This is from a ZenDesk (customer service issue tracker) search results page. It’s somewhat amusing that they call this ‘Sort by relevance’ because other than the search term the most important relevance factor is date of last activity, which I have to select every time I search. Thunderbird’s global search makes the same mistake.
You remember things that happened more recently better than older things. For searching “streams” (email, tweets, etc) you should default to the most recent items to take advantage of that.
And to assume you can derive relevance from a single word in a system like this is ridiculous. Google can do it because they have PageRank and millions of other searchers. In a small data set searched by a small amount of people, I don’t know what they could possibly be using for relevance other than ‘ORDER BY rand()’ (pseudocode)