Two Approaches to Solving the "Quiet Fediverse" Problem: Conversation Backfilling Mechanisms
hackers.pub
This post addresses the "quiet fediverse" problem, where users encounter fragmented conversations with missing replies in decentralized social networks. It explores the underlying causes related to the distributed nature of the ActivityPub protocol, which scatters conversations across different servers. The author delves into two primary approaches to solve this issue: reply tree crawling and the context owner-based method. Reply tree crawling involves fetching all replies to reconstruct the conversation, while the context owner method relies on a designated owner to manage and provide the complete conversation. The post examines the technical mechanisms, advantages, and disadvantages of each approach, including real-world implementations and ongoing debates within the fediverse community. It also touches on additional backfill mechanisms like periodic crawling, user-triggered backfills, and mention-based discovery. Ultimately, the author advocates for a hybrid approach that combines multiple strategies to enhance conversation completeness and user experience, highlighting the importance of ongoing standardization efforts and collaborative implementations within the fediverse. This article provides valuable insights into the challenges and potential solutions for creating a more connected and coherent social networking experience in decentralized environments.
  • freamon@preferred.socialEnglish
    0·
    21 days ago

    I’d be wary of getting a conversation node from anybody other than the original author (as described in the second approach).

    There’s a reason why, if you want to resolve a missing post in Lemmy, etc, you have to use the fedi-link to retrieve it from its source, not just from any other instance that has a copy (because, like the “context owner”, they could be lying).

    For Group-based apps, conversation backfill is mostly an issue for new instances, who might have a community’s posts (from its outbox), but will be missing old comments. Comments can be automatically and recursively retrieved when they are replied to or upvoted by a remote actor, but fetching from the source (as you arguably should do) is complicated by instances closing (there’s still loads of comments from feddit.de and kbin.social out there - it will be much worse when lemm.ee disappears). So perhaps Lemmy could also benefit from post authors being considered the trusted owner of any comments they receive.