TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

What Makes TikTok's Algorithms So Effective?

5 点作者 Babawomba4 个月前

2 条评论

PaulHoule4 个月前
That article misses that TikTok gets effective negative feedback: they know when you don&#x27;t like a video. Contrast that to the interface that YouTube cribbed from [1]. When you show people a list of ten videos on the side you can&#x27;t come to any conclusion of whether or not people liked the videos they didn&#x27;t click on, in fact, I&#x27;m frequently frustrated because I also wanted to watch another video that was recommended and now it&#x27;s lost. If Facebook had a &quot;dislike&quot; button that worked that would fix many of the things wrong with it. [2]<p>I read many papers on &quot;negative sampling&quot; which is used in western recommender algorithms to guess at what you didn&#x27;t like because there&#x27;s no dislike button: I always thought &quot;I can&#x27;t believe it works&quot; and fact is, it doesn&#x27;t or it doesn&#x27;t work very well.<p>My RSS reader gets a thumbs up&#x2F;thumbs down and gets AU-ROC of 0.79, TikTok gets 0.84 or so and has a lot more data and 1000x the budget.<p>I&#x27;d almost say &quot;throw the western recommendation literature in the trash and treat it as a classification problem [3]&quot; except that when you have less than &lt;1000 data points for a user there is a cold start problem. You could throw AI and the computational power of a Dyson sphere at the problem and it won&#x27;t do as much as a dislike button that speaks with authority.<p>[1] <a href="https:&#x2F;&#x2F;montevallotimetravel.wordpress.com&#x2F;2012&#x2F;05&#x2F;03&#x2F;idiocracy-supplemental-film-viewing&#x2F;" rel="nofollow">https:&#x2F;&#x2F;montevallotimetravel.wordpress.com&#x2F;2012&#x2F;05&#x2F;03&#x2F;idiocr...</a><p>[2] would be good for people who don&#x27;t like hysterical posts about politics but it&#x27;s an existential threat to the advertising economy, influencers, etc...<p>[3] calibrated, even
phillipseamore4 个月前
I&#x27;ve always suspected that the secret sauce was that there are plenty of humans in the loop tagging and classifying, that&#x27;s a cheap option in China. It&#x27;s from a censorship regime after all. Today they likely have enough data to automate most of it.