PUTGET SMART NEWS™ AI based article suggestions

by Magnus Enderstein — 18/01/2019

How AI is solving the information overload

Screenshoot Article Suggester panel 

Who has not faced the information overload problem today?

You work in a particular area, you have personal interests, you have kids who like a specific sport. All of this requires your attention and interest.

You like modern art and traveling, your kids love football, on your job you are expected to know everything about accounting or marketing.

Obviously, you want to keep informed about what is happening in these areas.

An internet search for “modern art” or “traveling” or “football” or for “accounting” returns a total search result of 2 040 000 000 items and about 116 000 000 news articles. 

How would you be able to cope with this situation without compromising your life?

Can technology help? Yes, we think so.

Take the case above, manually reading 116 000 000 news articles is completely impossible, anyone understands that.

Anyway, assume you would have the time and interest to read all these 116 000 000 news articles. How would you go for this task?

You might start splitting the material into different parts. For example, you may see articles in different languages, different topics, different dates, different lengths, with or without pictures etc. etc. Some articles are very positive and some articles are very negative towards the central topic. You will find long format articles and short format articles etc.

This sorting will help you categorize and classify the articles into manageable groups. You already now see that this task is not completely impossible to solve. A pattern begins to occur. Given your personal interest, this classification has significantly reduced the number of articles you really have to read. 

This is how we have thought when we developed our Smart Article Recommender based on AI Artificial Intelligence. By collecting thousands of articles, read them mechanically, processing the data applying text analysis, adding metadata, categorize and classify the articles, a pattern will begin to occur. The machine will do the job for you. You may say “How can a machine read and understand a text”. Well at the end of the day text is just a set of data, something that computers are very good at processing.

Sorting and classifying articles is not all you want. Every day you will face loads of new articles being published on the internet. How can you find all these, relevant to you? 

By teaching the computer to classify the news flow and knowing what interests areas you have, it can start working for you and just presenting what you're interested in. 

And now you may say that “I have different interests over time”. That's right. Therefore, we have created an additional process that will help solve this problem. We have applied Machine Learning also called  AI Artificial Intelligence. It is simply different methods that help us solve the problem sometimes also called algorithms. Algorithms are nothing new. The man has been using algorithms for thousands of years without computers. The difference between using computers to handle algorithms is that computers are much better than us to manage many and complex problem solutions at the same time and very fast.

So by classifying the texts and then using our algorithms to learn what the reader wants, we can deliver more accurate results than what a human can.

How well does this work in practice then? The answer is that we do not know exactly. However, we know that it works technical very well (the computer is doing the job) and that we ourselves are very pleased when we use our Article Recommender to monitor our interest areas. 

It's like all learning. You have to constantly learn new and practice over and over again. We also have some tricks we use to make learning easier for the machine. Exactly what any normal student does. The tricks, for example, is about making the sample selections not too extensive but rather focused. It facilitates the likelihood of the machine finding the right texts and it will be easier to learn.

I sometimes compare our approach to this with a Coca-cola recipe or a “best of breed” approach. We know what the customer wants, but we have to change the recipe from time to time as the world is constantly changing. 

If you want to try how this works please contact us and we will show you. Smart Recommender is a premium features we offer in our product range.