How to process text embeddings classification on n8n

Navigate to Settings > Credentials in your n8n workspace. Click + Add Credential and select your AI service (OpenAI, Cohere, or Hugging Face). Enter your API key and test the connection. Save the credential with a descriptive name like OpenAI-Embeddings.

Add a Manual Trigger or Webhook node to start your workflow. Configure the input to accept text data by adding a JSON payload structure:

{
  "text": "Your text to classify",
  "categories": ["category1", "category2"]
}

Insert a Code node after your trigger. Add JavaScript code to clean and prepare your text:

const cleanText = $json.text
  .toLowerCase()
  .replace(/[^\w\s]/g, '')
  .trim();

return {
  originalText: $json.text,
  cleanedText: cleanText,
  categories: $json.categories
};

Add an OpenAI node and select Get Embeddings operation. Set the Input Text field to {{ $json.cleanedText }}. Choose the embedding model (recommended: text-embedding-ada-002). Select your previously created credentials.

Add another OpenAI node to generate embeddings for your classification categories. Use an Item Lists node to split categories, then connect to the OpenAI node. Set the input to {{ $json }} to process each category name into embeddings.

Add a Code node to compute cosine similarity between text and category embeddings:

const textEmbedding = $('OpenAI').first().json.embedding;
const categoryEmbeddings = $('OpenAI1').all();

function cosineSimilarity(a, b) {
  const dotProduct = a.reduce((sum, val, i) => sum + val * b[i], 0);
  const magnitudeA = Math.sqrt(a.reduce((sum, val) => sum + val * val, 0));
  const magnitudeB = Math.sqrt(b.reduce((sum, val) => sum + val * val, 0));
  return dotProduct / (magnitudeA * magnitudeB);
}

const similarities = categoryEmbeddings.map(cat => ({
  category: cat.json.input,
  similarity: cosineSimilarity(textEmbedding, cat.json.embedding)
}));

return { similarities };

Add a final Code node to select the highest similarity score:

const similarities = $json.similarities;
const bestMatch = similarities.reduce((best, current) => 
  current.similarity > best.similarity ? current : best
);

return {
  originalText: $('Code').first().json.originalText,
  predictedCategory: bestMatch.category,
  confidence: bestMatch.similarity,
  allScores: similarities
};

Connect your final node to a Webhook Response or Set node to output results. Add Error Trigger and Stop and Error nodes to handle API failures gracefully. Configure retry logic in the OpenAI nodes with Retry on Fail: 3 times.