<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=979905748791482&amp;ev=PageView&amp;noscript=1">
AI ChatbotsFebruary 1, 2018Written by Alex Debecker

Sentiment Analysis: Enhance Your Chatbot Experience

Sentiment and chatbot in the same sentence? No, this isn't a science fiction article.

Sentiment analysis is a technique that has been taking the front seat in many chatbot builds, particularly in the customer service space. However, this technique remains somewhat unknown (and, to some, frightening).

In this article, I want to break the myths about sentiment analysis in chatbots, explain what it is, and how it can enhance your chatbot user experience.


What is sentiment analysis?

First things first, we need to understand what sentiment analysis is.

Sentiment analysis is a layer on top of a chatbot's natural language understanding (NLU) engine. It is a functionality that allows the chatbot to 'understand' the mood the user is by analysing verbal and sentence structuring clues.

An example will help. Consider this sentence:

'Hi, the website seems broken, I keep hitting an error when I try to checkout?'

And now consider this one:

'Checkout page is 404 again...'

Now, humans can picture the different moods these two people are in. The person saying the first sentence seems more neutral, not angry about the situation. She's trying to understand and get the help she needs.

The second person, however, seems annoyed and on the verge of giving up.

Two very different moods, which should be treated by a customer support rep in two very different ways.

This level of understanding is what sentiment analysis is trying to achieve, automatically and at scale. Beyond understanding that these two people need help with the checkout page, the chatbot can understand the mood they are in at the moment. This allows it to deliver the best user experience it can, based on this extra layer of information.


More information, better service

Great, so a chatbot can figure out if you are angry, sad, or happy based on your sentence. Now what?

Well, now we can build and optimise user experiences.

sentiment analysis chatbot definition

This extra layer of information is invaluable. For the chatbots we have deployed using this technology, this is how we've improved the experience.


Auto-transferring to humans

Sometimes, a bot just won't cut it. We need human help. Most of the time, this is when we are very angry about something. We don't want to stick around talking to a bot, we want the real deal.

With sentiment analysis, the chatbot can suss out the angry customers and auto-transfer the ticket to a human. It doesn't even get involved.

PS: Think this defeats the purpose of a chatbot? Think again. Since the chatbot can automatically answer up to 80% of customer enquiries 24/7, the human staff has lots of time to focus on these angry customers. This is part of the fantastic chatbot experience.


Data collecting, validation, and training

As our most avid readers know, we like to develop chatbots in phases. It's the best way to approach this new technology.

When it comes to deploying a sentiment analysis function to one of our bots, we take the same approach. The idea here is to teach the chatbot to understand human input, capture the data, send it to a human for validation, then train it.

In practice, this means the chatbot will intercept the user's question. Then, it will analyse what the user wants (intent) and how the user asks for it (sentiment).

The chatbot will then forward a message to a human on the team saying something like: 'This is Sally, she wants to know why she hasn't received her package yet. She seems angry and anxious.'

The human then validates or contradicts the chatbot's opinion of Sally's frame of mind. This feedback teaches the chatbot, and the loop continues.

Over a few weeks, the chatbot learns to properly assess sentiment and intent -- both at the same time.


Adapting on the fly

Finally, sentiment analysis allows the chatbot to adapt in real time.

Remember, users don't expect to be treated the same way if they are angry or happy. A chatbot should not deliver the same robotic, seemingly unaware experience to everyone it talks to.

After a decent amount of training, the UX team behind the chatbot can start drafting different personas with different language and tone. This will allow the chatbot to adapt in real time and deliver the best experience it can to the users, based on their current mood.


Beyond customer service chatbots

You are now thinking this is amazing technology. You're right, it is. Should it remain within customer service? Of course not!

Ultimately, every chatbot will have some sort of sentiment analysis functionality. It makes sense to provide an optimal experience to each individual user. The power of a sentiment analysis functionality is it offers personalisation and consideration, at scale.