Tweets2Cube: Interactive spatiotemporal knowledge acquisition from massive social media

Abstract

We present the Tweets2Cube system that uncovers the patterns underlying people's spatiotemporal activities from massive online social media. Tweets2Cube organizes unstructured social media records into a multi-dimensional data cube along three dimensions: (1) what is the user's activity; (2) where does that activity occur; and (3) when does that activity occur. As such, the end users can use simple queries to retrieve task-relevant sub-corpus from the data cube in a flexible way. Moreover, Tweets2Cube consists of a set of spatiotemporal modeling algorithms, which can be readily applied to the retrieved data for extracting knowledge about people's activities in the physical world. Such algorithms jointly model location, time, and text and are capable of discovering a variety of patterns, such as routine spatiotemporal activities, unusual events, and mobility patterns. With Tweets2Cube, the end users can interactively retrieve task-relevant social media and choose appropriate spatiotemporal modeling algorithms for knowledge acquisition, which makes Tweets2Cube highly useful for downstream tasks like disaster relief, targeted advertising, and location-based recommendation.