Turkish-WebVectors

Turkish WebVectors: Tool to explore semantic relations between words in distributional models. Based on WebVectors prepared by Kutuzov et al. [1].

Installation

• First, clone the repository from GitHub to local machine using the following command:

git clone https://github.com/akutuzov/webvectors.git .

• Then, create a virtual environment using the following command:

python -m venv env

• Then, install all the Python requirements using the following command:

pip install -r requirements.txt

• Gunicorn is a pure Python WSGI server with simple configuration and multiple worker implementations for performance tuning. It is used for production deployment. Install gunicorn as follows:

pip install gunicorn

How to run using Docker?

• First, clone the repository from GitHub to local machine using the following command:

git clone https://github.com/Turkish-Word-Embeddings/Turkish-WebVectors.git . 

• Launch a terminal in the root directory of the repo and build the Docker image where -t is the tag for the Docker image. You can provide any name you want. . is the relative path to the Dockerfile

docker build -t turkish-webvectors .

• Dockerfile downloads the necessary two models from our release page. You can also download them manually and put them in the models directory.
• Run the Docker image using the following command:

docker run -p 8000:8000 turkish-webvectors

Models

Both of the following models are trained with gensim using 300 dimensions, 10 epochs, 5 negative samples, minimum word count set to 10 and window size set to 5.

• Word2Vec skip-gram Negative Sampling
• FastText skip-gram Negative Sampling

---

1. Kutuzov, Andrei & Kuzmenko, Elizaveta. (2017). WebVectors: A Toolkit for Building Web Interfaces for Vector Semantic Models. Communications in Computer and Information Science. 661. 155-161. 10.1007/978-3-319-52920-2_15.