This is a case where an algorithmic feed would make this into a truly amazing product.
I don't want email spam from every follower, but do want to see the best snippets on substack.
Super easy to create a for you page, given text content:
Step 1: embed every article I ever read, or note I liked/comment/share and stuff it in a DB.
Step 2: every time a new note is posted (by anyone anywhere), embed it, search the db for my last 100 embedded items, and see if new note has relevance > 50%. If so, add it to my feed inventory. Resort my feed inventory by semantic relevance every hour. Remove items older than 7 days from feed inventory every hour.
Step 3: On page load, move everything in my feed inventory to my feed archive - never rerank again. (Bonus points for tracking note level views rather than assuming all were viewed, but small detail).
Bonus Step 4: Every 4th item in my feed inventory, intersperse something that's solely there based on popularity/top liked note of all notes visible to all of my followings. i.e., show me something possibly irrelevant but viral.
That'll get you pretty far, each step can be endlessly optimized over time.
I want to see the results of this so bad that I'll volunteer to build v1 this weekend if you really don't have time to do it internally. Tiktok for text... could be amazing.
The terrible idea is only having an algorithmic feed. If you always offer the user the option to not use it and instead have a feed they can control, then it’s a non issue
As soon as the site is owned by a public company, it's over. They are legally required to have an algorithm only feed, as this gives the best engagement, regardless of how much the userbase might hate it.
Algorithmic feeds generate more user engagement, that is what they are for. That's why platforms built around following updates from people that you like can't help themselves but implement them.
I don't want email spam from every follower, but do want to see the best snippets on substack.
Super easy to create a for you page, given text content:
Step 1: embed every article I ever read, or note I liked/comment/share and stuff it in a DB.
Step 2: every time a new note is posted (by anyone anywhere), embed it, search the db for my last 100 embedded items, and see if new note has relevance > 50%. If so, add it to my feed inventory. Resort my feed inventory by semantic relevance every hour. Remove items older than 7 days from feed inventory every hour.
Step 3: On page load, move everything in my feed inventory to my feed archive - never rerank again. (Bonus points for tracking note level views rather than assuming all were viewed, but small detail).
Bonus Step 4: Every 4th item in my feed inventory, intersperse something that's solely there based on popularity/top liked note of all notes visible to all of my followings. i.e., show me something possibly irrelevant but viral.
That'll get you pretty far, each step can be endlessly optimized over time.
I want to see the results of this so bad that I'll volunteer to build v1 this weekend if you really don't have time to do it internally. Tiktok for text... could be amazing.