Chatbot Training Made Simpler & Versatile With Dynamic Decision Trees
Welcome to the chatbot-centric world, which believe it or not, is estimated to reach a whopping $102.29 billion by 2026. Despite the strong capabilities to boost self-service success and customer experience, a lot of people still don’t believe in the magical power of chatbots. Pretty sure that they must’ve been stung by a chatbot’s out-of-context and wrong responses.
We don’t blame them, but we also don’t blame the chatbot! Poor fella can only do so much with its underlying structure. We are talking specifically about a chatbot’s decision tree structure, the network of IF/THEN statements (sorry developers) that decide the conversation flow of a chatbot. This is where chatbot admins or trainers spend most of their time, honing a chatbot’s capability to provide contextual responses.
Needless to say, training a chatbot is complex. Is there any way to simplify the process? Also, how can you reduce the dependency of chatbot experts so that when they leave your company, the entire bot knowledge doesn’t go out the window?
Three words: dynamic decision trees. This blog post illuminates what they are and how they simplify chatbot training for quicker ROI and time-to-value on your company portals. And as a bonus, you’ll be introduced to a completely revamped StoryBoard of SearchUnify’s Intelligent Chatbot that offers a great deal of versatility and is easy to use. Let’s get started.
First Things First—What are Dynamic Decision Trees in Chatbots?
Before we proceed to answer this question, let’s quickly glance through some important chatbot terminologies that will help you better understand the rest of the post.
1. Storyboarding: This is where chatbot trainers or admins create a layout of possible scenarios or stories and frame utterances, intents, and entities for a successful conversation flow.
2. Nodes: A story consists of nodes. Each node consists of an intent & its subsequent response.
This is an image of our chatbot’s previous storyboarding section, which consists of static decision trees, offering a linear flow of conversations. Greetings, Content Sources, Search Clients - are all stories, and the plus sign at the end is called a node.
And this is what revamped StoryBoard UI that’s powered by dynamic decision tree structures looks like. Now, the admin gets a clearer picture of the story flow, parent node, and child nodes. Also, they can create as many nodes as they want for a story.
“The Old (Bot) Must Always Make Way for The New”
While our old chatbot was working stupendously, there was room for improvement. In its static decision tree structure, every story had a single node. Also, the nodes (intent + responses) were listed one after the other, making it toilsome to find connections between them quickly. In simple words, it was difficult for an admin to look at the storyboard and identify the node that will trigger after the first one.
Previously each node had to be linked with a single intent. This led to the creation of multiple intents for a single story. In the new, upgraded bot, the entire chatbot story will be driven by single intent, thus making it simpler to manage.
How the Reworked Bot Ameliorates Chatbot Training
With the new cleaner, more visual StoryBoard, anyone can figure out immediately how nodes come together in a story. Admins can create as many nodes as they want for a story - all driven by a single intent. Consider this example.
Let’s say a travel company is using SearchUnify’s intelligent chatbot on its website. For the chatbot admin, one possible story for customers could be for booking. Now, that booking could be for a hotel, flight, or holiday home. With dynamic decision trees, the admin can create one story titled ‘Booking’ and create nodes for all three under it. The benefits, as mentioned earlier, are:
1. It’s easier to manage a single story by creating a single intent ‘booking,’ that is further branched out to nodes - ‘hotel,’ ‘flight,’ etc., as compared to creating multiple intents for a single story.
2. This also helps the chatbot add or cull steps on the go. For instance, if a customer chooses 'booking,' and then 'flight,' SearchUnify’s bot will ask for the destination, travel dates, and other details. But if the customer had straightaway typed that they want to book a hotel, the bot wouldn't have asked them to choose from a hotel, flight, and holiday home, and skipped straight to dates, room preferences, available suites, etc.
Takeaway
Dynamic decision trees shift the onus of quality chatbot interactions, which is usually on trainers, to artificial intelligence. Hence, when a dedicated admin leaves your company, the entire bot knowledge stays airtight. An easy-to-grasp UI helps the new moderators to understand the conversational flows and makes chatbot training easier and more engaging!
Unlike traditional standalone bots, [SearchUnify’s](https://www.searchunify.com/) [Intelligent Chatbot](https://chatbots.searchunify.com/#) is built on top of a unified [cognitive platform](https://applications.searchunify.com/) that combines artificial intelligence, NLU, and adaptive unsupervised learning.
