Paper Contribution -

Analyzed and put forward the the model properties required for capturing fine-grained semantic and syntactic regularities through vector arithmetic. New GLOBAL LOGBILINEAR REGRESSION model that combines global matrix factorization and local window method. How is it achieved?

model efficiently leaverages statistical information by training only on no-zero elements in word-word co-occurence matrix rather than on the entire sparse matrix or an individual context window in large corpus.

Outcome - vector space with meaningful sub-structure. 75% performance on word nalaogy task. also out performs related models on similarity and NER tasks.

Main contribution of the paper