Vector Embedding

Vector embeddings are dense, low-dimensional numerical representations of discrete entities (such as words, phrases, sentences, or documents) in a continuous vector space.

Key Characteristics:

  • Dimensionality: Typically range from 50 to 1000 dimensions
  • Density: All elements are non-zero, unlike sparse representations
  • Learned: Generated through machine learning techniques on large corpora


  • Capture semantic relationships and contextual information
  • Enable mathematical operations on textual data
  • Facilitate similarity comparisons between entities


  • Word Embedding (e.g., Word2Vec, GloVe, FastText)
  • Sentence Embeddings (e.g., BERT, USE, Sentence-BERT)
  • Document Embeddings (e.g., Doc2Vec, BERT-based models)

Generation Methods:

  • Prediction-based (e.g., Word2Vec)
  • Count-based (e.g., GloVe)
  • Transformer-based (e.g., BERT)



  • Preserve semantic relationships
  • Enable efficient computation
  • Generalize well to unseen data
  • Supports Transfer Learning


  • Require large amounts of training data
  • May struggle with rare words or domain-specific terminology
  • Can inherit biases present in the training data