This post has only scratched the surface. We’ve got a lot of exciting stuff coming up for chatbot enthusiasts in our next release, Colubridae ‘21, on June 17. Save the date! And get more insights into our new all-powerful chatbot along with several other new features and enhancements.
We don’t blame them, but we also don’t blame the chatbot! Poor fella can only do so much with its underlying structure. We are talking specifically about a chatbot’s decision tree structure, the network of IF/THEN statements (sorry developers) that decide the conversation flow of a chatbot. This is where chatbot admins or trainers spend most of their time, honing a chatbot’s capability to provide contextual responses.
Needless to say, training a chatbot is complex. Is there any way to simplify the process? Also, how can you reduce the dependency of chatbot experts so that when they leave your company, the entire bot knowledge doesn’t go out the window?
Three words: dynamic decision trees. This blog post illuminates what they are and how they simplify chatbot training for quicker ROI and time-to-value on your company portals. And as a bonus, you’ll be introduced to a completely revamped StoryBoard of SearchUnify’s Intelligent Chatbot that offers a great deal of versatility and is easy to use. Let’s get started.
First Things First—What are Dynamic Decision Trees in Chatbots?
Before we proceed to answer this question, let’s quickly glance through some important chatbot terminologies that will help you better understand the rest of the post.
1. Storyboarding: This is where chatbot trainers or admins create a layout of possible scenarios or stories and frame utterances, intents, and entities for a successful conversation flow.
2. Nodes: A story consists of nodes. Each node consists of an intent & its subsequent response.
This is an image of our chatbot’s previous storyboarding section, which consists of static decision trees, offering a linear flow of conversations. Greetings, Content Sources, Search Clients - are all stories, and the plus sign at the end is called a node.
And this is what revamped StoryBoard UI that’s powered by dynamic decision tree structures looks like. Now, the admin gets a clearer picture of the story flow, parent node, and child nodes. Also, they can create as many nodes as they want for a story.
“The Old (Bot) Must Always Make Way for The New”
While our old chatbot was working stupendously, there was room for improvement. In its static decision tree structure, every story had a single node. Also, the nodes (intent + responses) were listed one after the other, making it toilsome to find connections between them quickly. In simple words, it was difficult for an admin to look at the storyboard and identify the node that will trigger after the first one.
Previously each node had to be linked with a single intent. This led to the creation of multiple intents for a single story. In the new, upgraded bot, the entire chatbot story will be driven by single intent, thus making it simpler to manage.
How the Reworked Bot Ameliorates Chatbot Training
With the new cleaner, more visual StoryBoard, anyone can figure out immediately how nodes come together in a story. Admins can create as many nodes as they want for a story - all driven by a single intent. Consider this example.
Let’s say a travel company is using SearchUnify’s intelligent chatbot on its website. For the chatbot admin, one possible story for customers could be for booking. Now, that booking could be for a hotel, flight, or holiday home. With dynamic decision trees, the admin can create one story titled ‘Booking’ and create nodes for all three under it. The benefits, as mentioned earlier, are:
1. It’s easier to manage a single story by creating a single intent ‘booking,’ that is further branched out to nodes - ‘hotel,’ ‘flight,’ etc., as compared to creating multiple intents for a single story.
2. This also helps the chatbot add or cull steps on the go. For instance, if a customer chooses 'booking,' and then 'flight,' SearchUnify’s bot will ask for the destination, travel dates, and other details. But if the customer had straightaway typed that they want to book a hotel, the bot wouldn't have asked them to choose from a hotel, flight, and holiday home, and skipped straight to dates, room preferences, available suites, etc.
Takeaway
Dynamic decision trees shift the onus of quality chatbot interactions, which is usually on trainers, to artificial intelligence. Hence, when a dedicated admin leaves your company, the entire bot knowledge stays airtight. An easy-to-grasp UI helps the new moderators to understand the conversational flows and makes chatbot training easier and more engaging!
Unlike traditional standalone bots, [SearchUnify’s](https://www.searchunify.com/) [Intelligent Chatbot](https://chatbots.searchunify.com/#) is built on top of a unified [cognitive platform](https://applications.searchunify.com/) that combines artificial intelligence, NLU, and adaptive unsupervised learning.
This post has only scratched the surface. We’ve got a lot of exciting stuff coming up for chatbot enthusiasts in our next release, Colubridae ‘21, on June 17. Save the date! And get more insights into our new all-powerful chatbot along with several other new features and enhancements.
0
Please sign in to leave a comment.
Comments
0 comments