What is embedding drift in NLP models?

Embedding drift occurs when learned vector representations become outdated due to data shifts.

In production NLP systems, word meanings and distributions evolve. Without retraining embeddings, model performance degrades over time.