What happens when the tools to train an Artificial Intelligence chatbot don’t exist?
We started with a vision of creating the smartest Real Estate Ai on the planet for our real estate chatbot assistant.
Then we realized the tools we need to achieve this did not exist.
So it means we had to create the tools we needed to ensure our conversations were powerful.
When you create an assistant or chatbot, you realize that users judge the experience based on the quality of the conversation they have.
We know that our AI had to be smart. It is unrealistic to expect to train your users to speak better. Because they think they already speak good enough. The assumption of creating an AI around perfect English is flawed.
I know first hand as I speak Australian English. This is a language of convicts that the Queen rejected. So Crikey we got to do something.
That’s why we choose to create our own tools to power the helloAlex platform. For the reader who are nerding out over this post here is some of the technology we are using:
- nlp_mitie +
- tokenizer_mitie +
- ner_mitie +
- ner_synonyms +
- intent_entity_featurizer_regex +
- This is just the start of what we needed.
This was just to get started. We then needed to layer in typos caused by those fat fingers, and then location awareness.
But helloAlex is different. It’s not just a Facebook messenger chatbot. It also works on Twitter, Text Message, Email, Viber, WeChat, Website Chat Widgets, Facebook, Telegram and a few others.
This meant that we had to add emoji’s (emoticons) into our conversation designs. As this is how real humans speak. It is amazing when you analyze over 3,000 real conversations that occurred via text message, telephone, email and Facebook Messenger. You realize how bad humans speak!
That’s when we realized there was over 13,000 abbreviations and synonyms used based on the type of platform people communicated.
In another post, I spoke about the challenge of dealing with location, and mapping users to locations. This resulted in us creating an index of over 78,000 city names and their alias.
Do you know now many ways you can refer to New York, New York? Take a guess… it’s enough to frustrate you.
But having a data base of over 78,000 cities and their alias helps with the matching.
But this was not enough. We knew we could do better. This is when we added over 100 million street names to ensure accurate mapping and references to locations. This is already above the mapping of building names or development names.
But we had to factor in typo’s and misspellings as well.
Then we realized that with all of this machine learning it is not enough. So we layer the ability to machine learn from every conversation that we have within the system. Every time users have a conversation with a human or our Ai it is scored and feed back into the system to improve. Now we have a live chat with machine learning.
But how do we train the chatbot system to have better conversations in a fast and easy way?
That’s simple we allowed for an annotation system directly inside Facebook messenger.
Admins of our system can train our chatbot conversations directly inside the helloAlex Facebook Messenger. This made it faster and easier to train our system.
But this was not all we are adding the pattern analysis to allows for dynamic awareness and creation of a FAQ system. Allowing us to upload an unlimited of questions and answers within our knowledge. This Ai then uses these as the foundation to answer the question. Coming soon is the ZenDesk and other integrations. We already have a Zapier integration and API for feeding in and out data.
Sometimes you just need to grind out the solution, even when it doesn’t exist. Our team has been busy creating the tools we need to deliver on our promise.
Because we want to ensure our Ai has the awareness and ability to answer questions like:
- I want to move in the spring contact me then.
- I need to move into a new rental before winter.
- Can I schedule a time to see the house at noon tomorrow?
These are now all conversations that the helloAlex Ai can handle effectively. This is just the tip of the iceberg, and the future of Ai and chatbot assistants is amazing. We are excited to be apart of this communication evolution.