NLP - Text Similarity and Search
Why can cosine similarity sometimes give misleading results when comparing very sparse vectors in NLP tasks?
Why can cosine similarity sometimes give misleading results when comparing very sparse vectors in NLP tasks?
hard📝 Conceptual Q10 of 15
Step-by-Step Solution
Solution:
Step 1: Understand sparse vector behavior
Sparse vectors often have many zeros; normalization can inflate similarity even with little overlap in non-zero elements.Step 2: Identify why cosine can mislead
Vectors with few shared features can still have relatively high cosine similarity due to normalization effects.Final Answer:
Because sparse vectors may have little overlap but still high cosine similarity due to normalization -> Option BQuick Check:
Sparse vectors can mislead cosine similarity [OK]
Quick Trick: Normalization can inflate similarity for sparse vectors [OK]
Common Mistakes:
MISTAKES- Thinking cosine ignores vector length entirely
- Assuming cosine requires dense vectors
- Confusing sensitivity to magnitude with normalization effect
Master "Text Similarity and Search" in NLP
Want More Practice?
15+ quiz questions · All difficulty levels · Free
Free Signup - Practice All QuestionsMore NLP Quizzes