Some examples include your feed not reflecting tag preferences, too many listicles, and more recent posts rather than relevant posts... We realized we are unintentionally giving reactions too much weight when calculating the aforementioned score of a post.
It is worth noting, of course, that the extra weight (previously) put on reactions is not totally unfounded. After all, it is one of the main metrics for user engagement.
The "too-many-listicles" phenomenon only occurs simply because listicles are inherently designed to be bookmarked (i.e. one of the three reaction types that DEV considers). Hence, we end up in a situation where in addition to the hearts and unicorns a listicle receives, the extra bookmarks greatly amplify what would have otherwise been a regular article.
However, the same argument can be said about regular articles, tutorials, and visualizations. That is, they can also "game the Algorithm" by being designed to be bookmarked—intentionally or otherwise. But, what separates listicles from regular articles is that they are more easily automated than full articles, guides, etc. And that is exactly why the feed becomes bloated. To make a long story short, automation and bookmarks make a perfect storm.
Unfortunately, I have no recommendations for addressing this at the moment. It's really tricky because—as mentioned earlier—certain patterns are just inherently gameable.
Anyway, with that said, I just wanted to bring some additional perspective to the table. Looking forward to hearing about the proposed solutions! 🚀
It is worth noting, of course, that the extra weight (previously) put on reactions is not totally unfounded. After all, it is one of the main metrics for user engagement.
The "too-many-listicles" phenomenon only occurs simply because listicles are inherently designed to be bookmarked (i.e. one of the three reaction types that DEV considers). Hence, we end up in a situation where in addition to the hearts and unicorns a listicle receives, the extra bookmarks greatly amplify what would have otherwise been a regular article.
However, the same argument can be said about regular articles, tutorials, and visualizations. That is, they can also "game the Algorithm" by being designed to be bookmarked—intentionally or otherwise. But, what separates listicles from regular articles is that they are more easily automated than full articles, guides, etc. And that is exactly why the feed becomes bloated. To make a long story short, automation and bookmarks make a perfect storm.
Unfortunately, I have no recommendations for addressing this at the moment. It's really tricky because—as mentioned earlier—certain patterns are just inherently gameable.
Anyway, with that said, I just wanted to bring some additional perspective to the table. Looking forward to hearing about the proposed solutions! 🚀
This is a great insight! 🤔 We should test treating bookmarks differently from other reactions. I am logging it here for review.
Thanks for the great suggestion!