36

I always glance through the "Questions that may already have your answer" section on the Ask Question page before posting since it frequently contains a post that already answers my question, and is a good way to double-check that I'm not posting a duplicate.

The results used to be so relevant to my question that before you updated your search engine, I used to use the Ask Questions page a lot to search for questions because it was was better at bringing up relevant posts than the actual search page.

But lately it's been useless to me. It appears that the top results are almost always highly voted posts that contain one of the words in the title, and the algorithm is not very good at picking out which words in the title are actually important and which are not.

For example, take a look at the difference in results between the "Questions that may already have your answer" search results and some actual SO search results when looking up something like "C# Convert List to 2D Array":

enter image description here enter image description here

This is only something I started noticing in the past few days, so I'm assuming you made a change to this recently which broke it.

I think it's too strongly weighted against how many votes a post has. For example, almost any question containing C, C++, or C# in the title brings up the C++ book question as one of the top "Questions that may contain your answer". And SO's top voted post about sorted vs unsorted array performance comes up in most of my tests, including a title of "test test test test test"

Can you fix this query so it goes back to returning relevant search results instead of mostly returning popular questions containing the same words?

Improve this question
6
  • 3
    site:stackoverflow.com is 820% better (made up figure, but it's better in my experience)
    – Kermit
    Apr 3, 2013 at 15:50
  • 9
    Obviously you should either switch language or read a book. How can that not be helpful. :p
    – Bart
    Apr 3, 2013 at 15:53
  • Even if using a dumb matching you would think that multiple matches should weigh rather heavily (n^2 maybe? stronger?). Perhaps some tuning of the weights of multiple-matches versus votes is in order.
    – dmckee --- ex-moderator kitten
    Apr 3, 2013 at 18:11
  • I guess it is only for some queries it is behaving like that, they must have improvised the algorithm used and is applicable for most of the cases barring a few corner cases.
    – AurA
    Apr 3, 2013 at 18:51
  • @dmckee, AurA, I think it's too strongly weighted against how many votes a post has. For example, almost any question with C, C++, or C# in the title brings up C++ Book post as one of the top "Questions that may contain your answer". While I'm sure one of those books may contain the answer to most C questions, most of us are asking on SO to avoid having to look through so many said books (or other online resources). And SO's top voted post about sorted vs unsorted array performance comes up in most of my tests, including "test test test test test"
    – Rachel
    Apr 3, 2013 at 19:18
  • I noticed this, too!
    – ErikE
    Apr 3, 2013 at 21:59

2 Answers 2

Reset to default
17

The boosting based on score (obviously an extreme with huge scores, we'll normalize that) was never intended for this feature. Starting with the next build, score boosting will not apply to title searches.

Improve this answer
3
  • Thanks Nick. Out of curiosity, in what cases does score boosting actually apply? Full-post searches?
    – Rachel
    Apr 4, 2013 at 11:49
  • @Rachel - normal searches as well as related question searches, I will work on normalizing really high scoring posts later this week, likely a log scale after a certain point.
    – Nick Craver
    Apr 4, 2013 at 12:57
  • 1
    That explains why some of the Related Question links haven't been very related lately :) Thank you again
    – Rachel
    Apr 4, 2013 at 12:59
1

I do not want to create a duplicate question / bug-report about the same problem. But even in 2018 the "Questions That May Have Your Answer" list does not show duplicate questions.

Both questions had the same keywords in title.

Rewarded with negative points of reputation.

Can this bug report be here (in the first unduplicated question/topic) as a answer? (I really do not want to create another duplicate question with a negative rating.)

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .