Content-based filtering with elasticsearch

Asked May 19 '16 at 12:32

Active May 20 '16 at 07:30

Viewed 338 times

I am trying to build a content-based filtering system that classifies products by assigning them features that read like { salty: 0, sweet: 0.5, bitter: 0.7}. Which means I give every product a n-dimensional vector, in this case [0, 0.5, 0.7].

For a given product I would now like to find "similar" products by calculating the distance between the vectors. So for two products [0.2, 0.2, 0.8] and [0.4, 0.9, 0.9] the euclidean distance is roughly 0.78, which should be their 'score' (lower is better).

How do I do this with elasticsearch? Is elasticsearch the right tool for such a task?

Note that the real problem has a lot more than 3 dimensions.

edited May 20 '16 at 07:30

asked May 19 '16 at 12:32

Thorben Croisé

12,407
8
39
50

1

This answer could help you: http://stackoverflow.com/a/24951668/2483542 – TroyAndAbed May 19 '16 at 12:51
And how do you calculate this `distance`? Is it something that can be indexed along with the vector itself? – Andrei Stefan May 19 '16 at 14:21
I was thinking about plain and simple euclidean distance to find out how close two products are. – Thorben Croisé May 20 '16 at 06:10
I had one more question in my previous comment. Can you answer that one as well? – Andrei Stefan May 20 '16 at 07:20
Maybe I don't get it ... how would you index a distance from any query point? – Thorben Croisé May 20 '16 at 07:23
I edited the question to make it clearer what I want. – Thorben Croisé May 20 '16 at 07:32
Now I got it. I was thinking at scripting for this. And there is a solution with scripting in the comment from @TroyAndAbed. – Andrei Stefan May 20 '16 at 08:01
Let us [continue this discussion in chat](http://chat.stackoverflow.com/rooms/112482/discussion-between-thorben-bochenek-and-andrei-stefan). – Thorben Croisé May 20 '16 at 08:06
Ever found a solution to this problem? – Simon Steinberger Dec 01 '17 at 22:24

Content-based filtering with elasticsearch

0 Answers0