GloVe embeddings in NLP - Cheat Sheet & Quick Revision

Recall & Review

beginner

What does GloVe stand for in GloVe embeddings?

GloVe stands for Global Vectors for Word Representation. It is a method to create word embeddings by capturing global word co-occurrence statistics.

intermediate

How does GloVe differ from Word2Vec in learning word embeddings?

GloVe uses a matrix factorization approach on the global word co-occurrence matrix, while Word2Vec learns embeddings by predicting words in local context windows using a neural network.

beginner

What is the main input data structure used by GloVe to learn embeddings?

GloVe uses a word co-occurrence matrix that counts how often pairs of words appear together in a large text corpus.

beginner

Why are GloVe embeddings useful in natural language processing tasks?

They capture semantic relationships between words by encoding how frequently words co-occur globally, helping models understand word meanings and similarities.

intermediate

What kind of mathematical operation does GloVe use to learn word vectors from the co-occurrence matrix?

GloVe performs matrix factorization by minimizing a weighted least squares objective to find word vectors that reconstruct the co-occurrence counts.

What is the main data structure GloVe uses to learn word embeddings?

A Word frequency list

B Part-of-speech tags

C Word co-occurrence matrix

D Dependency parse trees

GloVe learns embeddings by factorizing the word co-occurrence matrix, which counts how often word pairs appear together.

Which of the following best describes GloVe's approach?

A Matrix factorization of global co-occurrence counts

B Using recurrent neural networks

C Clustering words by frequency

D Predicting words from local context windows

GloVe uses matrix factorization on the global co-occurrence matrix, unlike Word2Vec which predicts words from local context.

What does a GloVe embedding vector represent?

A The frequency of a word in the corpus

B Semantic meaning based on global word co-occurrence

C The position of a word in a sentence

D The length of a word

GloVe embeddings capture semantic meaning by encoding how words co-occur globally in the corpus.

Which is a key advantage of GloVe embeddings?

A They capture global statistical information

B They require no training data

C They only consider immediate neighbors

D They are random vectors

GloVe embeddings capture global co-occurrence statistics, unlike some methods that only consider local context.

What kind of loss function does GloVe minimize during training?

A Mean absolute error

B Cross-entropy loss

C Hinge loss

D Weighted least squares loss

GloVe minimizes a weighted least squares loss to factorize the co-occurrence matrix effectively.

Explain in your own words how GloVe embeddings are created from a text corpus.

Think about how often words appear together and how that information is turned into vectors.

Describe the main difference between GloVe and Word2Vec embeddings.

Focus on what data each method uses to learn embeddings.