Constrained Topic Model

A human-in-the-loop visual explanation system that bridges machine learning interpretations and human semantic understanding through interactive drag-and-drop clustering.

How It Works

Step	Action	Output
1. Human Input	Users drag-and-drop prototypes into semantic clusters	Grouped prototypes
2. Constraint Generation	System extracts pairwise relationships from user groupings	Must-link & cannot-link constraints
3. Model Update	Constrained clustering algorithm learns from human feedback	Refined prototype clusters
4. Explanation	Model generates interpretable visual explanations	Human-aligned outputs

Core Innovation

• 🎯 Interactive Interface: Drag-and-drop clustering captures human semantic knowledge
• 🔄 Constraint-Based Learning: Converts user interactions into machine-readable constraints
• 📊 Validated Approach: User study demonstrates improved alignment with human perception

Authentication System

Built for Interactive User Studies

This project implements a robust authentication architecture specifically designed for behavioral experiments requiring user login and data persistence. Common use cases:

• Longitudinal studies where participants return across multiple sessions
• Crowdsourcing experiments (e.g., Yahoo! Crowd Sourcing, MTurk) requiring participant tracking
• Interactive tasks generating sensitive behavioral data requiring secure storage

Architecture

Request Layer (useAxiosPrivate hook)

• Auto-injects JWT tokens via Axios interceptors (no manual header configuration needed)
• AbortController-based request cancellation prevents memory leaks
• Unified error/loading state management

Authentication Flow (RequireAuth component)

• Token validation and automatic redirect on expiration
• localStorage-based session recovery (survives page refresh)
• Flexible routing: unauthenticated users redirected to registration

For Researchers

This authentication layer is designed to be reusable. To adapt for your own study:

1. Configure backend API endpoints in axios.js
2. Customize registration form fields for your participant metadata
3. Modify route protection rules in RequireAuth.js if needed

The modular design means you can focus on experiment-specific UI/UX rather than rebuilding authentication infrastructure.