Word2Vec-Galaxy

An interactive 3D visualization tool for high-dimensional word vectors and performing vector arithmetic operations.

References

• Mikolov, Tomas, et al. "Efficient estimation of word representations in vector space." arXiv preprint arXiv:1301.3781 (2013).
• Mikolov, Tomas, et al. "Distributed representations of words and phrases and their compositionality." Advances in neural information processing systems 26 (2013).
• Gensim Word2Vec - Word2Vec model

Features

• 3D Word Vector Visualization - Explore semantically similar words in interactive 3D space using PCA dimensionality reduction
• Vector Arithmetic Operations - Perform and visualize classic word analogies (king - man + woman = queen) with step-by-step vector operations
• Interactive Controls - Intuitive sidebar controls for customizing visualizations and selecting parameters

Repository Structure

Word2Vec-Galaxy/
┣ .streamlit/
┃ ┗ config.toml
┣ main.py                         ← Streamlit frontend
┣ word_vectors.py                 ← Word2Vec model handler
┣ visualization.py                ← 3D plotting functions
┣ requirements.txt
┗ README.md

Getting Started

Follow these steps to set up and run the Word2Vec Galaxy visualization tool locally.

1. Prerequisites

• Python 3.10+ (Conda or Venv recommended)
• 4GB+ RAM (for handling the Word2Vec model)

2. Clone the Repository

git clone https://github.com/Gayanukaa/Word2Vec-Galaxy.git
cd Word2Vec-Galaxy

3. Create & Activate Conda Environment

conda create -n word2vec python=3.11 -y
conda activate word2vec

4. Install Python Dependencies

pip install -r requirements.txt

Running the Application

From the project root:

streamlit run main.py

The app will open in your browser (usually at http://localhost:8501).

First Run Setup

• The application will automatically download the Google News Word2Vec model on first launch
• Subsequent runs will use the cached model for faster startup

How to Use

Similar Words Visualization

1. Enter a word in the sidebar (e.g., "pet", "computer", "happiness")
2. Adjust the number of similar words using the slider (5-50)
3. Click "🔍 Visualize Similar Words"
4. Explore the 3D plot where colors indicate semantic similarity

Vector Arithmetic

1. Enter three words for the analogy:
   • Word 1 (subtract): e.g., "man"
   • Word 2 (add): e.g., "woman"
   • Word 3 (base): e.g., "king"
2. Click "🔢 Calculate Analogy"
3. View the result (e.g., "queen") and explore the vector visualization.

Example Analogies

• king - man + woman = queen
• walking - walk + run = running

Technical Details

Word Vector Model

• Uses Google's pre-trained Word2Vec model (300 dimensions)
• Vocabulary: ~3 million words and phrases
• Trained on Google News dataset (100 billion words)

Dimensionality Reduction

• PCA (Principal Component Analysis) for reducing 300D vectors to 3D
• PC1, PC2, PC3 represent the three principal components with highest variance

Known Limitations & Solutions

Input Word Echo Problem: Word2Vec analogies sometimes return input words instead of true analogical matches. This occurs because:

• Input words have high similarity to the computed result vector
• The model may prefer familiar words over novel analogical relationships
• Example: king - man + woman might return "king" instead of "queen"

Resolve: The application automatically filters out all input words from analogy results, forcing the model to find genuine analogical relationships rather than echoing familiar terms.

Common Issues

• Threading errors: Restart the application if model loading hangs
• Path issues: Ensure you're running from the project root directory
• Conda environment: Activate the correct environment before running

License

This project is licensed under the MIT License, allowing for open-source collaboration and modification.