compares performance of parsing word-coordinate data in different JSON source formats

word-coord-json-parse-test.md

Overview

For each of these JSON word-coordinate file formats of the same OCR word-coordinate data, representing a single newspaper page:

1. Open-ONI (58.9 kB)
2. IIIF Annotation List (826.4 kB)

I ran a test in the Rails console, benchmarking the time needed to:

1. load the source file into memory
2. parse as JSON
3. find nodes matching a search term ("October")
4. format the coordinate data as a hash matching an oa:Annotation
5. return an array of hashes that could be used for an IIIF Content Search response

1. Open-ONI

Benchmark.ms do
  file = File.read("open-oni_OCR-word-coordinates.json")
  ocr = JSON.parse(file)
  term = "October"
  term_for_match = term.downcase
  matches = ocr["coords"].select { |k,v| k.downcase =~ /#{term_for_match}/ }
  resources = []
  matches.each do |k,v|
    v.each do |coords_array|
      resource_hash = {}
      resource_hash["@id"] = "http://example.org/identifier/annotation/anno-line"
      resource_hash["@type"] = "oa:Annotation"
      resource_hash["motivation"] = "sc:painting"
      resource_hash["resource"] = { "@type" => "cnt:ContentAsText", "chars" => k}
      resource_hash["on"] = "http://example.org/identifier/canvas1#xywh=#{coords_array.join(',')}"
      resources << resource_hash
    end
  end
  resources
end

Total time (ms): 3.73

2. IIIF Annotation List

Benchmark.ms do
  file = File.read("open-oni_OCR-word-coordinates_iiif-anno-list.json")
  ocr = JSON.parse(file)
  term = "October"
  term_for_match = term.downcase
  matches = ocr["resources"].select { |anno| anno["resource"]["chars"].downcase =~ /#{term_for_match}/ }
  matches.each do |resource|
    resource["@id"] = "http://example.org/identifier/annotation/anno-line"
  end
end

Total time (ms): 10.08

Conclusion

I dug a littler deeper into the time needed for each part of the processing, and the biggest differences are in loading/parsing the file (Open-ONI: ~1.72ms; IIIF: ~6.33ms) and searching for terms (Open-ONI: ~1.76ms; IIIF: ~4.73ms).

It's worth noting that the source IIIF file (created using NLW's ALTO-to-IIIF XSLT) was missing @id values for each resource in the annotation list. The addition of these values would make the file a bit larger in size, and probably increase the time needed to load/parse the file.

Overall, while Open-ONI is 2.7x faster, the total difference of 6.35ms is not especially alarming.