Language detection with CLD3 (forem#19756)

* Language detection with CLD3 POC

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update app/models/article.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Clean up tests

* Clean up tests

* Move language detection to service

* rubocop

* Update app/services/languages/detection.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* More linting cleanup

* Remove allow_any_instance_of

* Update tests to better exercise different paths

* Fix flaky spec

* Update spec/services/languages/detection_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/models/article_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/services/languages/detection_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/services/languages/detection_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/services/languages/detection_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Update spec/services/languages/detection_spec.rb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Gemfile

@@ -21,6 +21,7 @@ gem "blazer", "~> 2.6" # Allows admins to query data
 gem "bootsnap", ">= 1.1.0", require: false # Boot large ruby/rails apps faster
 gem "carrierwave", "~> 2.2" # Upload files in your Ruby applications, map them to a range of ORMs, store them on different backends
 gem "carrierwave-bombshelter", "~> 0.2" # Protect your carrierwave from image bombs
+gem "cld3", "~> 3.5" # Ruby interface for Compact Language Detector v3
 gem "cloudinary", "~> 1.23" # Client library for easily using the Cloudinary service
 gem "counter_culture", "~> 3.2" # counter_culture provides turbo-charged counter caches that are kept up-to-date
 gem "ddtrace", "~> 1.3.0" # ddtrace is Datadog’s tracing client for Ruby.

Gemfile.lock

@@ -158,6 +158,7 @@ GEM
       fastimage
     cgi (0.3.6)
     chartkick (4.2.1)
+    cld3 (3.5.3)
     cloudinary (1.26.0)
       aws_cf_signer
       rest-client (>= 2.0.0)
@@ -982,6 +983,7 @@ DEPENDENCIES
   carrierwave (~> 2.2)
   carrierwave-bombshelter (~> 0.2)
   cgi (~> 0.3.6)
+  cld3 (~> 3.5)
   cloudinary (~> 1.23)
   counter_culture (~> 3.2)
   cuprite (~> 0.13)

app/models/article.rb

@@ -198,6 +198,7 @@ def self.unique_url_error
   before_save :calculate_base_scores
   before_save :fetch_video_duration
   before_save :set_caches
+  before_save :detect_language
   before_create :create_password
   before_destroy :before_destroy_actions, prepend: true
 
@@ -625,6 +626,12 @@ def collection_cleanup
     collection.destroy
   end
 
+  def detect_language
+    return unless title_changed? || body_markdown_changed?
+
+    self.language = Languages::Detection.call("#{title}. #{body_text}")
+  end
+
   def search_score
     comments_score = (comments_count * 3).to_i
     partial_score = (comments_score + (public_reactions_count.to_i * 300 * user.reputation_modifier * score.to_i))

app/services/languages/detection.rb

@@ -0,0 +1,22 @@
+module Languages
+  class Detection
+    attr_reader :text
+
+    PROBABILITY_THRESHOLD = 0.5
+
+    def self.call(...)
+      new(...).call
+    end
+
+    def initialize(text)
+      @text = text
+    end
+
+    def call(identifier: CLD3::NNetLanguageIdentifier.new(0, 1000))
+      language_outcome = identifier.find_language(text)
+      return unless language_outcome.probability > PROBABILITY_THRESHOLD && language_outcome.reliable?
+
+      language_outcome.language
+    end
+  end
+end

db/migrate/20230712195950_add_human_language_to_articles.rb

@@ -0,0 +1,5 @@
+class AddHumanLanguageToArticles < ActiveRecord::Migration[7.0]
+  def change
+    add_column :articles, :language, :string
+  end
+end

db/migrate/20230713150940_add_language_index_to_articles.rb

@@ -0,0 +1,6 @@
+class AddLanguageIndexToArticles < ActiveRecord::Migration[7.0]
+  disable_ddl_transaction!
+  def change
+    add_index :articles, :language, algorithm: :concurrently
+  end
+end

db/schema.rb

@@ -10,7 +10,7 @@
 #
 # It's strongly recommended that you check this file into your version control system.
 
