A collection of visualizations for exploring word embeddings in Word2Vec models. The code is dirty, the project not well set up, it may break, your computer may burn, but when it works it looks pretty.

• Python3
• Poetry

Clone the sources from the repository and compile it with

poetry install --no-dev

or if you need to development dependencies (for example to train models)

poetry install

This project uses streamlit. To run it type

poetry run streamlit run main.py

The models have to be in a subfolder called models besides main.py. So it should run out of the box with the example models when you clone this repository.

Build it with

docker build -t word-embeddings:latest .

then run it with

docker run -it -v `pwd`/models:/code/models word-embeddings:latest

where the volume that is mounted contains the Word2Vec models.

Note: You can add a description file by creating a text file with the same name as the model but with extension .txt.

The models must have the extension .bin and placed in the models directory.

The models must have the extension .model and placed in the models directory.

The models must have the extension .kv and placed in the models directory.

• word2vec_abc_news: A model trained from the https://www.kaggle.com/therohk/million-headlines dataset. The dataset contains over a million news headlines published over a period of eighteen years from the Australian Broadcasting Corporation (ABC).

• word2vec_bigram_abc_news: A model trained from the https://www.kaggle.com/therohk/million-headlines dataset. The dataset contains over a million news headlines published over a period of eighteen years from the Australian Broadcasting Corporation (ABC). It is similar to the word2vec_abc_news model, but it contains bigrams (try searching for new_york).

• https://github.com/marcellusruben/Word_Embedding_Visualization for PCA and t-SNA visualization and making me aware of Streamlit.
• streamlit for creating an awesome piece of software.