Stay with us

We are the human tech agency

We are Greenwood Campbell
16 October 2018 by Patrick Davis

Machine Learning & Natural Language Processing

Machine Learning & Natural Language Processing

Amongst the current digital buzz, certain technologies are left under appreciated. We will take a look at NLP (natural language processing) and ML (machine learning.) They’re both branches of artificial intelligence and they’re both very important, but it’s worth learning the difference and the reasons why a company might prioritise one over the other.

Let’s start with Machine Learning…

“The idea that systems can automatically learn from data in order to form patterns/stats and more - whilst simultaneously decision-making with minimal (perhaps none) human intervention. Their learning improves autonomously over time.”

Remember recently when Amazon recommended you a genuinely useful “related” product pre-checkout? Machine learning. Or perhaps the other night when Netflix recommended a show for you based on your habits? Machine learning. The recommendation engines are acting alone without the explicit need for programming - the basis of ML.  

Netflix have a very interesting way of utilising ML. Imagine a system split three ways, the first lot of data is collected from the global Netflix user base (what they’re watching & when.) The second lot of data is more specific, in-house Netflix staff and freelancers watch every single show (and every single minute of said shows) adding “tags” to briefly describe scenes or the entire show, such as “good cop” or “space thriller” for example. Finally, the third section is where a machine learning algorithm comes in. This is how the system combines the general user behaviour alongside the “tags” and then delivers recommendations. The algorithm even decides which shows are more relevant, for example a user watching 10 minutes of a show then abandoning it shouldn’t be considered, neither should a show watched 8 months ago (tastes can change!) so instead it prioritises recommendations based on your current or recent shows and films. The algorithm has created its own “taste communities” which are in the thousands across the globe, classifying their customers under types of “taste” and delivering content based on that.

This is really amazing when you think about it - there may be 500,000 users alone simply under the “sci-fi” community, but then there could be 1,768 users under a community that are almost exclusively “sci-fi and 80s drama fans” - all figured out by ML algorithms. It even handles language barriers and context issues, for example a “gritty drama” may not come across the same way as it does to an English user than to a French or German user.

Then there’s Natural Language Processing…

“A branch of AI where computers understand, interpret and manipulate human language just as we do. This can be in conjunction with machine learning or deep learning for example.”

A beautiful example of modern day NLP are voice assistants like Siri, Google or Alexa. Imagine you’re relaxing in your home and you happen to own a smart speaker, perhaps an Echo. A song is playing and you say “Alexa, I really love this song!” to which the volume lowers briefly and a human-like voice chirps back “Sure, I’ll save that rating!” From then on, the algorithm attached to your profile will begin to play that song more (and similar songs) when you listen to something like that again, or that radio station.  That’s natural language processing. The Alexa unit heard “I really love that song” and identified the unspoken intent - please play me this song more often and also more like it, and carried out the action of saving this rating for future reference. Not only this, but the interaction was answered by a completely coherent sentence in English in around 5 seconds.

NL & MLP together?

Here’s the best part - they’re not exclusive. The example of the Alexa interpretation also highlights uses of machine learning and/or deep learning in the background (for example, Alexa begins to understand that 77.4% of males and females aged 18-25 in the UK like the “Top 40” playlist, and so will recommend it more often to that demographic. This sort of thing is absolutely everywhere.) So, that’s NL & MLP! What would you like to see next from these marvellous technologies?