网络与分布式2024downloads.webis.de编辑精选
Google 变差了吗？对搜索引擎中 SEO 垃圾的纵向调查
Is Google Getting Worse? A Longitudinal Investigation of SEO Spam in Search
入门 · 约 45 分钟读完前置知识：搜索引擎工作原理、SEO 基本概念
导读
这篇论文对搜索引擎结果中的 SEO 垃圾内容做了纵向追踪：作者团队在一段时间内持续采集 Google、Bing、DuckDuckGo 等引擎针对商品评测类查询的结果，用分类器识别联盟营销/SEO 垃圾页面，观察其占比与排名的变化。结论指向 Google 的结果质量随时间下滑，垃圾内容不仅能进入结果页，还会逐步爬升到更靠前的位置。
HN 关注点：HN 读者长期抱怨搜索结果被内容农场和 AI 生成内容淹没，这篇论文给这种体感提供了量化证据，讨论迅速转向个人遭遇与替代方案。
HN 评论精选
这几条评论里：支持 2 补充 2
支持从内容农场、图片搜索到丢词和时效性偏好，逐一列举 Google 结果质量下滑的具体表现。
我已经看不懂 Google 了。大家应该都遇到过：带着一个技术问题点进某篇文章，文章把你的问题重复七遍，塞满看似相关却始终不回答问题的废话，最后来一句“试试重新插拔”。这明显是恶意内容农场，Google 有那么多技术却似乎检测不出来或不愿检测。技术做不到的话，搞人工策展或反馈机制行不行？图片搜索也一样，搜“泰国的红花”却出现各种颜色、各种产地的花。大多数人只输入 2-3 个词，现在 Google 却总是把最后一个词丢掉，好像它比我更清楚我要什么。精心撰写的高质量文章排不上去，过时的文章反而排在前面，你只能看到好内容就手动收藏，因为再也搜不回来了。Google 的人都睡着了吗？这不是小事，这是你的饭碗，青少年已经用 TikTok 搜索了。
iteratethis查看英文原文
I don't understand Google anymore. Surely people can relate to the situation where you end up on an article based on some technical query you have. The article repeats your question 7 times, has endless casually-related filler text that still does not answer the question and then ends with: try to unplug it. It is so freaking obvious that it's a malicious content farm, but Google with all of its technical might seem unable or unwilling to detect it. If tech can't do it, organize some type of curation or feedback? Same for image search. You search for "red flower Thailand" and flowers of various other colors from various locations appear. The idea that Google is spectacularly good at subject detection from imagery does not seem to actually work out in practice. Most people's search queries consist of just 2-3 words. Nowadays Google consistently just drops the last word as if it knows better than I do what I need. High value elaborate articles on various topics do not rank. Instead, dated articles do. You have to manually bookmark high quality content as you see it, because you'll never find it back via search. Is everybody asleep at Google? This is not a small thing, this is your bread and butter. Teens are using Tiktok for search, you're in real trouble and better start cleaning up your act.
补充图片搜索被 AI 仿作污染，且“偏好新鲜度”的排序逻辑在历史题材和已故艺术家上适得其反。
Google 图片正被 AI 垃圾淹没，某些用途下已不可用，甚至没有做 SEO 的内容也被推到前面。有时连普通搜索首屏展示的关键图片都是假的——比如搜“tank man”时，优先展示了一张伪造的、从当事人视角拍摄的 AI 自拍。那张图现在还在，只是不再是第一条，可能是舆论反弹后 Google 手动降权了。想浏览某位艺术家的作品？祝你好运。Google 显然分不清真正的 J. C. Leyendecker 作品和“Leyendecker 风格”生成模型吐出来的东西。搜“Yoji Shinkawa”，首图（链接）根本不像他的风格，Google 却认定它最能代表他；完整图片页里他的真作和 AI 仿作大约各占一半。我猜 Google 优先展示“最新”结果，但对天安门这样的历史事件或 1951 年去世的艺术家来说，某个时间点之后的新结果几乎全是 AI 伪造品。
jsheard查看英文原文
Google Images is filling up with AI slop to the point of unusability for some purposes, even content that isn't actively engaging in SEO shenanigans is getting pushed to the top. In some cases even the key image that Google displays first on a normal search is fake, like when searching for the "tank man" prioritized showing a fake AI selfie from the mans perspective. It's still there in fact, just not the first result anymore, but Google might have demoted it manually after the backlash. Looking to browse works by a specific artist? Good luck, Google evidently can't tell the difference between a genuine J. C. Leyendecker piece and anything shat out by a "Leyendecker style" image generation model. Search for "Yoji Shinkawa" and this https://i.imgur.com/RYghaoY.png is currently the first thing you see, which isn't even close to his style, but Google has somehow determined it's the image that best represents him. The full images page shows his actual work but interspersed about half-and-half with AI imitations. My speculation is that Google prioritizes showing recent results, presumed to be the freshest most up-to-date information, but of course for a historical event like Tiananmen Square or an artist who died in 1951, nearly all of the fresh results past a certain point are AI simulacra.
