Understanding Social Media Behavior in Philippines Presidential Election using Natural Language Processing

Abstract

Philippines Presidential election was conducted last May 9, 2022, and the voter turns out is 83% which is unprecedented. In this study, students at Cebu Technological University were asked for social media accounts and profiles and gathered Facebook posts 1 week before the Philippines May 9, 2022 election using a beautifulsoup and selenium scraper. A total of 300 student Facebook profiles were gathered and a total of 5321 Facebook posts were collected. Data pre-processing was performed to delete irrelevant data such as emoticons, images, and URL. After data pre-processing, a total of 3210 Facebook posts were processed using Latent Dirichlet Allocation (LDA) and K-means clustering. A total of 10 topic models were generated using LDA and 5 clusters were captured using K-means clustering. Open Coding technique was used to analyze the result of each algorithm and the result shows that the topic models are focused on the following narrative: {Marcos Golden Era, VP Leni Angat Buhay, Dilawan J}. For the clusters, the following narratives: { Marcos tops in Surveys, Presidential Debates, Unity of Candidates, Religious leaders endorsing, Marcos disqualification case}. The topic models and clusters were evaluated using human annotators and the result is 50.01% which denotes that the quality of topic and cluster models are on the average level. Sentiment analysis was also performed using Support Vector Machine with 70% training set and 30% evaluation. Each Facebook post was manually labeled in the category of “Negative”, “Neutral” and “Positive”. The accuracy of the model is 73%. All Facebook posts were classified and the result shows that 60% were negative.