This project is an interactive 3D visualization tool for exploring word embeddings and their relationships. It allows users to visualize word vectors in a 3D space, compare their similarities, and even add their own words to the visualization dynamically.
- Interactive 3D Space: Word vectors are represented as arrows in a 3D space that can be rotated and explored.
- Vector Comparison: Click on two vectors to compare them and see their similarity scores.
- Similarity Metrics: Cosine similarity is used to determine the relationship between vectors.
- Dynamic Word Addition: Add new words to the visualization and see how they relate to existing words.
- Controls: Easily reset the view, clear selections, or start with a fresh canvas.
- HTML, CSS, JavaScript: The core technologies for building the web application.
- Three.js: A 3D graphics library for rendering the vector space.
- @xenova/transformers: A library for generating word embeddings directly in the browser. Specifically used the Xenova/all-MiniLM-L6-v2 model.