经济与社会科学2019arxiv.org
YouTube 的激进化兔子洞
YouTube’s Rabbit Hole of Radicalization
进阶 · 约 35 分钟读完前置知识：推荐系统基础、内容分类方法、描述性统计分析
导读
这篇论文通过给近 800 个政治类 YouTube 频道分类，追踪其推荐流量如何在各类频道之间流动，以此量化"YouTube 算法助推激进化的兔子洞"这一流行说法。作者得出的结论与主流指控相反：算法倾向于把观众推向主流媒体和有线新闻，而非独立或极端频道，因此并未助长激进化内容。
HN 关注点：结论直接挑战了媒体和学界广泛接受的"YouTube 兔子洞"叙事，HN 读者围绕其方法和问题定义展开了激烈争论。
HN 评论精选
这几条评论里：质疑 3 补充 1
补充转述论文摘要：算法实际上抑制而非助长极端内容推荐
为方便无法在设备上读 PDF 的人：论文称记者和学者都暗示 YouTube 及其幕后推荐算法在助推网络激进化。本研究通过检验算法在推荐激进化内容中的角色，直接量化这些说法。在将近 800 个政治频道分类后，我们区分出不同的政治图谱，分析算法在各类之间的流量。详细分析后，我们反驳了流行的激进化说法：数据显示 YouTube 推荐算法反而主动劝阻观众访问激进或极端内容，算法偏爱主流媒体和有线新闻，而非有左倾或政治中立倾向的独立 YouTube 频道。作者推文补充：2019 年末的算法"摧毁"了阴谋论者、挑衅者和白人身份主义者，帮了党派人士，伤了几乎其他所有人。
dfabulich查看英文原文
For the convenience of those who can't conveniently read a PDF on their current device: > The role that YouTube and its behind-the-scenes recommendation algorithm plays in encouraging online radicalization has been suggested by both journalists and academics alike. This study directly quantifies these claims by examining the role that YouTubes algorithm plays in suggesting radicalized content. After categorizing nearly 800 political channels, we were able to differentiate between political schemas in order to analyze the algorithm traffic flows out and between each group. > After conducting a detailed analysis of recommendations received by each channel type, we refute the popular radicalization claims. To the contrary, these data suggest that YouTubes recommendation algorithm actively discourages viewers from visiting radicalizing or extremist content. Instead, the algorithm is shown to favor mainstream media and cable news content over independent YouTube channels with slant towards left-leaning or politically neutral channels. Our study thus suggests that YouTubes recommendation algorithm fails to promote inflammatory or radicalized content, as previously claimed by several outlets. Tweet from the author: https://twitter.com/mark_ledwich/status/1210743168246771716 > It turns out the late 2019 algorithm DESTROYS conspiracy theorists, provocateurs and white identitarians > Helps partisans > Hurts almost everyone else.
质疑方法学质疑：只测了匿名账号的基线推荐，未覆盖有观看历史后的推荐路径
这段引文值得质疑：抓取数据和 YouTube API 呈现的是"匿名账号"看到的推荐，即该账号没看过任何视频、保持中性基线推荐。而许多关于激进化的讨论谈的是某人看了若干视频之后出现的推荐——算法开始识别出用户所属的细分领域，然后推荐更多同类内容。盲目的推荐系统似乎无法摆脱这种处境，这更像是其固有问题。什么才是更好的办法？更好的教育让人具备辨别能力？还是让算法标注分类、让用户自己选择想看或屏蔽哪些类别？
joe_the_user查看英文原文
This quote raises questions: "The scraped data, as well as the YouTube API, provides usa view of the recommendations presented to an anonymous account. In other words, the account has not ”watched” any videos, retaining the neutral baseline recommendations, described in further detail by YouTube in their recent paper that explains the inner workings of the recommendation algorithm[38]. One should note that the recommendations list provided to a user who has an account and who is logged into YouTube might differ from the list presented to this anonymous account." Many discussions of radicalization talk about recommendations coming after someone has watched a number of videos. This situation is that after N videos, the algorithm starts to see a pattern, a niche, that the user fits into and then recommends more niche content. Once the user watches the niche content, just more niche content is recommended, or that-niche and mainstream content is recommended and the mainstream content is boring. In a sense, I don't see how a blind recommendation system could escape this situation. It seems more an inherent problem of blind recommendation systems. What could be a better approach? * Better education in the US and the world so people have BS detectors? * Instead of unlabeled recommendations, the algorithm categorizes what you're seen, categorizes what it's recommending and let's the user say what category they want and remove or add categories for recommendation. * Have more ordinary search tool and a better categorization system (Youtube is a horrible black-box now, search is terrible, recommendations is terrible, I get videos from other sites mostly).
