https://github.com/efeharmankaya/mp3-comp472

Mini-Project 3 Experiments with Word Embeddings

This project aims to experiment with word embeddings to solve a synonym test and compare the results between different models, random choice and a human gold-standard.

Setup

> pip install -r requirements.txt

Run Experiments

Each model run will generate a unique details file that shows the results of each synonym check in the form synonym word, correct answer, guess, result. If the result is 'guess' then none of the words given in the question were found in the model's corpus and a random guess was made.

> python mp3.py

Loading Models

The required model names are automatically loaded through the form below. Additional models can be found in the Genism Data Repo or by running api.info().get('models'). While running mp3.py the models listed in the models array are loaded and saved for future use. The inital loading process can take upwards of 5 minutes as the models can range from 100MB to 1.2GB. However, after the first load, the models can be accessed directly through the saved wordvector files in the models directory.

models = [
    {'model_name' : 'word2vec-google-news-300', 'model_file_name' : 'GoogleNews300.wordvectors', 'details_file_name' : 'GoogleNews300-details.csv'}
]