class EntityExtractor < Component::Service
        java_import "edu.stanford.nlp.ling.CoreAnnotations$TokensAnnotation"

        include_package "edu.stanford.nlp.pipeline"
        java_import "java.util.Properties"
        java_import "intoxicant.analytics.coreNlp.StopwordAnnotator"

        # Binds an RMQ consumer that receives jobs
        process do |msg|
          if msg["data"]["type"] == "html"
            page = Page.id_or_url(msg["id"], msg["url"])
            process msg, page if page
          end
        end

        @semaphore = Mutex.new

        def self.pipeline
          @semaphore.synchronize do
            @pipeline ||= begin
              props = Properties.new.tap do |p|
                p.set_property "annotators", "tokenize, ssplit, stopword, pos, lemma, ner"
                p.set_property "ner.applyNumericClassifiers", "false"
                p.set_property "ner.useSUTime", "false"
                p.set_property "parse.maxlen", "80" # See https://mailman.stanford.edu/pipermail/java-nlp-user/2014-November/006483.html
                p.set_property "customAnnotatorClass.stopword", "intoxicant.analytics.coreNlp.StopwordAnnotator"
              end
              StanfordCoreNLP.new props
            end
          end
        end

        def process(msg, page)
          entities = extract_entities annotate(msg["data"]["article"])
        end

        def annotate(article)
          Annotation.new(article).tap {|doc| self.class.pipeline.annotate(doc) }
        end

        def extract_entities(document)
          document.get(TokensAnnotation).map {|e| [e.original_text, e.ner] } # < -- undefined method `original_text' for #<Java::EduStanfordNlpLing::CoreLabel:0x412ce5ad>
          # ...
        end
      end