My name is Willian Magalhães and I work with Chatbots. I never thought I would put together my two bachelor’s into a new profession. Nowadays, I work developing Chatbots (Watson — IBM) and I am a Linguist as well.
Recently, something caught my attention. I was curating a Chatbot of ours and I came across a very interesting situation. In order to give you a quick explanation, when a Chatbot gets huge and you have more than 500 intents, there is a need to analyze a certain level of confidence to enter the right intent. Alright, with the confidence set to 0.5 (value previously studied), we were all set, the flow was being followed properly but… I could not enter an intent, a simple one, from a simple question.
Just to exemplify, here’s an example: Where is the bank?
I had the intent set upto understand this simple question, so why on earth wasn’t my intent being triggered? So, I decided to go to my intent’s configuration to understand what was happening.
Indirect question, my friends. I had configured my intent with such examples:
Do you know where the bank is?
Would you mind telling me where the bank is?
Can you tell me where the bank is?
I wonder if you know where the bank is?
Could you tell me where the bank is?
In conclusion, I realized these indirect questions prompts were killing my intent, giving it a confused confidence. My intent did not know where to focus, if it was to focus on everything else but “where is the bank”.
So, here is the tip: less is more when designing a good intent.