Embedding

Text Embedding

• The most widely used type of embedding
• Convert text into a dense vector representation of semantic understanding
• Real-world use cases
  • semantic search
  • question answering
  • chatbots
  • document clustering

Text Embedding Models

• BERT (Bidirectional Encoder Representations from Trnansformers)

  • created and used by Google
  • 768 dimenion ectors
  • context-aware Bidirectional encoding
  • pre-trained on 3.3B words

• SBERT (Sentence BERT (see BERT above))

  • created by UKP Lab
  • 5x faster than BERT for sentence comparison
  • siamese network architecture

• OpenAI text-embedding-3

  • Created by OpenAI
  • 3072-dimension vectors
  • better multi-lingual support
  • cost-effective API access