Neural networks and machine learning:
Get the news you didn´t even know you wanted
Neural network is the buzzword of the moment, but what is it, and how does it work? Swedish media group Mittmedia uses the machine learning method in several of their development projects in order to build a modern news network platform that’s ready to take on the future.
If Artificial Intelligence is the goal, artificial neural networks is one of the crucial tools required to make it happen. The focus is no longer on throwing massive amounts of computer power at a problem in order to realize the goal slightly faster than a human being would, like we did in the 90s. Today’s scope include identifying patterns in data that make it possible to search more efficiently, and to better categorize and understand content like images, speech and music – or to make predictions about what will happen next. An artificial neural network may already know how this article will end, but you yourself have to continue reading to find out!
How it works /Advanced machine learning
First of all – let’s get the terminology straight. A neural network is a brain. An artificial neural network is a simulated brain created for the purpose of recognizing patterns and making decisions in a humanlike way. What’s so great about the artificial neural network is that you don’t have to explicitly program it to learn: it does that all by itself, just like a brain. We call this process machine learning, and the artificial neural network is but one model of many used to make computers learn.
Just like a brain, an artificial neural network needs input in the form of data to process information and get results. The more data, the more accurate the network will become. Think of it as any task you do over and over again. With time, you’ll gradually become more efficient and make fewer mistakes. The same principle applies to the computer; by testing and checking the results, it evolves and learns how to achieve the best possible outcome.
In the computer’s case, interconnected nodes replace the brain cells, each one responsible for one simple computation. Working in conjunction with each other, they process the incoming data and forward the result to the next layer of nodes, and so on and so on until the task is completed.
So what’s it good for?
Well, almost anything actually. Neural nets are involved in many of the major breakthroughs in AI today. They are an important feature in everything from language translation and voice comprehension in apps, to the development of self-driving cars.
Media house employs neural networks
Mittmedia, a Swedish news publicist housing around 30 different titles is a Collab to Gävle Innovation Arena. Magnus Engström is Head of Data Strategies at Mittmedia/Bonnier and responsible for the Group’s IT strategies:
– We’ve worked hard on identifying what it is the customer wants over the last few years. As an example, we use artificial neural networks to predict future user subscription cancellations.
By processing data on subscriber behavior, their system can identify patterns in parameters for a certain type of subscriber before the subscription is cancelled. Such data include demographic parameters like age, sex or location, as well as behaviors such as daily usage, what time the usage occurs, amount of time spent etcetera.
Based on data pulled from other users, Mittmedia can then identify high risk subscribers before they cancel their subscription, and makes it possible for the system to take action – for example by sending push notifications encouraging the user to read more, or by applying other measures that will likely lower the risk of the subscription being cancelled.
Robot generated personal news
Mittmedia also uses machine learning to identify what news a user is most likely to read, and match it with new users showcasing the same parameter and behavior profile. This enables them to personalize the news and promote articles for a certain type of user and not others, and to make sure that the first piece of news you see is the one you are most interested in.
Additionally, the system has a social value, Magnus explains. Digital media outlets can for example use clustering based on where the users live and their behavior.
– If we have information about a water shortage that needs to reach the public in Gävle, personalization is all about making sure that news are distributed to those who are affected by it.
Magnus can see several other societal benefits with machine learning based on neural networks.
– If you visit a hospital today, it is likely because you are already sick or hurt. In the future we can train neural networks to predict who is at risk of developing which illnesses, and the health care industry can take action and apply preventive measures long before the symptoms show.
Machine learning in the form of neural networks is part of many great future inventions, such as self-driving vehicles, artificial limbs that can communicate with the brain and smart homes.
Magnus is convinced that this area of research is something that will affect the whole of society in the future.
– Things often take longer than expected, but then turn out to have a much greater impact than one was able to imagine. I think this is especially true for machine learning and the use of neural networks.
By Malin Hefvelin