News stories vary in importance and can signal both increased risk and new opportunities. Either way, catching them can be crucial. Here you can read more about how we use machine learning to separate what is relevant from what probably doesn't matter to you.
1. Connect company and news item
First, an algorithm works out which company an article is about. This is no simple task, and we are now on version 6 of this system. In short, it establishes the link between a company and a news item in a way that scales when you need to keep track of every company and organization in Norway.
2. Determine relevance
The next step is to work out how important or relevant a given article is. For this we use a variant of a language model called BERT. BERT is a language understanding technology from Google that can tell the difference between words used in different contexts. Traditional approaches have often relied on word counting, but those are easier to fool, so this deeper understanding gives you a more accurate assessment of the text.
3. Training with machine learning
The model still needs to know what to look for, so we have built a large set of examples: we have gone through many texts and assessed how relevant they are to different types of users. For instance, we distinguish between relevance for sales and relevance for risk. The machine is then trained to reproduce that classification.
We can also use anonymized customer behavior for classification, looking at which types of articles our users click through to. These data points are “noisier” and require more processing afterwards, but they are something we may consider in order to build an even broader data foundation. .
4. Classification of categories
We also assess which categories of articles are generally useful or not useful, often using manual methods. This is because a number of articles are machine generated, yet still contain the same expressions and phrases. One example is news stories about changes that correspond to announcements from Brønnøysund, which we already monitor separately. Those are therefore removed so you don't get the same alert twice.
How can I try it out?
By setting up a portfolio of your own customers, prospects or anyone else you want to keep an eye on in the Enin web portal, you can receive alerts whenever relevant events occur.
Events your competitors might miss. Which is rarely a bad thing.
Get in touch with us to find out more!
