Regional Analysis #2

[D2-02] Mapping Emotional Geographical Trends and User Behaviors Based on Japanese Tweets: A Case Study of Urban Green Spaces in Tokyo 23 Wards

*Mingchen Liu¹, Yuya Shibuya², Yoshihide Sekimoto² (1. Department of Civil Engineering, The University of Tokyo, 2. Center for Spatial Information Science, The University of Tokyo)

Keywords:Sentiment Analysis, Natural Language Processing, Twitter, Machine Learning, Behavior Analysis

Emotions are the basis of human behavior and an important part of spatial constructions. Social Networking Service has become an important source of spatio-temporal big data due to many users sharing their emotions and behaviors, which helps us to clarify the relationship between sentiment changes and geographical locations. This research focuses on sentiment analysis of the one-year geo-tagged Japanese Tweets to understand the sentiment and behavioral patterns in urban green spaces in Tokyo 23 wards. We estimate the sentiment of Tweets with a fine-tuned pre-trained Japanese BERT model. The results show that 30% of the Tweets are estimated to be positive and 5% are negative. In addition, we found that their behaviors are different when they are in different moods. Users in positive moods engaged in behaviors such as walking, taking pictures, flower viewing, and painting, while words such as “go home”, “go to sleep”, and “give up” appeared more often in negative moods.