Bringing Memes to LinkedIn
(via Content Classifiers)
The year is June 2021…
We are smack dab in the middle of COVID-19, and remote work is blurring the lines between everybody’s personal and professional lives.
In summer of 2021, I noticed LinkedIn posts geting… personal. My boss was constantly posting pictures of her dog, connections were raving about their favorite fitness apps or weekend reflections, and thought leaders were talking about their best tips on “decompressing” and “WLB”. I wanted our Feed to reflect this new “remote work” attitude, and worked with my AI team to build several new content classifiers: “personal” content classifiers, “promotional” content classifiers, “gray” content classifers. My favorite one — of course— was our “memes” content classifer (it was a hot topic in our company Slack too!)
No doubt, this was an exciting opportunity to boost the DAUs and Macrosessions on our site, but I was also excited to make content FUN again on LinkedIn - instead of seeing the same humble brag posts and influencers selling courses about B2B SaaS sales. We’ve seen over 40M+ posts a month now become eligible for viral distribution through these new classifers, and spark some very interesting conversations on the platform.
Defining a “Meme”
If you told me 10 years ago that my first big exec presentation at LinkedIn would be me showing a “shocked Pikachu face” meme to my VP - I would’ve thought you were crazy. But a little crazy is needed to pitch a platform overhaul like this. In my first PRD draft, I took the liberty of describing a meme as “an image with a text overlay that makes a reference to pop culture”.
Training the Classifier
We trained the meme classifier on over 10M+ posts each week, manually inspecting a few posts in each batch, and correcting the classifer until the accuracy of the classification was validated as statistically significant.