5
$\begingroup$

I've written a simple recommender which generates recommendations for users based on what they have clicked. The recommender generates a data file with the following format:

userid,userid,simmilarity (between 0 and 1 - closer to 0 the more similar the users)
a,b,.2
a,c,.3
a,d,.4
a,e,.1
e,b,.3
e,c,.5
e,d,.8

I've looked at some graphs, but I'm not sure which one to use, or if are there other ones that will better display the user similarities from the dataset above. Any suggestions?

I'm aiming this visualization at business users who are not at all technical. I would just like to show them an easy to understand visual that details how similar some users are and so convince the business that for these users the recommendation system is useful.

@Steve Kallestad do you mean something like this :

enter image description here

$\endgroup$
5
$\begingroup$

I think you're looking for a similarity matrix (see bottom of the page). If you don't have data on similarity between certain pairs, you can always leave them as grey or white. Also, this will only work for data sets small enough to actually make out what's going on. I'd say 25 rows / columns maximum.

In a similarity matrix, x, and y coordinates correspond to the two things you're comparing, while a colormap magnitude represents similarity

EDIT: One thing you could do to replace the colormap is the insert, say, circles of different sizes according to the similarity metric. Or you could insert the numbers themselves, again, varying the size of the number as the magnitude of that number varies. Size usually works best is business visualizations.

| improve this answer | |
$\endgroup$
  • $\begingroup$ thanks but I'm aiming this at non technical business users, ive updated question. I think a colormap would go over their heads $\endgroup$ – blue-sky Jun 24 '14 at 20:34
  • $\begingroup$ @blue-sky Updated my answer. May not be what you're looking for, but that's all I've got =) $\endgroup$ – Phonon Jun 24 '14 at 20:37
2
$\begingroup$

Personally, I think Netflix got it right. Break it down into a confidence rating from 1-5 and show your recommendations based on the number of yellow stars.

It doesn't have to be stars, but those icon based graphs are very easy to interpret and get the point across clearly.

| improve this answer | |
$\endgroup$
  • $\begingroup$ please see question update, so for my use case instead of a movie display a user (that is similar) , the stars indicate the degree of similarity ? $\endgroup$ – blue-sky Jun 25 '14 at 8:01
  • $\begingroup$ Exactly - instead of a movie poster, you can display a user profile picture. So several rows - Users similar to A, Users similar to B, etc. and across each row, user profile pictures. Below each user profile picture a star rating. You indicate that the similarity is stronger the closer to 0, so .2-.0 = 5 stars, .4-.2 = 4 stars, etc. If there are no profile pictures available, a user name would do, but people will respond to photos better. $\endgroup$ – Steve Kallestad Jun 25 '14 at 9:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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