GloVe embeddings in NLP - Model Metrics & Evaluation

Metrics & Evaluation - GloVe embeddings

Which metric matters for GloVe embeddings and WHY

GloVe embeddings create word vectors that capture meaning by looking at word co-occurrence in text. To check if these vectors are good, we use cosine similarity. This measures how close two word vectors are in meaning. A higher cosine similarity means words are more related. For example, "king" and "queen" should have a high similarity.

We also use analogy tests like "king - man + woman = ?" to see if the embeddings capture relationships. These tests show if the model understands word connections beyond just frequency.

Confusion matrix or equivalent visualization

Since GloVe embeddings are not classifiers, we don't use confusion matrices. Instead, we look at similarity scores between word pairs.

    Word Pair        Cosine Similarity
    ----------------------------------
    (king, queen)          0.78
    (king, man)            0.75
    (king, apple)          0.12
    (apple, fruit)         0.82
    (apple, car)           0.10
    

High similarity for related words and low for unrelated words shows good embeddings.

Precision vs Recall tradeoff with concrete examples

For GloVe embeddings, the tradeoff is between specificity and generalization.

• High specificity: Embeddings capture very detailed word meanings but may miss broader connections. This is like remembering exact details but missing the big picture.
• High generalization: Embeddings capture broad relationships but may confuse similar words. This is like understanding the theme but mixing up characters.

Choosing the right balance depends on the task. For example, in sentiment analysis, generalization helps group similar feelings. In translation, specificity helps pick exact words.

What "good" vs "bad" metric values look like for GloVe embeddings

Good embeddings:

• Cosine similarity close to 1 for related words (e.g., > 0.7 for synonyms or related concepts)
• Cosine similarity close to 0 or negative for unrelated words
• Analogy test accuracy above 70% on standard benchmarks

Bad embeddings:

• Cosine similarity high for unrelated words (e.g., > 0.5 for random pairs)
• Low accuracy on analogy tests (below 40%)
• Vectors that do not cluster similar words together

Metrics pitfalls

• Ignoring context: GloVe embeddings are static and do not change with sentence meaning, so similarity may be misleading for words with multiple meanings.
• Overfitting to frequent words: Very common words may dominate co-occurrence counts, skewing embeddings.
• Using cosine similarity alone: It does not capture all semantic nuances; relying only on it can miss problems.
• Not testing on analogy or downstream tasks: Embeddings may look good by similarity but fail in real tasks like classification or translation.

Self-check question

Your GloVe embeddings show cosine similarity of 0.85 for "king" and "queen", but only 0.3 for "apple" and "fruit". Is this good? Why or why not?

Answer: This is not ideal. "King" and "queen" are related, so 0.85 is good. But "apple" and "fruit" are also related and should have a high similarity. A low 0.3 suggests the embeddings do not capture this relationship well. You may need to retrain or check your data.

Key Result

Cosine similarity and analogy test accuracy are key to evaluating GloVe embeddings' quality.

Practice

1. What is the main purpose of GloVe embeddings in natural language processing?

easy

A. To generate random text based on input

B. To translate text from one language to another

C. To count the frequency of words in a document

D. To convert words into numerical vectors that capture meaning and relationships

Solution

1. Step 1: Understand what embeddings do

   Embeddings convert words into numbers so machines can understand text.

2. Step 2: Identify GloVe's role

   GloVe embeddings specifically capture word meanings and relationships in vector form.

3. Final Answer:

   To convert words into numerical vectors that capture meaning and relationships -> Option D

4. Quick Check:

   GloVe = word vectors capturing meaning [OK]

Hint: Remember: embeddings = words to numbers showing meaning [OK]

Common Mistakes:

• Confusing embeddings with translation
• Thinking embeddings count word frequency
• Assuming embeddings generate text

2. Which of the following is the correct way to load pre-trained GloVe embeddings in Python using the gensim library?

easy

A. glove = gensim.models.FastText.load('glove.txt')

B. glove = gensim.models.Word2Vec.load('glove.txt')

C. glove = gensim.models.KeyedVectors.load_word2vec_format('glove.txt', binary=False)

D. glove = gensim.load('glove.txt')

Solution

1. Step 1: Recall GloVe loading method

   GloVe embeddings are loaded as KeyedVectors using load_word2vec_format with binary=False.

