Jacob O'Bryant
Home · Archive · Twitter · GitHub · LinkedIn
Why did The Sample send me this newsletter?
27 July 2021

When you sign up for The Sample, you can specify different topics that you're interested in. However, these topics are used differently than many people expect.

screenshot of The Sample's landing page

The algorithm uses the keywords you choose to improve your recommendations, but it doesn't restrict itself to recommending things that fall within those keywords (indeed, a lot of newsletters don't fit any of the keywords very well). If you say you're interested in design, then you'll get design-related newsletters more often than most people, but the algorithm will still try to introduce you to new things outside your current interests.

Why does it do that? Partly it's because the whole point of The Sample is to increase the variety of information you get. If you want to stay solidly within a certain niche, you can join relevant communities. For example, I'm into the Clojure programming language. I frequent the Clojure subreddit and the Clojure Slack workspace, and I follow various Clojurists on Twitter. I don't need a fancy algorithm to help me discover new Clojure-related content. The Sample is meant to span across different communities and make connections that normally would go unnoticed. At some point we'll redesign the landing page to hopefully make that clear.

The other reason is practical: the algorithm's topic modeling is extremely fuzzy. Even if we wanted the algorithm to stay within a particular niche, the algorithm isn't smart enough to do it. It doesn't think "this is a finance newsletter," it thinks "this newsletter is 10% more likely to be related to finance than the average newsletter."

With more work, we could change that. The Sample is only a few months old, and there's plenty of room for improvement. However, as mentioned, rock-solid topic modeling isn't a huge priority for us. We mainly use an approach called collaborative filtering—"people who like X also like Y," without any thought for what X and Y are. In other words, when you give a particular newsletter a positive rating, we look for other users who also liked that newsletter, and we send you other newsletters that those people liked. It works similarly with the topics: if you're interested in productivity, we don't necessarily send you newsletters about productivity. We send you newsletters that are rated positively by people who are into productivity. Or at least, we try to!