Integrating Prior Knowledge in Mixed-Initiative Social Network Clustering.

Area of Science:

• Social Network Analysis
• Data Mining
• Human-Computer Interaction

Background:

• Social scientists often struggle with complex clustering algorithms and lack guidance in selecting and evaluating them.
• Existing tools do not adequately incorporate scientists' prior knowledge into the clustering process.

Purpose of the Study:

• To introduce PK-clustering, a novel approach and visual analytics interface for social network analysis.
• To address the challenges social scientists face in understanding, selecting, and evaluating clustering algorithms by integrating prior knowledge.

Main Methods:

• Capturing scientists' prior knowledge as incomplete clusters.
• Running multiple clustering algorithms and evaluating their results against prior knowledge.
• Utilizing a visual analytics interface for ranking, summarizing, and iteratively updating cluster analysis.

Main Results:

• A functional prototype of the PK-clustering approach was developed.
• Initial use cases and feedback from social scientists demonstrate the approach's utility.
• The method provides a constructive way to build knowledge iteratively.

Conclusions:

• PK-clustering offers a novel, user-centered method for social network analysis.
• The approach empowers social scientists to create meaningful clusters by leveraging their expertise.
• It mitigates over-reliance on "black-box" algorithms by integrating domain knowledge.