28 September 2020
I’m thinking of making a simplified, email-only version of Findka just for essays (call it Findka Essays). Every week, you’ll get an email with links to 10 essays. Findka will record which links you click on, and the links in future emails will be chosen specifically for you (by algorithm). To submit essays, reply to the email and include the link(s) in your message.
It would look like this:
If you click on an essay but end up not liking it, just click “Show less like this.”
The goal is to deliver information with an extremely high signal-to-noise ratio and with minimal effort on your part. I also want to help out on the producer side of the problem: if you write an essay, Findka should find interested readers for you.
Since everyone gets a different set of links each week, we can avoid popularity bias. On Hacker News (and many other sites), a small percentage of the articles end up with most of the eyeballs. That's OK because HN is for discussion. If everyone read different articles, there wouldn't be much to talk about. However, that trade-off means (1) you see articles that are interesting to the community, not necessarily interesting to you, and (2) submitting an article to HN is like playing the lottery: you might get a ton of traffic, but usually you get none.
Findka Essays will also counter recency bias. The weekly links will be sampled from Findka's entire catalog of submitted essays, not just whatever was submitted during the past week. As a writer, this means you'll get a small amount of traffic on the week after you submit, and if your essay is good, it'll build up momentum from week to week. As a reader, you'll get content that stays relevant over time—the opposite of what social media promotes.
The focus on essays is important too. Good essays in my experience give you better return on your time spent than any other type of content. The main Findka website at the moment works for all content types, but doing just essays would make it easier to have high signal-to-noise in these early stages.
For the algorithm, I'll use standard recommender system techniques. Findka will use link click data to learn which articles appeal to similar people (collaborative filtering), and for new articles (i.e. articles for which we don’t have link click data) Findka will analyze the content and guess its similarity to other articles (content-based filtering). There’ll be plenty of opportunities for improving the algorithm as the user base grows.
I'll also do some manual curation for as long as I can manage. For example, I'll tag essays to help the algorithm gauge similarity, I'll filter out spam etc, and I'll boost essays that I think are particularly good.
One last thing. At the end of each email, I'll include the reminder: "Liked an essay? Email the author to say thanks!" Findka Essays may not facilitate community discussion, but it should be great for networking.
If you want to join the beta, send me an email: email@example.com. Include 3 – 5 links to essays you like. I’ll start sending the weekly emails once I get enough signups (50, perhaps).