Geotemporal analysis and topic modelling of Twitter data: study in nine big city areas of Indonesia

Abstract

Nowadays, Twitter data is significant to many studies since there is a shift in the data collection paradigm. As one of the contemporary social media with many active users, Twitter provides geotagging facilities to create a geotagged Tweet. Various spatial based studies use geotagged Tweet data. This paper aims to review the geo-temporal characteristics of geotagged Twitter data in nine major cities in Indonesia, namely five cities in the Greater Area of Jakarta, Surabaya, Bandung, Medan, and Makassar. Twitter data was collected by the streaming method for two years (January 2019- December 2020). The temporal analysis was carried out by graphing the number of Tweets with 30-minute intervals. Weekly Twitter activities were also visualized to get a specific understanding of when the optimum time to post a Tweet was. Density analysis was employed to Twitter data to find out the spatial patterns in the study area. Kernel Density Estimation (KDE) was used to determine the Tweets Density in the day and night. This study also used a simple framework of text analysis of topic modelling using Latent Semantic Indexing (LSI) to use the Twitter data better. Overall, Central Jakarta and South Jakarta have a significant number of Tweets compared to other cities. The study results show that, in general, big cities in Indonesia have almost the same temporal curve and the peak time for making geotagged tweets occurs from 4 pm to 8 pm. Our finding also points out that a high number of the population in a city does not always produce a high number of Tweets. The results of topic modelling in the Greater Area of Jakarta show that the themes of traffic jams/congestion, entertainment, and culinary tourism are widely mentioned by Twitter users, thus opening opportunities for research on these subjects.