Regular Activity Patterns in Spatio-Temporal Events Databases: Multi-Scale Extraction of Geolocated Tweets
Pablo López-Ramírez, Alejandro Molina-Villegas, Oscar S. Siordia, Mario Chirinos-Colunga, Gandhi Hernández-Chan · Research in Computing Science · 2018
This paper proposes a new technique for the extraction of regular activity patterns at different scales (resolution levels), mined from the microblogging platform Twitter.The approach is based on the recursive application of the DBSCAN clustering algorithm to the geolocated Twitter feed.The proposed technique includes a novel way to obtain 'averaged' regular activity zones based on the rasterization and aggregation of the Concave Hull of the clusters identified at each resolution level.This technique uses only the spatio-temporal characteristics of the geolocated Twitter feed and does not depend on the data content; therefore it can be extended to work with different spatio-temporal event sources such as mobile telephone records.An experiment was carried out to demonstrate the effectiveness of our technique in the extraction of known activity patterns in the Mexico City Metropolitan Area.