Network selection: a method for ranked lists selection.

Area of Science:

• Computational Social Science
• Data Mining
• Bioinformatics

Background:

• Social choice theory and voting often require aggregating individual preferences into a single group ranking.
• Heterogeneous preference lists can prevent the identification of a unique, true underlying ranking.
• Existing aggregation methods may fail when dealing with diverse or conflicting preference data.

Purpose of the Study:

• To develop a novel algorithm for aggregating heterogeneous preference lists.
• To address the challenge of non-existent unique rankings in social choice and voting theory.
• To improve the accuracy and relevance of aggregated rankings by considering data homogeneity.

Main Methods:

• Proposing the Network Selection algorithm, inspired by graph theory.
• Identifying distinct communities of homogeneous rankings within a heterogeneous dataset.
• Aggregating rank orderings only within their respective identified communities.
• Applying the algorithm to simulated, financial, and biological datasets.

Main Results:

• Network Selection significantly outperforms existing related methods on simulated data.
• On real financial data, the algorithm effectively selects relevant variables for data mining predictive models, enhancing predictive power.
• Demonstrated potential in bioinformatics through an application to a biological microarray dataset.

Conclusions:

• Network Selection offers a robust approach to aggregating heterogeneous preference lists by leveraging community detection.
• The algorithm provides superior performance in both simulated and real-world scenarios, including data mining and bioinformatics.
• This method offers a more feasible solution for problems where a single underlying ranking is not apparent.