Forecasting Election Polls with Spin Systems

TL;DR

This paper models political forecasting as a spin system ground state problem, enabling the use of optimization techniques, and demonstrates its effectiveness with Twitter data for trend detection and prediction.

Contribution

It introduces a novel approach mapping election prediction to spin systems and reformulates it as HUBO/QUBO problems for classical and quantum optimization.

Findings

Good agreement between model predictions and Twitter poll data.

Shows political forecasting is NP-Hard.

Applicable to trend detection and fake news identification.

Abstract

We show that the problem of political forecasting, i.e, predicting the result of elections and referendums, can be mapped to finding the ground state configuration of a classical spin system. Depending on the required prediction, this spin system can be a combination of XY, Ising and vector Potts models, always with two-spin interactions, magnetic fields, and on arbitrary graphs. By reduction to the Ising model our result shows that political forecasting is formally an NP-Hard problem. Moreover, we show that the ground state search can be recasted as Higher-order and Quadratic Unconstrained Binary Optimization (HUBO / QUBO) Problems, which are the standard input of classical and quantum combinatorial optimization techniques. We prove the validity of our approach by performing a numerical experiment based on data gathered from Twitter for a network of 10 people, finding good agreement between results from a poll and those predicted by our model. In general terms, our method can also be understood as a trend detection algorithm, particularly useful in the contexts of sentiment analysis and identification of fake news.