's particular expertise.

Try to come up with at least 10-20 samples for each intent, and as you type them assign the propper entities to extract. Most importantly, have other people test your bot and use the Inbox to learn from the examples.

Running locally

Note than when you deply to Heroku a PostgreSQL database will be created that you will use when running locally as well. Do this once:

git clone https://github.com/<USER_ID>/messenger-bot-nlp.git
cd messenger-bot-nlp
# Clone the values in your heroku apps to your local machine
heroku config -s --app <HEROKU_APP_NAME> > .env
npm install

Do this every time to start a new webserver

heroku local web

and to expose the server to the world via ngrok

node_modules/ngrok/bin/ngrok http 3000

This will print a URL of the form https://<SOME_CODE>.ngrok.io/webhook that you should use to reconfigure your webhook to point to your local machine. When you are done make sure to reverse this to point back to your heroku app.

Review process

Once you're happy with your bot you will want to show it to others, sadly you need Facebook's approval before letting your app out in the wild. The good thing is that you'll only need to this once.

First go the App Review and make sure to toggle your app state to Public.

In the Roles section on your app's dashboard you can add as many friends as you want as test users, who will be able to talk to your bot regardless of its approval status.

Finally, in the Messenger section you can create a new submission. You will only need the pages_messaging permission. Add the relevant details and click on Submit for Review. It should take only a few days to get approved.

Join the Messenger Platform Developer Community to get help with anything Messenger Bot related.

Good luck!