1
$\begingroup$

I have a set of points which is a count of links that users have clicked on : 

         link1,link2,link3,link4
user1,   0    ,2    ,3    ,2     
user2,   1    ,4    ,0    ,5

I am not sure which distance function I should use to calculate the distance between users. I think Euclidean distance is a good fit (reading http://en.wikipedia.org/wiki/Euclidean_distance) ?

Improve this question
$\endgroup$
1
  • 3
    $\begingroup$ What would be the reason for the distance calculation? That ought to play an important role in evaluating whether your choice of distance is useful or appropriate. $\endgroup$
    – whuber ♦
    Commented May 13, 2014 at 22:51

2 Answers 2

Reset to default
0
$\begingroup$

Mahalonobis distance could be better, because it takes into account the variability of number of clicks on the links. Say you have an unpopular link, then when someone clicks on it often, the weight of this link in the distance will be much higher, so you'll separate this user from others.

Improve this answer
$\endgroup$
0
$\begingroup$

I think cosine distance will be better in this situation and it gives you direction and is more robust than the usual distanc measures like euclidean/mahalonobis. You can also try manhattan distance and check how's the result.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.