Predicting Election Results Via Social Media: A Case Study for 2018 Turkish Presidential Election

Abstract

Social media platforms provide massive amounts of data that can be used to analyze social issues and forecast events in the future. However, it is a challenging task due to the biased and noisy nature of the data. In this work, we propose a method to predict election results via Twitter. In particular, we first detect the stance of social media accounts using their retweets. Subsequently, we develop four different counting methods for our prediction task. In the simple user counting (SC) method, we count labeled users without taking any further steps to reduce bias. In the city-based weighted counting (CBWC) method, we apply a weighted counting based on the number of electorate in each city. The closest-city-based prediction (CCBP) method utilizes sociological similarity between cities to predict results for cities with limited sample sizes. The using former election results (UFERs) method compares predictions for each city against former election results to detect data bias and uses them accordingly. We evaluate our proposed methods with the data collected for the presidential election of Turkey held in 2018. In our extensive evaluation, we show that utilizing domain-specific information and location-based weighted counting is effective in reducing bias. CBWC, CCBP, and UFER methods outperform tweet-counting-based baseline methods. Furthermore, UFER and CCBP outperform almost all traditional polls, suggesting that social media platforms are alternative mediums for conducting election polls.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 13 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.