Measuring similarity in co-occurrence data using ego-networks.

Area of Science:

• Network Science
• Data Mining
• Computational Social Science

Background:

• Co-occurrence data analysis is crucial for understanding complex systems like social networks, ecosystems, and brain networks.
• Traditional network-based similarity measures often incorporate unwanted indirect relationships from aggregated networks.
• Accurate entity similarity measurement is vital for effective information extraction from co-occurrence data.

Purpose of the Study:

• To propose a new similarity measure for co-occurrence data that addresses limitations of traditional aggregated network approaches.
• To introduce a method that accounts for changes in entity centrality across different ego networks.
• To provide a computationally efficient and interpretable index for quantifying entity similarities.

Main Methods:

• Developed a novel similarity index based on the ego network of each entity.
• Compared the proposed index against traditional network-based measures and embedding methods using two distinct datasets.
• Evaluated the index's performance in capturing unique similarity dimensions.

Main Results:

• The proposed ego network-based similarity measure outperforms traditional aggregated network approaches.
• The new index demonstrates superior performance, sometimes surpassing embedding methods.
• The measure shows weak correlation with existing methods, indicating a distinct dimension for similarity quantification.

Conclusions:

• The novel ego network-based similarity measure offers an effective extension to network analysis for co-occurrence data.
• This approach mitigates issues of indirect relationships present in aggregated network methods.
• The proposed index provides a valuable new tool for analyzing complex systems and related data mining tasks.