About CombFlow and HiveComb
10 years of social media on the blockchain. Yes, today is my 10th Hive anniversary!
We started just when the big web2 networks concluded their switch from a chronological feed to an algorithmic one.
We do have our own version of an algorithm, but as that only reflects upvotes its output isn't made to give users content they really care about. And for creators it means you either build something great for hive, or you get a good amount of autovotes, there's no other way to hit trending and be seen.
While "the algorithms" are under constant criticism for being addictive and messing with our dopamine systems, everyone who uses social media today relies on them to be entertained. It's like with any other drug, we know it's bad for us, but life without is just so dull.
For years I have said that this dullness is one of the biggest reasons that Hive is bleeding users. They don't get the hits they are used to get elsewhere. There's nothing to do except writing a post and having look what the people you're already following post. Tags help a bit, but they're a mess with everyonee trying to use them to fish for attention. All of that is especially problematic for newly onboarded users.
Luckily, the state of AI solves a lot problems I had with just going ahead and implementing something. It can create a set of vectors (called centroids) from thousands of posts it autonomously classifies, which then allows a simple comparison of new posts against these vectors to classify the whole blockchain.
So here we are, CombFlow is the first content discovery engine on Hive with real life categories. To show how this changes the browsing experience, it is accompanied by a dedicated user interface: HiveComb
But the real kicker is that the API is completely open, and can easily be used by all Hive interfaces to offer the same filters to their users.
It's far from the algorithms elsewhere, as it doesn't take user actions into account at all. Yet at least. That can only be done on the frontend side, and if it's advisable to have this data stored for and used by a public API is a discussion for another place and time.
About the proposal
The amount is obviously symbolic. It's calculated to cover the server cost and Claude subscription, and it honestly doesn't matter if its funded or not.
What I do need is attention and help though. The list of categories is still worth improving, and there are also quite a few misclassified posts. There's a report feature for misclassifications on the site for logged in users. Every manual correction of current classifications helps to improve the next round of centroid generation.
If you want to get involved beyond that, have discussions about algorithms, user data, and how we can ethically create a more engaging experience on Hive, come and visit the Discord