feat: Retrieve cosine similarity from Postgres query

lib/medical_transcription/coding.ex

@@ -8,7 +8,7 @@ defmodule MedicalTranscription.Coding do
   import Ecto.Query
   import Pgvector.Ecto.Query
 
-  alias MedicalTranscription.Coding.CodeVector
+  alias MedicalTranscription.Coding.{CodeVector, CodeVectorMatch}
 
   def insert_vector(params) do
     changeset = CodeVector.changeset(%CodeVector{}, params)
@@ -37,14 +37,13 @@ defmodule MedicalTranscription.Coding do
 
   def process_chunk(text, opts \\ []) do
     k = Keyword.get(opts, :num_results, 5)
+    similarity_threshold = Keyword.get(opts, :similarity_threshold, 0.80)
 
     search_vector_for_db = compute_vector_as_list(text)
 
     find_similar(search_vector_for_db, k)
-
     # -- Remove matches that don't exceed a given threshold
-    # TODO: This depends on receiving a similarity score from the Postgres query
-    # |> Enum.filter(fn {_index, score} -> score >= similarity_threshold end)
+    |> Enum.filter(fn %CodeVectorMatch{} = score -> score.cosine_similarity >= similarity_threshold end)
   end
 
   @doc "Creates a vector embedding for text using the text embedding serving in the application's supervision tree."
@@ -59,7 +58,8 @@ defmodule MedicalTranscription.Coding do
     Repo.all(
       from v in CodeVector,
         order_by: cosine_distance(v.description_vector, ^search_vector),
-        limit: ^limit
+        limit: ^limit,
+        select: %CodeVectorMatch{id: v.id, code: v.code, description: v.description, cosine_similarity: 1 - cosine_distance(v.description_vector, ^search_vector)}
     )
   end
 end

lib/medical_transcription/coding/code_vector_match.ex

@@ -0,0 +1,6 @@
+defmodule MedicalTranscription.Coding.CodeVectorMatch do
+  @moduledoc """
+  Represents a vector match found in the database, along with its similarity score.
+  """
+  defstruct [:id, :code, :description, :cosine_similarity]
+end

lib/medical_transcription_web/components/transcription_text_component.ex

@@ -6,7 +6,7 @@ defmodule MedicalTranscriptionWeb.Components.TranscriptionTextComponent do
   use MedicalTranscriptionWeb, :live_component
   import MedicalTranscriptionWeb.Components
   import MedicalTranscriptionWeb.Components.KeywordHighlighter
-  alias MedicalTranscription.Coding.CodeVector
+  alias MedicalTranscription.Coding.CodeVectorMatch
 
   @impl Phoenix.LiveComponent
   def update(assigns, socket) do
@@ -53,10 +53,8 @@ defmodule MedicalTranscriptionWeb.Components.TranscriptionTextComponent do
       </div>
 
       <div class="flex-1 flex flex-col items-stretch gap-3">
-        <%= for %CodeVector{id: id, code: code, description: label} <- @row.tags do %>
-          <!-- NOTE: We're not currently deriving a similarity score from our Postgres query using cosine similarity.
-               For now, pass 1.0 until we research whether we can determine that score from the query. -->
-          <.tag_result code_vector_id={id} code={code} label={label} score={1.0} text={@row.text} />
+        <%= for %CodeVectorMatch{id: id, code: code, description: label, cosine_similarity: score} <- @row.tags do %>
+          <.tag_result code_vector_id={id} code={code} label={label} score={score} text={@row.text} />
         <% end %>
       </div>
     </div>