Estimation of 2017 Iran's Presidential Election Using Sentiment Analysis on Social Media

Nowadays with growth of social media, they become a part of every man's life; so, they can be used in analysis and prediction tasks. People share their feelings, opinion and viewpoints on these media. One of the most important uses of social media is in national elections. In days near the election, people share their opinions, candidates share their plans and channels try to broadcast election events. So, data scientists can analyze these widely broadcasting messages to predict the election results. In this paper, we propose using both text and meta data analysis methods including sentiment analysis of hashtags and messages, time and reputation analysis to predict Iran's 2017 presidential election. We used sentiment analysis of messages on words with positive and negative polarities for text analysis and hashtags to determine the polarity of messages for metadata analysis. In addition, we used time analysis to weight messages score by their closeness to the election. Finally, we used reputation analysis of messages to calculate the impact of messages on people's opinion. For doing so, we used the number of views on telegram messages and numbers of members of the channels to weight messages by an appropriate weight. Our experiments on twits and telegram data show that the proposed model achieved 97.3% accuracy compares to the real results of the election.