By the time threads start getting multiple comments, they've gotten enough exposure from humans to upvote them to the front page if they deserve it.
More interesting is to scan new submissions, download the article and run a model to predict whether it's interesting enough to upvote. You can start by scanning old articles and trying to predict votes.
A major challenge is detecting dupes, follow-on articles and blogspam where the text looks relevant, but the topic has already been covered.
By the time threads start getting multiple comments, they've gotten enough exposure from humans to upvote them to the front page if they deserve it.
More interesting is to scan new submissions, download the article and run a model to predict whether it's interesting enough to upvote. You can start by scanning old articles and trying to predict votes.
A major challenge is detecting dupes, follow-on articles and blogspam where the text looks relevant, but the topic has already been covered.