Right, so it's Friday again. That probably means I should add some generic comment about the week flying by too fast (because it has, dammit). Rather than dawdle on that subject, I'll launch right into what's kept me busy this week.
There's a new "Trending on TikTok" chart, visible here. It's still super duper beta, but chances are you'll be hearing more about this in the future :)
For TikTok videos I've also added the ability to track how long the average video actually is. Here's the breakdown by percentile:
So, what's the takeaway? The average video is 15 seconds long, with the bulk falling between 11 and 50 seconds. Anything below 10 seconds is probably worth warning artists, so I've added a note to each influencer's profile where said case applies.
Fully moved away from the "upvote" / "downvote" approach. The new breakdown (which has been live in some form or shape for a little while now) can be seen below:
I also made it easier to cancel vote requests and/or add more votes to an existing request. Hard to believe that for the last 2 years this hasn't been possible at all, so... that's probably why Hot or Not is still in "beta" :)
Geez this one was tough. We're using a mongodb database here. One way to search is using a $text query, which requires creating an index. Mongodb then gets really smart and tries to "stem" your possible search terms. The word "amazing" becomes "amaze" and "amazed" so that they can do fuzzy searches and figure out if someone's near the target. Only problem with this system is that it can't handle partial queries. Doing a $regex search is the solution, except that you can't actually use an index when searching this way.
If you're confused, that's okay. So was I.
Finally, after spending some time with mongodb's customer support team, they suggested that I set up an autocomplete Atlas Search index. And after a bit of jiggling I did indeed manage to get it working. The end result is that curators can now search their feed with partial searches.
There's a pretty in-depth audio fingerprint that Spotify gives for each song via their API, measuring things like bpm, danceability, energy, vocal presence, etc. I decided to play with it a bit this week. Below is an example of the outcome.
The practical application of this data does hold some potential, though I'll need to dig closer to see how dependable it actually will be.
Some examples of how it could be used: