Searching for common words produces unexpected result


(Christoph) #1

Continuing the discussion from Read time metrics:

Looking at the posts that have been moved into the above topic, several of which contain the words “read” and “time”, I’m wondering why none of those posts showed up in my search results when I searched for those words…

I know that it is not a new insight that discourse search results can be improved, I just wanted to mention this as a particularly astonishing case. What might be the reason? (I just realize that we can’t really check the search results because the posts have been moved, but anyway.)


(Jeff Atwood) #2

Read and time are super common words. Try a less common word, or try “read time” in quotes to make it more specific. You can also try searching Google for a very common word like “them” to see what you get.

I’m not really seeing a problem though because the title was correctly prioritized in search results, based on @sam’s work in 2.0 (with the help of a community PR)


(Christoph) #3

Yes, title prioritization is good. All I’m saying is that I didn’t find any of those posts and was wondering if that might provide some hints for how the search results ranking could be tweaked. For example, I noticed (also in a number of previous searches) that the top results are often rather old posts. Nothing against old posts, but perhaps they are slightly overrated? Perhaps because they are likely to have more likes and links? In that case, perhaps age could be factored in just a little bit.

I dunno. I’m not a search algorithm expert.


(Jeff Atwood) #4

If you want to prioritize for latest add the shortcut letter “l” to your search or use the advanced search UI:

read time l

or

"read time" l

You can also select latest sort from the search result page at upper right like so:

image


(Christoph) #5

I suspected you would say that. But my point was that when you’re searching for the most relevant posts, there might be an (unjustified) bias towards older post.

In saying that, I’m assuming that likes (and possibly incoming links) are part of the formula for relevance search. Maybe that assumption is wrong?