支持学习类查询被带货内容和标题党视频占据，促使他回归读书。
Google、YouTube 等等都在变差，我越来越难从互联网上学到东西。无论是学吉他、钓鱼、健身还是几乎任何事，得到的都是垃圾结果。Google 结果全是浅薄内容，大多想卖我点什么；YouTube 则充斥着标题党的短视频，和我输入的完全无关：“别犯这个 {topic} 错误！”“这一招让你 {topic} 更厉害！”“两周成为 {topic} 高手的十个方法！”。这些算法不断投喂垃圾，我认为它在让社会变蠢，所以我自然转向了书。书的内容质量高出一大截，更详细、更透彻，体验也更丰富。
foofoo4u查看英文原文
Google, YouTube, etc., they're all getting worse. I'm finding it ever more difficult to learn anything anymore from the internet. Whether it's learning how to play the guitar, how to fish, weight lifting or how to do mostly anything, I get trash results. Google results are exclusively shallow content, most of which are trying to sell me on something. YouTube is full of short videos with click baity subjects entirely irrelevant to what I actually entered: "Don't make this {topic} mistake!", "This one trick will make you better at {topic}!", "Top 10 ways to become a pro at {topic} in two weeks!". Due to the constant trash these algorithms insist on feeding me, which I believe is actively contributing to the dumbing-down of society, I've naturally gravitated towards books. Books are leagues better in terms of quality of content. They are more detailed and thorough. It's a richer experience so far.
补充补充一个可自定义域名权重的替代搜索引擎 Kagi。
Kagi 允许用户调整特定域名的排名权重（链接），我不确定这个设置是否在所有用户间共享，但这肯定会有帮助。
barbazoo查看英文原文
Kagi allows users to adjust the ranking [0] of certain domains. I'm not sure if that's shared among all users but it would certainly help. [0] https://blog.kagi.com/kagi-features
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
- 2024Hacker News 讨论帖276 分 · 274 评
同领域还值得读
Paxos vs. Raft：分布式共识是否已达成共识？
Paxos vs. Raft: Have we reached consensus on distributed consensus?
重新审视了 Paxos 与 Raft 两种分布式共识算法之间的差异，试图在一个统一的框架下比较它们，并论证二者的相似程度远高于通常的认知，许多所谓的差别只是表述层面的。
网络与分布式2020 · 343 分 · 65 评 · 引用 10进阶 · 约 28 分钟编辑精选Stellar 协议：互联网级共识的联邦模型
Stellar Protocol: A Federated Model for Internet-Level Consensus (2016)
提出 Stellar Consensus Protocol（SCP），一种面向互联网级共识的联邦拜占庭协议模型。
网络与分布式2018 · 301 分 · 61 评专业 · 约 90 分钟编辑精选拜占庭将军问题（1982）
The Byzantine Generals Problem (1982)
这篇 1982 年论文提出并形式化了拜占庭将军问题：在分布式系统中，部分节点可能叛变或故障、通信不可完全信任时，忠诚节点如何达成一致行动。
网络与分布式2018 · 290 分 · 77 评进阶 · 约 75 分钟编辑精选使 CRDT 能够容忍拜占庭故障
Making CRDTs Byzantine Fault Tolerant
Martin Kleppmann 等人的论文研究如何让 CRDT（无冲突复制数据类型）抵御拜占庭故障节点，即部分参与者恶意篡改数据时仍能保证一致性。
网络与分布式2022 · 266 分 · 40 评进阶 · 约 90 分钟编辑精选Filecoin：去中心化存储网络
Filecoin: A Decentralized Storage Network
Filecoin 白皮书介绍了一个去中心化存储网络：用户付费让矿工存储数据，矿工通过复制证明和时空证明来证明自己确实持续存储了数据，并以此获得代币奖励。
网络与分布式2017 · 196 分 · 162 评入门 · 约 45 分钟编辑精选