Merge pull request #98 from identity-research-lab/fix-bad-relation

Fix an issue with categories creating invalid relations with codes

README.md

@@ -101,6 +101,14 @@ To create your local `.env` file, make a copy of `.env.example`. (This filename
 
 Now, when you open the `.env` file in your text editor, you'll see a list of key/value pairs that need filling in. These are configuration options including database credentials, customizations, and the API keys that TMI-Web needs to communicate with third-party services. Follow the instructions provided in the file to register for the appropriate API keys.
 
+### Run the test suite
+
+    rspec --format documentation spec/
+
+To always run rspec with the documentation flag:
+
+    echo "--format documentation" > .rspec
+
 ### Start the background job runner
 
 From the root directory of tmi-web, launch Sidekiq by typing:
@@ -135,14 +143,6 @@ To update the Neo4j database, type:
 
 ## Developer tips
 
-### Run the test suite
-
-    rspec --format documentation spec/
-
-To always run rspec with the documentation flag:
-
-    echo "--format documentation" > .rspec
-
 ### Clear the Sidekiq (background job) queue
 
 Launch the local interactive Rails console:

app/models/category.rb

@@ -15,36 +15,23 @@ class Category
 
   has_many :out, :codes, rel_class: :CategorizedAs, dependent: :delete_orphans
 
-  # This is the prompt sent to the selected AI agent to provide instructions on category derivision.
-  PROMPT = %{
-    You are a social researcher doing data analysis. Please generate a list of the 20 most relevant themes from the following list of codes. The themes should be all lowercase and contain no punctuation. Codes should be stripped of quotation marks. Return each code with an array of its categories in JSON format. Use this JSON as the format:
-
-    {
-      "themes" : [
-        {
-          "theme": "foo",
-          "codes": [ "bar", "bat", "baz"]
-        }
-      ]
-    }
-
-    The codes are as follows:
-  }
-
   # Regenerates Category objects based on codes within a given context.
   # This method uses the Clients::OpenAi client passing the codes as an argument to the prompt.
   # The agent returns an array of themes, which are then captured as Category objects.
   def self.from(context)
     codes = Code.where(context: context)
-    response = Clients::OpenAi.request("#{PROMPT} #{codes.map(&:name).join(',')}")
-    return unless response['themes']
+    return unless codes.any?
+
+    text = codes.map(&:name).join(',')
+    return unless text.present?
+    return unless themes = DeriveThemes.perform(text)
 
     Category.where(context: context).destroy_all
 
-    response['themes'].each do |record|
-      category = Category.find_or_create_by(name: record['theme'], context: context)
+    themes.each do |theme|
+      category = Category.find_or_create_by(name: theme['theme'].strip.downcase, context: context)
       codes.each do |code|
-        next unless record['codes'].include?(code.name)
+        next unless theme['codes'].include?(code.name)
         CategorizedAs.create(from_node: category, to_node: code)
       end
     end

app/models/code.rb

@@ -13,7 +13,7 @@ class Code
 
   validates :name, presence: true
   validates :context, presence: true
-  validates_uniqueness_of :name, :scope => :context
+  validates_uniqueness_of :name, scope: :context
 
   has_many :out, :personas, rel_class: :Experiences
   has_many :in, :categories, rel_class: :CategorizedAs

app/services/derive_themes.rb

@@ -0,0 +1,39 @@
+class DeriveThemes
+
+	attr_accessor :text
+
+	# This is the prompt sent to the selected AI agent to provide instructions on category derivision.
+	PROMPT = %{
+		You are a social researcher doing data analysis. Please generate a list of the 20 most relevant themes from the following list of codes. The themes should be all lowercase and contain no punctuation. Codes should be stripped of quotation marks. Return each code with an array of its categories in JSON format. Use this JSON as the format:
+	
+		{
+			"themes" : [
+				{
+					"theme": "foo",
+					"codes": [ "bar", "bat", "baz"]
+				}
+			]
+		}
+	
+		The codes are as follows:
+	}
+
+	def self.perform(text)
+		new(text).perform
+	end
+
+	def initialize(text)
+		@text = text
+	end
+
+	# Uses the OpenAI client to pass the prompt and text through the API for sentiment analysis.
+	def perform
+		return false unless text.present?
+
+		response = Clients::OpenAi.request("#{PROMPT} #{self.text}")
+		return false unless response['themes'].present?
+		return response['themes']
+ 	end
+
+
+end