质疑质疑频道分类的主观性，CNN 与 TYT 不可相提并论
把 CNN 和 The Young Turks 都归类为"党派左翼"，就很难认真对待这个分析了。无论你怎么看这两个频道，它们的节目内容差异极大。而且这些标签是实验者手工创建、手工分配的，这让结果更可疑。
faizshah查看英文原文
Difficult to take this analysis seriously when CNN is categorized with The Young Turks as “Partisan Left.” Regardless of what you think of either channel the programming of these two channels is dramatically different. The fact that the tags were manually created by the experimenters and were manually assigned by the experimenters throws the results even more into question.
质疑问题定义有误：真正的风险是同类内容内的自我强化，而非跨类别的跳跃
我认为这项研究检验的是错误的假设。他们的设想是人们看了主流新闻后被推荐极端内容而"激进化"，比如福克斯新闻→9·11 阴谋论视频，并证明这并不常发生。但 YouTube 的真正问题是同类别内的推荐：如果某人看了一个阴谋论视频，其推荐就会全变成阴谋论视频，不可避免地让人消费越来越多极右内容。这看起来是扎实的好研究，只是不确定他们对问题的理解是否正确。
hooande查看英文原文
I think this study is testing the wrong hypothesis. Their idea is that people get "radicalized" by being recommended extremist content after watching mainstream news videos. ie, fox news -> 911 conspiracy videos. They demonstrate that this doesn't happen often. But the real problem wirh youtube is intracategory recommendations. If someone watches one conspiracy video, then their recommendations become all conspiracy videos which inevitably leads to them consuming more and more far right content. This seems like good science and well conducted research. I just don't know if they had the correct view of the problem
相关资源
从这篇论文所有 HN 讨论帖的全部评论里整理出来的链接。
历次讨论
- 2019Hacker News 讨论帖115 分 · 111 评
同领域还值得读
一切的回报率，1870–2015
The Rate of Return on Everything, 1870–2015
利用1870–2015年多国的长期数据，估算股票、住房等各类资本的回报率，并将其与经济增长率进行比较。
经济与社会科学2019 · 487 分 · 260 评进阶 · 约 100 分钟编辑精选一切资产的回报率，1870–2015
The Rate of Return on Everything, 1870–2015
构建了 1870–2015 年 16 个国家的长周期资产回报数据集，覆盖股票、住房、政府债券和短期票据，系统比较各类资产的真实回报与波动性。
经济与社会科学2018 · 215 分 · 151 评进阶 · 约 120 分钟编辑精选一切资产的回报率，1870–2015（2019）
The Rate of Return on Everything, 1870–2015 (2019)
这篇发表于《经济学季刊》的论文构建了 1870–2015 年间 16 个主要发达国家的股票、债券、房地产等资产收益率的长周期数据库。
经济与社会科学2024 · 190 分 · 148 评进阶 · 约 80 分钟编辑精选想法是否越来越难找？
Are ideas getting harder to find?
探讨“想法是否越来越难找到”：通过衡量多个领域的“研究生产率”，发现维持相同的增长需要投入指数级增长的研究资源，即研究效率在持续下降。
经济与社会科学2016 · 182 分 · 106 评入门 · 约 60 分钟编辑精选比特币与区块链的经济极限
The Economic Limits of Bitcoin and the Blockchain
芝加哥大学 Eric Budish 的论文从经济学角度分析比特币与区块链的根本限制：核心论点是只要攻击网络（双花或砸盘做空）的收益超过挖矿成本，51% 算力攻击就有利可图，而挖矿…
经济与社会科学2018 · 176 分 · 37 评进阶 · 约 90 分钟编辑精选