Vector Space Visualization
About This MicroSim
This MicroSim demonstrates how embeddings position similar items near each other in vector space by visualizing MicroSims as points in a 2D projection.
Iframe Embed Code
1 2 | |
Description
Understanding Vector Space
| Concept | Meaning in This Visualization |
|---|---|
| Point | A MicroSim positioned by its embedding |
| Cluster | MicroSims with similar content |
| Distance | Semantic dissimilarity (closer = more similar) |
| Color | Subject area classification |
How to Use
- Click any point to select it and see its neighbors
- Adjust the neighbor count using +/- buttons
- Toggle cluster backgrounds to see subject groupings
- Toggle distance lines to see connections to neighbors
- View the side panel for similarity percentages
Visual Elements
- Colored points: MicroSims grouped by subject
- Cluster shading: Light backgrounds showing subject regions
- Distance lines: Dashed lines to nearest neighbors
- Information panel: Selected point details and neighbor list
Learning Objectives
After using this MicroSim, students will be able to:
- Explain how embeddings position similar items near each other
- Identify clusters of related content in vector space
- Interpret the meaning of distance in semantic similarity
Lesson Plan
Introduction (5 minutes)
- Explain that embeddings convert text to numerical vectors
- Introduce the concept of "semantic space"
- Discuss how similar meaning = close position
Exploration Activity (15 minutes)
- Click different points, observe which neighbors appear
- Notice how Physics items cluster together
- Compare distances within vs. between clusters
- Adjust K to see how neighborhood size affects results
Analysis Questions (10 minutes)
- Why are all Physics MicroSims near each other?
- What does it mean when a Chemistry item is close to Physics?
- If you added a new MicroSim about "sound waves," where would it appear?
Discussion (5 minutes)
- Limitations of 2D projection (information loss)
- Real embeddings have 384 dimensions
- Applications: recommendation systems, search