Word2Vec Skip-Gram and CBOW Quiz

Answers and explanations appear as you take the quiz.

1. What does the Skip-Gram architecture predict?
   • Surrounding context words from the center word
   • The center word from surrounding context words
   • The next sentence in a document
   • The part-of-speech tag of each word
2. What is the actual useful output of training a Word2Vec model?
   • The prediction accuracy on context words
   • The learned weight matrix, whose rows are the word vectors
   • A parse tree of the sentence
   • A probability for each document class
3. Why does Word2Vec use negative sampling?
   • To increase the vocabulary size
   • To translate between languages
   • To avoid the expensive full softmax over the whole vocabulary
   • To remove stop words automatically
4. What does subsampling frequent words do in the later Word2Vec training method?
   • It reduces the influence of very frequent words and can speed training
   • It turns every word into a character n-gram
   • It changes CBOW into a sentence classifier
   • It guarantees better vectors for every rare word
5. What famous property of Word2Vec vectors is illustrated by "king - man + woman ≈ queen"?
   • Vector analogies via linear arithmetic
   • Exact dictionary lookup
   • Grammatical error correction
   • Sentence-length normalization
6. Who introduced Word2Vec and in what year?
   • Geoffrey Hinton at Toronto, 2006
   • Tomas Mikolov and colleagues at Google, 2013
   • Yann LeCun at Bell Labs, 1998
   • OpenAI, 2018
7. What does the CBOW architecture predict from its input context?
   • Each future paragraph in the document
   • The sentiment label of the full corpus
   • The center word surrounded by the context window
   • The spelling of a character n-gram
8. What distribution does the original Word2Vec paper use to draw negative samples?
   • A uniform distribution over all character n-grams
   • A distribution based only on word length
   • A distribution proportional to squared inverse frequency
   • A unigram word-frequency distribution raised to the 0.75 power