Predicting election outcomes based on voters' weighted networks

摘要

Researchers study election process from different approaches to predict results or analyze social behavior of voters. We propose a model based on the network science to model voters' mutual opinions similarity. To model voters opinions, weighted networks (agreement and disagreement) are build from voters' previous votes. Connections in these networks shows score of individuals' opinions similarities. Applying community detection algorithms help us to have high-level view of voters' behavior as groups. Building weighted hyper-graph from groups help to analyze voters' parties opinions. A linear probabilistic model predicts nominee's success chance in polling using features extracted from networks' weights. The model is verified on Wikipedia admin-ship election data-set with good precision results.