Mining exceptional social behavior on attributed interaction networks.

Area of Science:

• Social Network Analysis
• Data Mining
• Computational Social Science

Background:

• Social interactions are increasingly recorded via online and offline sensors, forming time-stamped interaction networks.
• Analyzing these networks often involves identifying subgroups with unique behavioral patterns, typically focusing on network structure.
• Attributed social networks offer richer data by including additional attributes beyond simple connections.

Purpose of the Study:

• To propose a novel approach for identifying exceptional social behavior patterns in attributed social interaction networks.
• To focus on the dyadic structure of these networks, enabling a compositional analysis.
• To adapt local pattern mining and subgroup discovery techniques to spatio-temporal attributed interaction data.

Main Methods:

• The approach models spatio-temporal interaction data as attributed social networks.
• It adapts local pattern mining and subgroup discovery to analyze the dyadic structure and attributes.
• Seven novel quality functions are proposed to measure the interestingness of identified subgroups, considering interaction duration and frequency.

Main Results:

• The method was evaluated on four real-world datasets, including academic conferencing and school playground interactions.
• The approach successfully identified interesting, meaningful, and valid subgroups exhibiting exceptional social behavior.
• Results demonstrate the efficacy of exploiting dyadic structures and attributes for subgroup discovery in social interaction networks.

Conclusions:

• The proposed method effectively identifies subgroups with deviant social behavior by analyzing spatio-temporal attributed interaction networks.
• Focusing on dyadic structures and attributes provides a powerful compositional perspective for subgroup discovery.
• The novel quality functions enhance the ability to find significant patterns in complex social interaction data.