Understanding Pol.is Algorithms
Learn how Pol.is analyzes and visualizes participant opinions
Overview
Pol.is is an open-source platform that uses machine learning to facilitate large-scale, real-time conversations. Unlike many other platforms, Pol.is doesn't analyze text or use natural language processing. Instead, it focuses on analyzing structured responses (agree, disagree, or pass) to create a mathematical understanding of opinions.
The Opinion Matrix
| Participant | Comment 1 | Comment 2 | Comment 3 |
|---|---|---|---|
| User A | Agree | Disagree | Pass |
| User B | Agree | Agree | Disagree |
Key Algorithms
1. Dimensionality Reduction
Principal Component Analysis (PCA)
PCA reduces the complexity of high-dimensional data by identifying the most important patterns in participants' responses. Think of it as finding the "main themes" in how people vote.
UMAP (Uniform Manifold Approximation and Projection)
UMAP helps visualize complex relationships between opinions while preserving important structures in the data. This creates the opinion group visualizations you see in Pol.is.
2. Clustering Algorithms
K-Means Clustering
Groups participants with similar voting patterns into distinct clusters, helping identify major opinion groups.
Leiden Algorithm
A sophisticated method for finding communities within the network of participants based on their voting patterns.
Benefits of This Approach
Scalability
Can handle conversations with hundreds or thousands of participants effectively
Language Agnostic
Works across any language since it analyzes votes, not text
Real-Time Analysis
Provides immediate insights as participants engage
Understanding the Visualizations
Reading Opinion Group Maps
What You'll See
- Colored clusters representing different opinion groups
- Dots representing individual participants
- Distances between groups showing how different their opinions are
- Group sizes indicating how many participants share similar views
What It Means
- Closer groups = more similar opinions
- Larger groups = more common viewpoints
- Scattered dots = unique or diverse perspectives
- Overlapping areas = potential consensus points
Common Patterns to Look For
Consensus
When groups are close together or overlapping, it suggests shared views and potential for agreement
Polarization
Distinct, widely separated groups indicate strongly differing viewpoints on key issues
Bridging
Participants between groups often hold views that could help bridge differences
Making Use of These Insights
Tips for Discussion Owners
- Monitor how opinion groups form and evolve over time
- Look for comments that receive broad agreement across groups
- Identify potential areas of consensus for further exploration
- Use insights to guide additional seed comments or discussion topics
- Pay attention to bridging opinions that connect different groups
What Makes a Good Discussion?
Healthy Patterns
- Multiple distinct but not extremely polarized groups
- Presence of bridging opinions
- Mix of consensus and divergent viewpoints
- Active participation across groups
Warning Signs
- Extreme polarization with no bridging views
- One dominant group with minimal other participation
- No clear opinion patterns emerging
- Lack of engagement with different viewpoints
Frequently Asked Questions
How many participants do I need for meaningful results?
While Pol.is can work with any number of participants, you typically need at least 15-20 active participants to start seeing meaningful opinion patterns emerge. The more participants, the more reliable the patterns become.
How long should I run my discussion?
This depends on your goals, but most discussions benefit from running for at least a week to allow time for different viewpoints to emerge and for participants to engage with various comments. Some discussions run for months to capture evolving opinions.
What if no clear groups are forming?
This might indicate that your seed comments aren't capturing the key points of disagreement, or that the topic needs more specific focus. Consider adding new seed comments that address potential areas of disagreement or refinining your discussion prompt.