Negative Sampling Improves Hypernymy Extraction Based on Projection Learning

We present a new approach to extraction of hypernyms based on projection learning and word embeddings. In contrast to classification-based approaches, projection-based methods require no candidate hyponym-hypernym pairs. While it is natural to use both positive and negative training examples in supervised relation extraction, the impact of negative examples on hypernym prediction was not studied so far. In this paper, we show that explicit negative examples used for regularization of the model significantly improve performance compared to the state-of-the-art approach on three datasets from different languages.

The russian model.

$ python -V; pip show tensorflow numpy scipy scikit-learn gensim | egrep -i '(name|version)'
Python 3.5.2 :: Continuum Analytics, Inc.
Name: tensorflow
Version: 0.12.1
Name: numpy
Version: 1.12.0
Name: scipy
Version: 0.18.1
Name: scikit-learn
Version: 0.18.1
Name: gensim
Version: 0.13.4.1

The english-combined model has been trained using the well-known word embeddings dataset based on Google News: GoogleNews-vectors-negative300.bin on EVALution, BLESS, K&H+N, ROOT09 combined. The english-evalution model is traned on EVALution only.

$ python -V; pip show tensorflow numpy scipy scikit-learn gensim | egrep -i '(name|version)'
Python 3.5.2 :: Anaconda custom (64-bit)
Name: tensorflow
Version: 0.12.1
Name: numpy
Version: 1.11.3
Name: scipy
Version: 0.18.1
Name: scikit-learn
Version: 0.18.1
Name: gensim
Version: 0.13.4.1