When I was at Wired, I dutifully added tags and categories to stories I wrote and edited, but I was never sure I ever understood why we all bothered, other than that’s what one was supposed to do. It just seemed like drudgery.
But earlier this month, I discovered a new appreciation for tags and categories.
(A confession first: I was never a great tagger. If you want a great tagger and a great writer, Tim Carmody is who you want. His tagging for Wired’s business section was exquisite.)
I’ve always hated the experience for readers who land on a tag or category page. Inevitably, it will just be a giant list of posts in reverse chronological order, with no sense of what’s important or even what that tag or category is about. They are like anti-topic pages, and the reason that Wikipedia has thrived.
I asked once, somewhere on social, probably on Twitter, why we all even bothered. If my memory serves, Dylan Tweney, now editor in chief of VentureBeat, said tags and categories were good for SEO. I retorted that Google ought to ban those pages from search results since they are just awful for users.
Noah Shachtman, now the editor in chief of The Daily Beast, said he actually likes clicking on tags. I will admit I occasionally do this, but only when it’s a very obscure topic and want to see if there’s more on it.
But earlier this month, one of Contextly’s clients accidentally published the exact same news story twice and didn’t notice. The only differences between the two posts were that one had a photo, 2 categories and 4 tags; the other had only 2 categories, but no tags and no photo.
From a data perspective, this was *really* interesting.
Contextly uses quite a few factors to make our related recommendations (which are only half of what we do). These factors include semantic entities we identify in headlines and body text, link structure, curation (by which we mean editors and writers using our tools to specify what stories are related), along with tags and categories.
And just that difference in tags led to significantly different sets of recommendations, and the ones on the version with more tags looked much more interesting to me. Our system optimizes over time, so we’re keeping our eye on this interesting pair to see what happens. (These are the identical twins of publishing for us.)
But, the key thing that was made clear to me is how tags and categories work as curation or smart human input into machine learning. That fits so well with Contextly’s model that marries curation to machine learning. We do that to create engaging content recommendations that keep readers reading.
So, tags and categories, I’m sorry I doubted you. It turns out you are useful after all.
And as for those artless tag and category pages? Don’t worry, we’ve got plans for them, so just keep up the tagging for now.
Image: Matthew Anderson/CC