C5W2 Word Embeddings Flashcards
1
Q
Name two ways of word representation?
A
One hot representation and embedding representation
2
Q
How to visualise N-dimensional space?
A
Use t-SNE algorithm
3
Q
What are the advantages of word embeddings?
A
- You can use unlabelled data to learn the embedding(or take open source), and then use transfer learning for your task
- Word embedding can help to learn analogies
4
Q
Difference between one-hot vectors and word embedding vectors?
A
One hot is sparse, while embedding is dense
5
Q
What is a embedding matrix?
A
Matrix where height is a number of embeddings, and width is a size of a vocabulary
6
Q
How model learns embeddings?
A
You need to learn model to predict target word based on context word.
7
Q
What is word2vec?
A
Mapping from word to embedding vector
8
Q
What is cosine similarity?
A
Function of similarity of two words based on their embedding vectors
9
Q
Main advantage of word embeddings?
A
It allows your model to work on words that are not in train set
