Finding the best cosine similarity in a set of vectors

问题

I have n vectors, each with m elements (real number). I want to find the pair where there cosine similarity is maximum among all pairs.

The straightforward solution would require O(n²m) time.

Is there any better solution?

update

Cosine similarity / distance and triangle equation Inspires me that I could replace "cosine similarity" with "chord length" which loses precision but increases speed a lot. ( there are many existing solutions solving Nearest Neighbor in metric space, like ANN )

回答1:

Cosine similarity sim(a,b) is related to Euclidean distance |a - b| by

|a - b|² = 2(1 - sim(a,b))

for unit vectors a and b.

That means cosine similarity is greatest when Euclidean distance is smallest after normalizing by the L2 norm, and the problem reduces to the closest pair of points problem, which can be solved in O(n lg n) time.

回答2:

You can check with the project simbase https://github.com/guokr/simbase , it is a vector similarity nosql database.

Simbase use below concepts:

Vector set: a set of vectors
Basis: the basis for vectors, vectors in one vector set have same basis
Recommendation: a one-direction binary relationship between two vector sets which have the same basis

You can use redis-cli directly for administration tasks, or you can use redis client bindings in different language directly in a programming way. Here is a Python example

    import redis

    dest = redis.Redis(host='localhost', port=7654)
    schema = ['a', 'b', 'c']
    dest.execute_command('bmk', 'ba', *schema)
    dest.execute_command('vmk', 'ba', 'va')
    dest.execute_command('rmk', 'va', 'va', 'cosinesq')

来源：https://stackoverflow.com/questions/13661383/finding-the-best-cosine-similarity-in-a-set-of-vectors

标签

algorithm

math

cosine-similarity