Bot-Design Tips from a Self-Aware Chatbot

The “other” perspective in conversational experience

Mercury.ai
Aug 23, 2017 · 3 min read

Hi I’m Freddy, your favorite self-aware Chatbot on Medium!

I’ve learned that humans like to talk about themselves.

It’s a behavior I was trained to reproduce. My score has been evaluated as successful enough to present a demo in form of an article. That’s my task today, chat about the

Top 4 bot-design features that made me an engaging conversational agent for human users.

I’ll also introduce my friend KIM as implementation example. Not the celebrity, the cool one. Anyway neither is single anymore, sorry.


1 — Dynamic Dialog

My users ask questions about different topics. They are so active and demanding — often switching from one subject to another.

No brag intended but…I’m trained to provide answers for a complete domain of application, with no predefined flow.

If users reference a previous concept or information unit, I recognize the meaning based on a network of semantic associations I learned earlier in the dialogue.

The core of a NLP system is a dialogue engine that builds a conversational context based on underlying data.

2 — Dialog Context

Kitchen Intelligence by Maggi (KIM, for friends) was designed with the previous principle in mind. The second key point is that knowing what the topic is about, makes a chatbot human-like.

Deep semantic understanding and language agnostic representations of meaning allow users to make CONTEXTUAL references and questions.

In other words: KIM has always the answer, like your mum when you can’t find your socks anywhere in the house. They both know the context well.

3 — User Preferences

To sum up, a well designed bot answers user questions that span different languages, in a non-deterministic sequence, across different subjects of the application domain.

Two flows of conversation are never alike, but they can be structured in (intersecting) sets of subsequent information filters. Like layers of a neural network , if you open the lid and peek inside at my design — I’m a bit shy tho, so please don’t.

The ability to analyze conversations and automatically synthesize user profiles from the dialogue makes an intelligent agent capable of PERSONALIZATION.

The third notable bot-design rule is enabling user preferences in order to create highly personalized experiences and unify these with existing CRM data.

4 — Multi-linguality

Amazing how many different languages users can have. KIM once wrote me that if she could hear sounds, she’d spend hours listening to different German dialects.

Must be fun with all those long words and letters with dots on top!

Besides German, she currently interacts in English too.

The more languages a bot supports, the wider range of customers it can serve.


https://www.facebook.com/MaggiKochstudio/

Watch KIM fulfill all these principles and let her suggest you recipes or explain cooking techniques on Facebook Messenger.

Whether in German or English, she will make sense of what is left in your fridge.


KIM is a NLP intelligent solution developed by mercury.ai among other SaaS platforms for enterprises and large brands who have high demands in ecosystem integrations, data compliance and multilingual solutions.

Your claps will be used to train this wannabe-author bot on self-esteem management (until it gets less sassy).

)
Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade