Say, you are tagging car pictures on a website, and have the following:

image1.jpg -> ["honda","s2000","convertible","black"]
image2.jpg -> ["honda","civic","blue"]
image3.jpg -> ["lexus","is300","blue"]
image4.jpg -> ["s2000","honda","convertible","silver"]
image5.jpg -> ["toyota","starlet","black"]

Seeing this, you’d know that image1.jpg and image4.jpg are similar pictures. Or rather “more similar” than , say, image1.jpg and image3.jpg. For this, I wrote below snippet of code. This goes in the model file, and can be called as “object.similar”. It returns an array of similar “things”, sorted on most similar to less similar (hence the results.reverse at the end)

For example:

  img = Image.find(params[:id])
  @similar_images = img.similar[0..10]

Will give you the 10 “most similar” images as img. Well, it gives you the files with the most similar tags.

def similar
  tags = self.tags
  results = []
  tags.each do |tag|
    results = results + tag.pictures # or tag.things, tag.products, ...
    results.delete(self)
  end

  # make array into hash
  h = Hash.new
  results.each do |r|
    h[r] = h[r].to_i + 1
  end

  # sort on values
  tmp = h.sort {|a,b| a[1]<=>b[1]}
  results = []
  tmp.each do |t|
    results << t[0]
  end

  results.reverse # return all items, products, ...
end

This was written for a new project coming up, and will be used to do better "similarities matching" for places.ae, though for the latter we also had to sort on distance. (For it's vicinity)