app/services/export_to_graph.rb

@@ -23,7 +23,7 @@ def perform
 	# Hydrates the associated Persona with data from the SurveyResponse.
 	# Note that this operation is destructive to a Persona that already exists.
 	def persona
-		@persona ||= Persona.create(
+		@persona ||= Persona.find_or_create_by(
 			name: "Persona #{survey_response.identifier}",
 			survey_response_id: survey_response.id,
 			permalink: survey_response.permalink
@@ -46,9 +46,11 @@ def populate_experience_codes
 			"notes" => survey_response.notes_codes
 		}
 		contexts_and_codes.each do |context, codes|
-			codes.each do |name|
-				code = Code.find_or_create_by(name: name, context: context)
-				Experiences.create(from_node: persona, to_node: code)
+			codes.compact.uniq.each do |name|
+				if code = Code.find_or_create_by(name: name, context: context)
+					next unless code.valid?
+					Experiences.create(from_node: persona, to_node: code)
+				end
 			end
 		end
 
@@ -67,9 +69,11 @@ def populate_id_codes
 			"pronouns" => survey_response.pronouns_id_codes
 		}
 		contexts_and_codes.each do |context, codes|
-			codes.each do |name|
-				identity = Identity.find_or_create_by(name: name, context: context)
-				IdentifiesWith.create(from_node: persona, to_node: identity)
+			codes.compact.uniq.each do |name|
+				if identity = Identity.find_or_create_by(name: name.strip, context: context)
+					next unless identity.valid?
+					IdentifiesWith.create(from_node: persona, to_node: identity)
+				end
 			end
 		end
 	end

lib/clients/open_ai.rb

@@ -1,11 +1,7 @@
-# A wrapper class for the OpenAI API.
+# A wrapper class for the OpenAI API client.
 class Clients::OpenAi
 
   # Sends a prompt to the configured model and returns the relevant portion of the response.
-  #
-  # @param prompt [String] the text of the prompt to send to the model.
-  # @return [String] the JSON response returned by the API.
-
   def self.request(prompt)
     client = OpenAI::Client.new
     response = client.chat(

spec/services/export_to_graph_spec.rb

@@ -7,12 +7,17 @@
 		allow(Persona).to receive(:find_or_initialize_by).and_return(persona)
 		allow(Persona).to receive(:create).and_return(persona)
 		allow(persona).to receive(:destroy)
+		allow(IdentifiesWith).to receive(:create)
 	end
 
+	let(:service) 	{ ExportToGraph.new(1) }
+
+	let(:code) 			{ Code.new(name: "not okay", context: "age") }
+	let(:identity) 	{ Identity.new(name: "genx", context: "age") }
+
 	let(:survey_response) {
 		SurveyResponse.new(
 			id: 1,
-			response_id: 1,
 			age_exp_codes: ["not okay"],
 			age_id_codes: ["genx"]
 		)
@@ -23,35 +28,11 @@
 			survey_response_id: 1
 		)
 	}
-	let(:service) {
-		ExportToGraph.new(1)
-	}
-
-	it 'assigns its SurveyResponse' do
-		expect(service.survey_response.id).to eq(1)
-	end
-
-	context "populates its codes" do
-
-		it 'finds or creates an associated code' do
-			allow(Identity).to receive(:find_or_create_by)
-			allow(IdentifiesWith).to receive(:create)
-
-			expect(Code).to receive(:find_or_create_by).with(name: "not okay", context: "age").and_return(Code.new)
-			expect(Experiences).to receive(:create).and_return(Experiences.new)
-
-			service.perform
-		end
-
-		it 'finds or creates an associated identity' do
-			allow(Code).to receive(:find_or_create_by)
-			allow(Experiences).to receive(:create)
-
-			expect(Identity).to receive(:find_or_create_by).with(name: "genx", context: "age").and_return(Identity.new)
-			expect(IdentifiesWith).to receive(:create).and_return(IdentifiesWith.new)
-			service.perform
-		end
 
+	it 'populates experience and identity codes' do
+		expect(Code).to receive(:find_or_create_by).with(name: "not okay", context: "age")
+		expect(Identity).to receive(:find_or_create_by).with(name: "genx", context: "age")
+		service.perform
 	end
 
 end