NIPS Proceedingsβ

Streaming Pointwise Mutual Information

Part of: Advances in Neural Information Processing Systems 22 (NIPS 2009)

[PDF] [BibTeX]



Recent work has led to the ability to perform space efficient, approximate counting over large vocabularies in a streaming context. Motivated by the existence of data structures of this type, we explore the computation of associativity scores, other- wise known as pointwise mutual information (PMI), in a streaming context. We give theoretical bounds showing the impracticality of perfect online PMI compu- tation, and detail an algorithm with high expected accuracy. Experiments on news articles show our approach gives high accuracy on real world data.