You are viewing a single comment's thread from:

RE: [Witness] Update – August 2025 | Backup Server | New Trending | Steemd

in #witness • 28 days ago

Twitter is Open Sourcing their recommendation algorithm. I haven't looked, yet, but maybe worth scanning through it for your Trending experimentation.

https://x.com/XEng/status/1965226798460887127

28 days ago in #witness by
$0.16
  • Past Payouts $0.16
  • - Author $0.08
  • - Curators $0.08
2 votes
Reply 3
Sort:  
  • Trending
    • [-]
     23 days ago 

    Vielen Dank für den Hinweis.

    Topic Social Proof Service (TSPS) serves as a centralized source for verifying topics related to Timelines and Notifications. By analyzing user's topic preferences, such as following or unfollowing, and employing semantic annotations and tweet embeddings from SimClusters, or other machine learning models, TSPS delivers highly relevant topics tailored to each user's interests.

    Sounds interesting in relation to the user feed. The question is whether and how, for example, each post is rated and whether the results are stored centrally somewhere so that the relevant topics for the user are then determined.

    Representation Scorer (RSX) serves as a centralized scoring system, offering SimClusters or other embedding-based scoring solutions as machine learning features.

    This is probably the central element of scoring...

    Page-Rank algorithm for calculating X User reputation.

    Very interesting...

    The languages are a little unusual (for me):

    Scala 66.4%
    Java 19.7%
    Starlark 5.5%
    Python 3.5%
    Thrift 2.1%
    C++ 1.6%
    Other 1.2%

    [-]
     23 days ago 

    I only looked briefly so far, but the follow recommendation service was the thing that caught my attention. Something like that might be a nice addition for condenser.

    I was also thinking of, maybe, another browser extension for that purpose or an update the the Steem Curation Extension. I think it might be challenging without a hivemind node or standalone DB, though.

    $0.22
    • Past Payouts $0.22
    • - Author $0.11
    • - Curators $0.11
    2 votes
    Reply
    [-]
     22 days ago 

    Yes, this service is also interesting. I often read about machine learning in the descriptions. Python, which is used in Hivemind, would be very suitable for this in principle. But I haven't worked with ML yet.

    In any case, it would certainly be a much better approach than looking at rigid criteria such as number of votes, resteems or comments...