The second half of the article describes that in some detail.
The "is here" has no link going anywhere.
> we look at nine metrics that we think are good proxies for measuring what languages people are programming in. Sources include Github, Google, Stack Overflow, Twitter and IEEE Xplore. The raw data is normalized and weighted according to the different rankings offered—for example, the Spectrum default ranking is heavily weighted towards the interests of IEEE members, while Trending puts more weight on forums and social media metrics
So they looked at sources, calculated metrics, and then weighed them by rankings? Still no idea what they've actually done.