Semantic distance measures how dissimilar two language representations are in meaning or contextual association. Nearby words in vector space tend to share surrounding words; distant words tend not to. section 6.4.5
Read distance values with these practical implications:
- Values near zero indicate high similarity, as with San Francisco and California. section 6.4.5
- Large distances indicate weakly shared context, as with cars and peanuts. section 6.4.5
Comparing word associations
For the word nurse, Word2Vec reports a smaller distance to woman (0.5586) than to man (0.7453). Such comparisons can reveal associations in training data, but they should not be treated as neutral facts.
