GloVe Global Vectors Quiz

Answers and explanations appear as you take the quiz.

1. What core data structure does GloVe build from the corpus?
   • A global word-word co-occurrence count matrix
   • A parse tree for every sentence
   • A list of part-of-speech tags
   • A character-level frequency table
2. According to GloVe, the dot product of two word vectors is trained to approximate what?
   • The raw co-occurrence count
   • The logarithm of their co-occurrence count
   • The number of letters they share
   • Their alphabetical distance
3. What key quantity does GloVe argue carries semantic meaning?
   • The total corpus size
   • The length of each word
   • Ratios of co-occurrence probabilities
   • The order words appear in the dictionary
4. What is the purpose of GloVe's weighting function f(X_ij)?
   • To translate words between languages
   • To randomly shuffle the training data
   • To convert vectors into one-hot encodings
   • To cap the influence of very frequent pairs and ignore zero counts
5. Where and when was GloVe developed?
   • Stanford, 2014
   • Google, 2013
   • MIT, 2010
   • Facebook AI, 2016
6. How does GloVe's training approach differ fundamentally from Word2Vec's?
   • GloVe uses no vectors at all
   • GloVe factorizes a precomputed global count matrix rather than predicting from local windows online
   • GloVe only works on a single sentence
   • GloVe requires labeled training data
7. What type of word relationship can a difference between GloVe vectors represent?
   • The chronological order of sentences in the corpus
   • The exact frequency count of every word in isolation
   • A relation shared by word pairs, such as the relation between country and capital
   • The punctuation pattern used by a document author
8. In a vector analogy a:b::c:?, what operation can be used to search for the missing word with GloVe vectors?
   • Find the word with the longest spelling
   • Sort words by corpus count and choose the median
   • Add all context-window vectors without comparing candidates
   • Find the vector nearest to b − a + c