-ActiveRecord::Schema[7.0].define(version: 2023_06_27_154435) do
+ActiveRecord::Schema[7.0].define(version: 2023_07_13_150940) do
   # These are extensions that must be enabled in order to support this database
   enable_extension "citext"
   enable_extension "pg_trgm"
@@ -105,6 +105,7 @@
     t.boolean "featured", default: false
     t.string "feed_source_url"
     t.integer "hotness_score", default: 0
+    t.string "language"
     t.datetime "last_comment_at", precision: nil, default: "2017-01-01 05:00:00"
     t.datetime "last_experience_level_rating_at", precision: nil
     t.string "main_image"
@@ -160,6 +161,7 @@
     t.index ["feed_source_url"], name: "index_articles_on_feed_source_url_unscoped"
     t.index ["hotness_score", "comments_count"], name: "index_articles_on_hotness_score_and_comments_count"
     t.index ["hotness_score"], name: "index_articles_on_hotness_score"
+    t.index ["language"], name: "index_articles_on_language"
     t.index ["path"], name: "index_articles_on_path"
     t.index ["public_reactions_count"], name: "index_articles_on_public_reactions_count", order: :desc
     t.index ["published"], name: "index_articles_on_published"

spec/models/article_spec.rb

@@ -1420,4 +1420,27 @@ def foo():
       end
     end
   end
+  describe "#detect_language" do
+    let(:detected_language) { :kl } # kl for Klingon
+
+    before do
+      allow(Languages::Detection).to receive(:call).and_return(detected_language)
+    end
+
+    it "detects language using title and body for newly created articles" do
+      article = create(:article)
+      expect(Languages::Detection).to have_received(:call).with("#{article.title}. #{article.body_text}")
+    end
+
+    it "detects language using title and body for updated articles" do
+      article.update(body_markdown: "---title: This is a new english article\n---\n\n# Hello World")
+      expect(Languages::Detection).to have_received(:call).with("#{article.title}. #{article.body_text}")
+    end
+
+    it "does not call detection when title and body_markdown are unchanged" do
+      article.language = "es"
+      article.update(nth_published_by_author: 5)
+      expect(Languages::Detection).not_to have_received(:call)
+    end
+  end
 end

spec/services/languages/detection_spec.rb

@@ -0,0 +1,76 @@
+require "rails_helper"
+
+RSpec.describe Languages::Detection, type: :service do
+  subject(:language_detection) { described_class.call(text) }
+
+  context "when the text is clearly identifiable as English" do
+    let(:text) { "This is clearly English text." }
+
+    it "returns en" do
+      expect(language_detection).to eq(:en)
+    end
+  end
+
+  context "when the text is clearly identifiable as Spanish" do
+    let(:text) { "Esto es claramente un texto en español." }
+
+    it "returns es" do
+      expect(language_detection).to eq(:es)
+    end
+  end
+
+  context "when probability and reliability vary" do
+    let(:text) { "This is some dummy text." }
+    let(:identifier) { instance_double(CLD3::NNetLanguageIdentifier) }
+
+    before do
+      allow(CLD3::NNetLanguageIdentifier).to receive(:new).and_return(identifier)
+      allow(identifier).to receive(:find_language).with(text).and_return(language_outcome)
+    end
+
+    context "when probability is low" do
+      let(:language_outcome) do
+        instance_double(
+          CLD3::NNetLanguageIdentifier::Result,
+          language: :es,
+          probability: 0.4,
+          reliable?: true
+        )
+      end
+
+      it "returns nil" do
+        expect(described_class.call(text)).to eq(nil)
+      end
+    end
+
+    context "when reliability is low" do
+      let(:language_outcome) do
+        instance_double(
+          'CLD3::NNetLanguageIdentifier::Result',
+          language: :es,
+          probability: 0.9,
+          reliable?: false
+        )
+      end
+
+      it "returns nil" do
+        expect(described_class.call(text)).to be(nil)
+      end
+    end
+
+    context "when probability and reliability are high" do
+      let(:language_outcome) do
+        instance_double(
+          CLD3::NNetLanguageIdentifier::Result,
+          language: :es,
+          probability: 0.9,
+          reliable?: true,
+        )
+      end
+
+      it "returns es" do
+        expect(described_class.call(text)).to eq(:es)
+      end
+    end
+  end
+end