Beyond word2vec: GloVe, fastText, StarSpace
summary
Compares word2vec, GloVe, fastText, StarSpace by claim, tool and result: uses summed context vectors and off‑the‑shelf embeddings as ML features, reaches about 78% accuracy on a 2,000‑sentence metaphor dataset, and reports weaker but promising results on sarcasm. Notes subword n-grams in fastText handle misspellings, GloVe builds a co‑occurrence matrix so training is fast after that step, and StarSpace generalizes embeddings to non‑text entities with hinge/softmax plus negative sampling.
AI-written summary · about
chapters (9)
- 0:00Word embeddings intro
- 2:44Why embeddings matter
- 8:54word2vec to GloVe fastText
- 15:07StarSpace beyond text
- 19:23Embedding research questions
- 23:17Using embeddings in machine learning
- 27:19Metaphor detection apps
- 30:02Transfer learning with embeddings
- 31:29Q&A
related talks
39mNext generation of word embeddings in Gensim
40mText similiarity with the next generation of word embeddings in Gensim
01h 01mWord Embeddings - An Alternative & Efficient Approach to Search for Docs
34mWord Embeddings for fun and profit in Gensim
01h 03mWord Embedding For Fun and Profit
01h 28mDocument/Sentence Similarity Solution
28m"Embeddings: How Computers Learned to Read"
32mGenerative AI Text and Multimodal Embedding Models for Real World Use Cases