SGNS-PFNE: PATI Filtered Segmentation-free word Embedding.

SGNS-PFNE is an open source implementation of PFNE with skip-gram with negative sampling.

Environment

• Linux(Tested with Ubuntu Linux release 16.04)

Requirment

• Python3
• gcc(>=5)
• Numpy
• argparse
• pickle

Contents

1. data/: Pre-processed corpous.
2. ngram2~ngram6/: Computing PATI score from 2-gram to 6-gram.
3. embedding/: Training distributed word vectors based on SGNS-PFNE.

Reference

1. Oshikiri, T. (2017). Segmentation-Free Word Embedding for Unsegmented Languages. In Proceedings of EMNLP2017.
2. Mikolov, T., Corrado, G., Chen, K., & Dean, J. (2013). Efficient Estimation of Word Representations in Vector Space. In Proceedings of ICLR2013.