FEATURE: Backfill posts sentiment.

app/jobs/scheduled/sentiment_backfill.rb

@@ -0,0 +1,30 @@
+# frozen_string_literal: true
+
+module Jobs
+  class SentimentBackfill < ::Jobs::Scheduled
+    every 5.minutes
+    cluster_concurrency 1
+
+    def execute(_args)
+      return if !SiteSetting.ai_sentiment_enabled
+
+      base_budget = SiteSetting.ai_sentiment_backfill_maximum_posts_per_hour
+      return if base_budget.zero?
+      # Split budget in 12 intervals, but make sure is at least one.
+      #
+      # This is not exact as we don't have a way of tracking how many
+      # posts we classified in the current hour, but it's a good enough approximation.
+      limit_per_job = [base_budget, 12].max / 12
+
+      classificator = DiscourseAi::Sentiment::PostClassification.new
+      return if !classificator.has_classifiers?
+
+      posts =
+        DiscourseAi::Sentiment::PostClassification.backfill_query(
+          max_age_days: SiteSetting.ai_sentiment_backfill_post_max_age_days,
+        ).limit(limit_per_job)
+
+      classificator.bulk_classify!(posts)
+    end
+  end
+end

config/settings.yml

@@ -17,6 +17,15 @@ discourse_ai:
   ai_sentiment_model_configs:
     default: ""
     json_schema: DiscourseAi::Sentiment::SentimentSiteSettingJsonSchema
+  ai_sentiment_backfill_maximum_posts_per_hour:
+    default: 250
+    min: 0
+    max: 10000
+    hidden: true
+  ai_sentiment_backfill_post_max_age_days:
+    default: 60
+    hidden: true
+
 
   ai_openai_dall_e_3_url: "https://api.openai.com/v1/images/generations"
   ai_openai_embeddings_url: "https://api.openai.com/v1/embeddings"

lib/post_extensions.rb

@@ -0,0 +1,16 @@
+# frozen_string_literal: true
+
+module DiscourseAi
+  module PostExtensions
+    extend ActiveSupport::Concern
+
+    prepended do
+      has_many :classification_results, as: :target
+
+      has_many :sentiment_classifications,
+               -> { where(classification_type: "sentiment") },
+               class_name: "ClassificationResult",
+               as: :target
+    end
+  end
+end

lib/sentiment/post_classification.rb

@@ -3,6 +3,47 @@
 module DiscourseAi
   module Sentiment
     class PostClassification
+      def self.backfill_query(from_post_id: nil, max_age_days: nil)
+        available_classifier_names =
+          DiscourseAi::Sentiment::SentimentSiteSettingJsonSchema
+            .values
+            .map { |mc| mc.model_name.downcase }
+            .sort
+
+        base_query =
+          Post
+            .includes(:sentiment_classifications)
+            .joins("INNER JOIN topics ON topics.id = posts.topic_id")
+            .where(post_type: Post.types[:regular])
+            .where.not(topics: { archetype: Archetype.private_message })
+            .where(posts: { deleted_at: nil })
+            .where(topics: { deleted_at: nil })
+            .joins(<<~SQL)
+              LEFT JOIN classification_results crs
+                ON crs.target_id = posts.id 
+                AND crs.target_type = 'Post'
+                AND crs.classification_type = 'sentiment'
+            SQL
+            .group("posts.id")
+            .having(<<~SQL, available_classifier_names)
+              COUNT(crs.model_used) = 0
+              OR array_agg(
+                  DISTINCT LOWER(crs.model_used) ORDER BY LOWER(crs.model_used)
+                )::text[] IS DISTINCT FROM array[?]
+            SQL
+
+        base_query = base_query.where("posts.id >= ?", from_post_id.to_i) if from_post_id.present?
+
+        if max_age_days.present?
+          base_query =
+            base_query.where(
+              "posts.created_at > current_date - INTERVAL '#{max_age_days.to_i} DAY'",
+            )
+        end
+
+        base_query
+      end
+
       def bulk_classify!(relation)
         http_pool_size = 100
         pool =
@@ -13,6 +54,7 @@ def bulk_classify!(relation)
           )
 
         available_classifiers = classifiers
+        return if available_classifiers.blank?
         base_url = Discourse.base_url
 
         promised_classifications =
@@ -25,9 +67,13 @@ def bulk_classify!(relation)
                 .fulfilled_future({ target: record, text: text }, pool)
                 .then_on(pool) do |w_text|
                   results = Concurrent::Hash.new
+                  already_classified = w_text[:target].sentiment_classifications.map(&:model_used)
+
+                  classifiers_for_target =
+                    available_classifiers.reject { |ac| already_classified.include?(ac.model_name) }
 
                   promised_target_results =
-                    available_classifiers.map do |c|
+                    classifiers_for_target.map do |c|
                       Concurrent::Promises.future_on(pool) do
                         results[c.model_name] = request_with(w_text[:text], c, base_url)
                       end
@@ -52,12 +98,17 @@ def bulk_classify!(relation)
 
       def classify!(target)
         return if target.blank?
+        return if classifiers.blank?
 
         to_classify = prepare_text(target)
         return if to_classify.blank?
 
