Election Prediction Based on Messages Feature Analysis in Twitter Social Network

Abstract

With the emergence of virtual social networks, predicting social events such as elections using social network data has attracted the attention of researchers. In this paper, three indicators for election prediction have been proposed. First, the tweets are grouped based on a specific time window. Next, the indicator values for each candidate in each time window are calculated based on the sentiment scores and re-tweet numbers. In fact, the indicators are calculated based on the ratio of features related to positive to negative sentiments. Finally, using the aging estimation method, the indicator values for each party on the election date are predicted. The party with larger predicted indicator values will be considered as the winner. Investigations into Twitter data related to 2016 and 2020 US presidential elections on a four-month time span indicate that the indicator values and elections can be predicted with a high accuracy.

Sentiment Analysis on Marketplace in Indonesia using Support Vector Machine and Naïve Bayes Method

“…After preprocessing, data is then labeled. Data labeling in sentiment analysis can be done using transformer [44], TextBlob techniques [16], or VADER [45] for sentiment scores [46] -1 to 1 respectively i.e. negative and positive classes.…”