yagays/pretrained_doc2vec_ja

pretrained doc2vec models on Japanese Wikipedia

This is a repository of doc2vec models trained on Japanese Wikipedia corpus.

Please See my blog for more details.

Pretrained models

• dbow300d
  • https://www.dropbox.com/s/j75s0eq4eeuyt5n/jawiki.doc2vec.dbow300d.tar.bz2?dl=0
  • Compressed file size: 5.48 GB
• dmpv300d
  • https://www.dropbox.com/s/njez3f1pjv9i9xj/jawiki.doc2vec.dmpv300d.tar.bz2?dl=0
  • Compressed file size: 8.86 GB

Training setting

Parameters

param	dbow300d	dmpv300d
dm	0	1
vector_size	300	300
window	15	10
alpha	0.025	0.05
min_count	5	2
sample	1e-5	0
epochs	20	20
dbow_words	1	0

Corpus

• jawiki-20190114-cirrussearch-content.json.gz

Enviornments

• Tokenizer
  • MeCab: 0.996
  • NEologd: Periodic data update on 2019-01-17(Thu)
• requirements
  • gensim:3.7.0
  • numpy:1.16.0