Being a conversation artist takes practice. You have to be extremely socially aware, have the right body language, say the right things, ask the right questions to keep the conversation going. The person you’re talking to may reciprocate energy, they may bring the tone down to a whisper, or they might carry the conversation without your help. We have studied how to effectively communicate with humans for centuries. We think about it all the time when we construct our texts and emails or asking someone on a date.
But, when we talk to our data, the conversation might be a little stale.
As a Product Manager, the conversation with your data usually takes place in the form of a SQL query. You have the option of writing the query yourself—which is tedious and often fails—or asking an engineer to write it for you. Instead of talking to data in a SQL query, it’s much more manageable and less time consuming to ask your data questions another way.
Enter Natural Language.
How we communicate with AI technology has a different set of rules that seem rather simple. We can say whatever we want and still get an output. We can yell mean things at Siri or Alexa and they will answer in a cordial and sometimes witty way. However, how we communicate with AI in a natural language is just as complicated.
By definition, a natural language is a form of language (English, French, Spanish, etc.) that humans speak. A computer language or programming language is a language computers speak (SQL, Java, C++). Humans have cracked the code on talking to computers, given the popularity of coding careers and customizability in programming.
Having computers understand human language is more complicated. A big term in AI is Natural Language Processing (NLP), which is a sub-field of AI that takes text or audio and—based on data—begins to understand the human language. NLP uses ML to understand natural language. However, to avoid complication we will approach this article from a no-code machine learning standpoint.
Read more on the reasons to master no-code ML.
This Leads Me Into Talking to Our Data—Meet Natural Language Powered Data Science.
When talking to our data, the conversation might be stale at first because you don’t know what to talk about. You’re looking at all this data and it can be overwhelming. I can relate it to not knowing what to talk about with someone you don’t know too well.
We will apply these tips to two categories of natural language powered data science:
Natural Language Predictions - Based on a dataset, predicting in a natural language how likely something is to happen or not happen.
Natural Language Analytics - Uncovering hidden insights in your data, by asking a question in a natural language and being answered in text or graph format.
Get to Know Your Data
How do you break the ice? The first step is identifying what you want to get out of the data. You should learn what attributes are in your dataset and how they relate to the overall problem you’re trying to solve. Remember, you’re probably trying to create action items from your dataset. Out of your dataset, identify a predictor column and an identifier column.
Example: You’re trying to figure out whether to cut out a feature in your product. You don’t want any unnecessary features that don’t contribute to another in-app purchase. You narrowed it down to three features. Image the dataset had a “1”, meaning Yes, or a “0”, meaning No in the features column indicating if the user uses those features.
The question should look like this:
How does [Feature 1], [Feature 2] and [Feature 3] cause [In-App Purchase] for all [Users]?
Talk to your data in X’s and Y’s
Your dataset will be great at comparing attributes. For comparisons and analyzing your data, you want an X and Y category, preferably one that compares qualitative data vs quantitive data. Say you’re in the retail space and you want to do a simple price vs purchase graph. With Natural Language Analytics, you could ask your data “What’s the total price by purchases” and get a comparison that can create value for your team.
Okay, now you’ve passed the small talk phase.
You usually get more information from asking someone “What was the best part of your day?” instead of asking someone “How was your day?”
The same goes for data. Asking “What customers will churn?” instead of asking “Which of my top 100 customers will churn in the next month?” doesn’t create as much actionable value.
Don’t Only Talk to Your Data 1 on 1
Sorry to say this, but your data doesn’t just want to be in a conversation with just you. Let it meet your friends in a group setting.
Collaborative ML is extremely valuable to avoid bias or misinformation. Adding collaboration to your machine learning process spreads information faster and increases the governance of your inputs and outputs. Read more on how collaborative ML decreases bias.
Natural Language Data Science Makes Any Data Queryable.
Not everyone speaks programming languages. But, with a machine learning process that uses natural language predictions and analytics makes any dataset queryable. With Natural Language Powered Data Science, you can make your ML process collaborative and more efficient by simply asking a question in plain English.