L2AP - Fast Cosine Similarity Search With Prefix L-2 Norm Bounds

Current version: 1.0, 12/11/2013

L2AP is a program that provides high-performance implementations of several methods for finding all pairs of vectors whose cosine similarity is greater than a user-specified threshold. These vectors are often sparse and high-dimensional, e.g., document-term vectors, user-item ratings, etc. The methods that are implemented include approaches developed by our group that prune the search space using L2 norm bounds (L2AP and L2AP-approx) and various other state-of-the-art approaches such as AllPairs, MMJoin, and IdxJoin.

L2AP can be downloaded from here.