+        already_classified = target.sentiment_classifications.map(&:model_used)
+        classifiers_for_target =
+          classifiers.reject { |ac| already_classified.include?(ac.model_name) }
+
         results =
-          classifiers.reduce({}) do |memo, model|
+          classifiers_for_target.reduce({}) do |memo, model|
             memo[model.model_name] = request_with(to_classify, model)
             memo
           end
@@ -69,6 +120,10 @@ def classifiers
         DiscourseAi::Sentiment::SentimentSiteSettingJsonSchema.values
       end
 
+      def has_classifiers?
+        classifiers.present?
+      end
+
       private
 
       def prepare_text(target)

lib/sentiment/sentiment_site_setting_json_schema.rb

@@ -27,6 +27,7 @@ def self.schema
       end
 
       def self.values
+        return {} if SiteSetting.ai_sentiment_model_configs.blank?
         JSON.parse(SiteSetting.ai_sentiment_model_configs, object_class: OpenStruct)
       end
     end

lib/tasks/modules/sentiment/backfill.rake

@@ -2,18 +2,8 @@
 
 desc "Backfill sentiment for all posts"
 task "ai:sentiment:backfill", [:start_post] => [:environment] do |_, args|
-  public_categories = Category.where(read_restricted: false).pluck(:id)
-
-  Post
-    .joins("INNER JOIN topics ON topics.id = posts.topic_id")
-    .joins(
-      "LEFT JOIN classification_results ON classification_results.target_id = posts.id AND classification_results.target_type = 'Post'",
-    )
-    .where("classification_results.target_id IS NULL")
-    .where("posts.id >= ?", args[:start_post].to_i || 0)
-    .where("category_id IN (?)", public_categories)
-    .where(posts: { deleted_at: nil })
-    .where(topics: { deleted_at: nil })
+  DiscourseAi::Sentiment::PostClassification
+    .backfill_query(from_post_id: args[:start_post].to_i)
     .find_in_batches do |batch|
       print "."
       DiscourseAi::Sentiment::PostClassification.new.bulk_classify!(batch)

plugin.rb

@@ -89,6 +89,7 @@ def self.public_asset_path(name)
   reloadable_patch do |plugin|
     Guardian.prepend DiscourseAi::GuardianExtensions
     Topic.prepend DiscourseAi::TopicExtensions
+    Post.prepend DiscourseAi::PostExtensions
   end
 
   register_modifier(:post_should_secure_uploads?) do |_, _, topic|

spec/jobs/scheduled/sentiment_backfill_spec.rb

@@ -0,0 +1,52 @@
+# frozen_string_literal: true
+
+require_relative "../../support/sentiment_inference_stubs"
+
+RSpec.describe Jobs::SentimentBackfill do
+  describe "#execute" do
+    fab!(:post)
+
+    before do
+      SiteSetting.ai_sentiment_enabled = true
+      SiteSetting.ai_sentiment_backfill_maximum_posts_per_hour = 100
+      SiteSetting.ai_sentiment_model_configs =
+        "[{\"model_name\":\"SamLowe/roberta-base-go_emotions\",\"endpoint\":\"http://samlowe-emotion.com\",\"api_key\":\"123\"},{\"model_name\":\"j-hartmann/emotion-english-distilroberta-base\",\"endpoint\":\"http://jhartmann-emotion.com\",\"api_key\":\"123\"},{\"model_name\":\"cardiffnlp/twitter-roberta-base-sentiment-latest\",\"endpoint\":\"http://cardiffnlp-sentiment.com\",\"api_key\":\"123\"}]"
+    end
+
+    let(:expected_analysis) { DiscourseAi::Sentiment::SentimentSiteSettingJsonSchema.values.length }
+
+    it "backfills when settings are correct" do
+      SentimentInferenceStubs.stub_classification(post)
+      subject.execute({})
+
+      expect(ClassificationResult.where(target: post).count).to eq(expected_analysis)
+    end
+
+    it "does nothing when batch size is zero" do
+      SiteSetting.ai_sentiment_backfill_maximum_posts_per_hour = 0
+
+      subject.execute({})
+
+      expect(ClassificationResult.count).to be_zero
+    end
+
+    it "does nothing when sentiment is disabled" do
+      SiteSetting.ai_sentiment_enabled = false
+
+      subject.execute({})
+
+      expect(ClassificationResult.count).to be_zero
+    end
+
+    it "respects the ai_sentiment_backfill_post_max_age_days setting" do
+      SentimentInferenceStubs.stub_classification(post)
+      SiteSetting.ai_sentiment_backfill_post_max_age_days = 80
+      post_2 = Fabricate(:post, created_at: 81.days.ago)
+
+      subject.execute({})
+
+      expect(ClassificationResult.where(target: post).count).to eq(expected_analysis)
+      expect(ClassificationResult.where(target: post_2).count).to be_zero
+    end
+  end
+end