By The Numbers: New York Fashion Week
Or, How ZeroBot Helped Celebrate Michael Jordan’s Birthday
Organizing New York Fashion Week is hard. But ZeroBot (ZB), our algorithm, has been working hard to produce accurate stories of all the unique shows taking place by automatically sorting through the tens of thousands of photos, videos and Tweets being shared on socials, grouping them into individual stories based on the keywords and finally, naming each story.
ZB is by no means perfect (yet), but its ability to group and name stories has become incredibly robust. NYFW actually helped us refine its abilities to organize; and seeing the discrepancies and gaps in knowledge helped our CTO/Co-founder, Masha, make it better and more badass.
Let’s take a look at the before and after:
Before: Poor ZeroBot
We’ll be honest — NYFW helped us refine ZB because the first stories that came out of it were, for lack of a better word, ugly. Separate shows were bleeding into one another and creating one massive, disconnected story because the words “New York” and “Fashion” were used in almost every post.
On the first day of NYFW, ZB couldn’t discern that there were multiple stories happening:
ZB was doing the best job it knew how: it detected a spike in posts, found major keywords being used across them, saw that they were being used across a relatively “small” area (one city), and named it “New York Fashion Week.” As our CEO/Co-founder Ryan alluded to in his recent blog post, “Phish Phucked Us,” getting a clean story in a location as busy and stacked as New York City is extremely hard to do. As you can see above, there is a lot of disconnected noise that makes its way into our first NYFW story; some noise has absolutely nothing to do with it.
During: Theoretical ZeroBot Experimenting
Masha is constantly coming up with theories on how to enhance ZeroBot and experiments like crazy (no one has been able to disprove her theories so far).
She saw that ZB’s parameters were too limited to account for the nuances and challenges presented by massive social media inundation.
Masha presented a different way for ZB to approach story creation — instead of finding all the similarities between posts, find all the differences — and use that as the way to determine whether a post belongs within the set used to tell a story.
To dive further into this, take a look at the below snapshot of our title image:
These are the keywords that every post from NYFW have in common — cool! We can tell there’s a big event going on in town because thousands of people are referencing the same seven keywords. But just because they all share these commonalities does not mean everyone is referencing the same exact thing.
If we zoom back out to look at the entire map, we see that every individual story (green circle) has its own set of keywords that set it apart from the others:
After: Good Job, ZeroBot!
Masha’s theory proved to be right. After updating ZB to look for the differences in keywords, one giant, noisy story about NYFW became many nuanced stories representing designer-specific shows.
A few examples:
We’ve seen this more targeted approach work incredibly well across all our stories; we even found a story in Chicago, IL celebrating Michael Jordan’s 53rd birthday yesterday:
Storytelling with Social Media
It’s pretty amazing how, literally overnight, such a small change in our algorithm made an incredibly big difference in Zero Slant’s storytelling abilities. We’ll continue to share our mistakes, problems, and — eventually — solutions as we keep moving down this path and towards our official launch (coming soon!).
As always, feel free to reach out to us with any questions or comments at team@zeroslant.com.
