What is easy NLP?
Natural language processing (NLP) provides immense power to chatbots, allowing bot builders to move from structured decision-tree based conversations to one that facilitates organic conversations about hundreds, if not thousands, of topics. (If you’re unfamiliar with NLP, we have a blog post to explain it here.)
One of the challenges of NLP is that it requires training data. Training data for chatbots are requests/responses to the chatbot about a certain topic. For example, “I would like oatmeal, please.” and “May I have oatmeal?” are two different ways to phrase the same “intent”. We suggest 150-200 pieces of training data per intent in order to give your users an accurate experience but most companies won’t have this quantity of training data on-hand when they’re first starting with NLP.
We are changing this model by providing all the training data, and eliminating the need for a third party integration. All you need to do is provide the custom responses. Now you can start using NLP the first day you launch your chatbot!
How does it work?
As you are building out your bot you can add a “Natural Language Processing” node that is pre-trained.
1.Invite the user to talk: The first step is to create a node where you invite your website visitors to ask a question or say something to your bot. For example, you can add a node that says, “Ask me anything!”
Click, “+Node”, select “Free Text Reply” under Add a Node menu. Once you add your question, make sure to check the box, “Save the user’s response to their profile” and name it something like, Question. Once you’re done, click “Add node.”
2. Add the NLP bot node: Next, we’ll add a bot node that will “read” the user’s input and determine what topic or Intent the user was asking about. Go to your bot and click “+Node”, select “Advanced” menu and choose, “Natural Language Processing” node.
3. Reference question + choose intents: Next, we’ll add a node that will reference the response/question your user provided to the bot, determine which intent is being referenced and allow you to create a response for each common intent.
First we need to reference the bot node with the question. In this particular example, we asked the question to the website visitor in “Node 121.Free Text”. So we’ll reference the node by typing the number of the node, or choosing it from the selection list from the carrot at the right.
We’ve pre-trained the bot on a couple topics, called, “Intents”. We’ll be adding to the library of intents, but in this first version, you can choose from topics, such as Pricing and About Us.
For example, if someone asks, “What is the price?” Instabot will immediately know that they are talking about pricing. If you want your Instabot to understand that intent, then just check that box to include it in your bot. If the bot doesn’t know which topic the user is asking about, it will use what is called a “Fallback” intent. This is a mandatory field for using this node and you will not be able to uncheck it because the bot needs to know what to do if it doesn’t understand the question.
4. Write responses for each intent: Now that the bot “understands” what the user says, we still need to make responses that are right for your business. For example, the bot understands it’s being asked about pricing, but you need to tell it how to respond about your own company’s pricing model. For example, we can write something to the effect as, “Pricing ranges but our base plan starts at $3,600.”
You’ll need to write a Fallback Intent. This will be the response for when the bot doesn’t understand. We suggest using something generalized that says that your team will follow-up on the question. We DO NOT recommend that you use anything that says, “I don’t understand.” It’s a poor practice in conversational design and leads to high abandonment rates. It’s much better to let users know that they should expect a follow up from a human!
Try something like, “This would be a better question for my colleague. Let me get some information and I can have them follow-up.” Then ask for their email address so your team can respond when you get a chance. You can even add a custom bot goal to track the times when the bot didn’t understand and alert your team to follow-up and route the conversation to the right team.
Still have questions? Just ask! Email us at [email protected]