2. Step 2: Check options for correct syntax

   glove = gensim.models.KeyedVectors.load_word2vec_format('glove.txt', binary=False) uses the correct function and parameters for GloVe format.

3. Final Answer:

   glove = gensim.models.KeyedVectors.load_word2vec_format('glove.txt', binary=False) -> Option C

4. Quick Check:

   Use load_word2vec_format with binary=False for GloVe [OK]

Hint: Use load_word2vec_format with binary=False for GloVe files [OK]

Common Mistakes:

• Using Word2Vec.load for GloVe files
• Forgetting binary=False parameter
• Using FastText load for GloVe

3. Given the following Python code snippet using pre-trained GloVe embeddings, what will be the output?

from gensim.models import KeyedVectors

glove = KeyedVectors.load_word2vec_format('glove.6B.50d.txt', binary=False)
result = glove.similarity('king', 'queen')
print(round(result, 2))

medium

A. 0.00

B. 0.78

C. 1.00

D. -0.50

Solution

1. Step 1: Understand similarity method

   The similarity method returns a cosine similarity score between two word vectors, usually between 0 and 1 for related words.

2. Step 2: Interpret expected similarity for 'king' and 'queen'

   These words are closely related, so the similarity is high but less than 1, typically around 0.78.

3. Final Answer:

   0.78 -> Option B

4. Quick Check:

   Similarity('king','queen') ≈ 0.78 [OK]

Hint: Related words have similarity close to but less than 1 [OK]

Common Mistakes:

• Assuming similarity is always 1 for related words
• Confusing similarity with distance
• Expecting negative similarity for related words

4. You try to find the vector for the word 'unseenword' using GloVe embeddings with this code:

vector = glove['unseenword']

But it raises a KeyError. What is the best way to fix this error?

medium

A. Check if the word exists in the embeddings before accessing it

B. Use glove.get_vector('unseenword') without checking

C. Ignore the error and continue

D. Restart the Python kernel

Solution

1. Step 1: Understand cause of KeyError

   The word 'unseenword' is not in the GloVe vocabulary, so direct access raises KeyError.

2. Step 2: Use safe access method

   Check if the word exists using 'if word in glove' before accessing to avoid errors.

3. Final Answer:

   Check if the word exists in the embeddings before accessing it -> Option A

4. Quick Check:

   Check word presence before access to avoid KeyError [OK]

Hint: Always check word in embeddings before access [OK]

Common Mistakes:

• Trying to access vectors without checking existence
• Ignoring errors instead of handling them
• Restarting kernel does not fix missing words

5. You want to improve a text classification model by using GloVe embeddings. Which approach best combines GloVe vectors with your model to handle words not in the GloVe vocabulary?

hard

A. Initialize an embedding layer with GloVe vectors and allow it to be trainable with random vectors for unknown words

B. Use only GloVe vectors and ignore unknown words during training

C. Replace unknown words with a fixed zero vector and freeze the embedding layer

D. Train a new embedding from scratch without using GloVe

Solution

1. Step 1: Understand embedding layer initialization

   Initializing with GloVe vectors provides good starting word representations.

2. Step 2: Handle unknown words and training

   Allowing the embedding layer to be trainable lets the model learn vectors for unknown words starting from random initialization.

3. Final Answer:

   Initialize an embedding layer with GloVe vectors and allow it to be trainable with random vectors for unknown words -> Option A

4. Quick Check:

   Trainable embeddings + GloVe + random unknown vectors = best practice [OK]

Hint: Use trainable embeddings with GloVe plus random unknown vectors [OK]

Common Mistakes:

• Ignoring unknown words instead of learning their vectors
• Freezing embeddings and losing adaptability
• Not using pre-trained GloVe vectors at all