PyTorch Word2Vec implementation

Description

Unofficial implementation of the Efficient Estimation of Word Representations in Vector Space paper written in PyTorch with code for training and demonstration of the properties of the trained model. Emphasis was placed on the Skip-gram Model only.

Content

Files to be familiarized with:

1. word2vec.pth is a pre-trained model on the Amazon Fashion dataset with a 4000-word vocabulary,
2. inference.ipynb contains the playground and demonstrates some properties of the model,
3. train.ipynb trains word2vec from scratch. Use it if you want to customize the training process for yourself,
4. extra/cloud.svg shows t-SNE visualization of the most distinct word clusters.

Installation

git clone https://github.com/tejpaper/word2vec.git
cd word2vec
pip install -r requirements.txt

Some clusters

Emotions and feelings

Family

Seasons

Numbers

Colors

Body parts

Clothes

Sizes

License

MIT