Algorithmic feeds in clients
Markus Unterwaditzer wrote a blog that does into the practical problems that he encounters with Mastodon’s chronological timeline. He notes that the lack of algorithmic feed makes it hard to connect with followers in different timezones, or followers who do not post a lot. Mastodon proposes lists as a partial solution to help with organising and taking control of your feed. But as the blog notes, lists require a high amount of effort of people to use and maintain, significantly more than what the average person can be expected to be willing to put into.
Algorithmic feeds are not impossible in the fediverse. Most fediverse servers however have pushed the option to implement algorithmic feeds away from the server to the client. There are separate solutions that provide an algorithmic view on someones personal Mastodon timeline, such as fediview and FeedSeer. The way these systems work is that they ingest all the posts on your timeline, and then sort them via simple, easy-to-understand algorithms that the user can choose from. They effectively function quite similar to other 3rd party clients as Phanpy or Elk, with the exception that they only provide the algorithmically sorted timelines.
@Moof says here that he was hoping there to be more mobile clients that would have incorporated some sort of algorithmic view at this point. I personally also would have also expected more to be happening in this space. The most notable clients who provide some form of algorithmic feed are Mammoth and SoraSNS. One issue seems to be that placing the algorithm in the client instead of in the server introduces significant lag in the client. Both Fediview and Feedseer can take quite a while to load, Fediview took more than 20 seconds to load. This is also visible in SoraSNS, where I scroll quicker than the algorithm can calculate the next post, introducing frequent pauses where the timeline is still loading the next post. Although I would love to see more experimenting with algorithmic timelines in fediverse clients, I’m not sure if the barrier of long loading times will be easy to overcome if the client also has to do the algorithmic calculations.
