案例库 · 软件与 IT · 技术决策 · 1996–1998
这条还没译成中文,下面是英文原文。
Google ranked web pages by who linked to them, not what they said.
Brin and Page treated every hyperlink as a vote, then weighted votes by the voter's own rank, so keyword-stuffed pages sank.
那一手
In the mid-1990s search engines ranked pages mostly by matching query words against page text, so pages stuffed with keywords or hidden text rose to the top. Larry Page and Sergey Brin, then Stanford PhD students, built a prototype called BackRub to make the web's link structure do the ranking instead.
PageRank treated each hyperlink as a vote, then weighted votes by the rank of the page casting them, so a page linked by authoritative pages inherits authority. Their 1998 paper, The Anatomy of a Large-Scale Hypertextual Web Search Engine, presented the link-based metric as the heart of the engine and reported that it ran on a cluster of commodity machines.
The ranking rule was the mechanism, not an add-on: the paper described how crawling, indexing and ranking fit together so the link graph did the quality work. Google went live in 1998, and its results were visibly harder to spam than the keyword-matching engines of the time.
为什么管用
- Links are created by other people, so a page cannot easily inflate its own score.
- Weighting votes by rank made authority recursive and hard to game.
- Ranking reused data the crawler already collected, so it cost little extra.
- Keyword stuffing stopped working, which changed what search optimization had to be.
可以搬走什么
When everyone can game the signal you measure, use a signal they cannot fake in isolation: reputation propagated through the network instead of claimed by each actor.
后来呢
Google grew from a Stanford project into the default web search engine, and the link-graph idea was copied by every later engine. The ranking technology became the base of an advertising business that later auctioned the attention those rankings created.
资料来源
发现哪里写错了?告诉我们。