Phrase Embeddings

Distributed vector representations of multi-word units (phrases, idioms, named entities) that capture meaning beyond the composition of individual word vectors. Phrase embeddings address a fundamental limitation of word embeddings: many phrases (e.g., "Boston Globe", "New York Times") have non-compositional meanings not captured by combining word vectors.

Key Papers

• Distributed Representations of Words and Phrases and their Compositionality — Data-driven phrase identification via unigram/bigram frequency scoring; shows phrase embeddings achieve 72% accuracy on phrase analogy tasks; demonstrates compositional structure via vector addition (Czech + currency → koruna)

Related topics

• Word Embeddings (atomic units)
• Text Representations (broader category)
• Natural Language Processing (NLP applications)