Live Engine
Select Topic
easyWord Representations
A team trains Word2Vec Skip-gram on a 10GB news corpus with window size 5 and 300 dimensions. They query
model.most_similar("king") and get ["queen", "prince", "monarch", "emperor", "throne"]. A colleague claims the model "understands" kingship. What does the model actually encode, and why is "understanding" an overstatement?