1. Engagement Metrics
When a search engine delivers a page of results to you, it can measure the success of the rankings by observing how you engage with those results. If you click the first link, then immediately hit the back button to try the second link, this indicates that you were not satisfied with the first result. Search engines seek the “long click” – where users click a result without immediately returning to the search page to try again. Taken in aggregate over millions and millions of queries each day, the engines build up a good pool of data to judge the quality of their results.
2. Machine Learning
In 2011 Google introduced the Panda update to its ranking algorithm, significantly changing the way it judged websites for quality. Google started by using human evaluators to manually rate thousands of sites, searching for low quality content. Google then incorporated machine learning to mimic the human evaluators. Once its computers could accurately predict what the humans would judge a low quality site, the algorithm was introduced across millions of sites spanning the Internet. The end result was a seismic shift that rearranged over 20% of all of Google’s search results.
3. Linking Patterns
The engines discovered early on that the link structure of the web could serve as a proxy for votes and popularity; higher quality sites and information earned more links than their less useful, lower quality peers. Today, link analysis algorithms have advanced considerably, but these principles hold true.