Feature Request: Tag's Closest Neighbors

[bouchard]

Thanks for the great work on this gem, it works wonderfully.

One feature I'm missing that'd I'd like to be considered for implementation is, for lack of a better term, a tag's "closest neighbors".

Given a tag A, what tags B, C, etc. are most commonly associated with a given object along with A?

This would allow for much more intuitive autocomplete on the client side, say if I often tag a user as "unprivileged, beginner, newbie", then:

User.first.nearest_tags("beginner") => [#<ActsAsTaggableOn::Tag id: 2, name: "unprivileged", associations: 6>, #<ActsAsTaggableOn::Tag id: 1, name: "newbie", associations: 4>]

Thoughts?

[bouchard]

I've coded this up - my SQL-fu isn't amazing, the best I could come up with uses two nested queries. Feel free to hack it apart, but it works :)

brady8@586078a3d2b62ec5fe976d2272ae5f291ef18a84

[bouchard]

A better version is up now, with support for arrays or strings of multiple tags (delimited using TagList.delimiter).

brady8@7012685626f66975c863bef0c8ca2c446ac2a9d5

[bouchard]

I've squashed the last two commits, and changed behaviour to include the tags being searched for in the results - my use case for this was with partial tag matching: If you start with 'card', and find all tags that are named_like, say 'cardio', and 'cardiology', then 'cardio' and 'cardiology' should be included in the neighboring_tags array.

Commit is here: brady8@ebbd7e25433b0ad4149127384d9ff0004e9becfc

[YavorIvanov]

Hey have you given up on this?

I can extract some "pseudo code" from our app. It has a tag_relation.rb model which I believe does what you want. Unfortunately it works with some hacking through generating a tag.rb which extends the ActsAsTaggableOn and its in a Rails 2.3.14 app so it uses an older monkey patched version of ActsAsTaggableOn. The result is pretty good though.

Let me know if I can help you with this.

[bouchard]

I should've posted this up in a Gist, as I managed to push to Github and lose my commits when working on an unrelated project.

All I had was a find_neighbouring_tags method added to tag.rb that would do a complicated join query to find the nearest neighbours, based on the frequency that any two tags where tagged on the same model instance together. Should be quite a bit simpler than what you're suggesting?

I'll need this again for a project soon, so I'll give a go a re-implementing it again, unless you'd like to show your work!

[YavorIvanov]

Have you tested the performance of you implementation? I believe we implemented our approach after stumbling on some performance issues. We have couple of tables including the taggings one that are in the millions.

[YavorIvanov]

Yep. Just tested an sql statement. The taggings table has 8M rows so its using a filesort at the end. This is going to be very costly. I'll try to come up with a simpler solution.

[bouchard]

Good points. Let me know if you come up with something, and/or I should hopefully be able to come up with something performant soon.

[YavorIvanov]

I tried decomposing it but no luck for now.

[artemk]

@YavorIvanov looking forward on this

[tilsammans]

It would be a nice feature, but I'm inclined to close it for now.

It seems like something that would be better handled by a full-text engine.

If anyone has thoughts about the implementation, I will gladly reopen it.

[trostli]

Best way to accomplish this IMO is to index the tags in soemthign like Elastic Search and do a similarity query. Check out the great gem Searchkick for this.