Mining multiple spatial–temporal paths from social media data

Abstract

Characterizing and mining people’s individual mobility pattern (IMP) has recently attracted much attention, and has shown great potential significance in many application fields, e.g., transportation planning, activity prediction, supermarket location decision, resource allocation in distributed and communication systems, etc. Social media has been widely regarded as an effective data source for IMP mining as it contains huge records with location and temporal tags. Existing social media based IMP mining methods mainly focus on the spatial aspect (e.g., points-of-interests). Some recent studies argue that, besides spatial aspect, IMP also exhibits temporal feature, and therefore it is more significant to find out the spatial–temporal path (STP). However, we notice that existing studies can only generate single STP for each individual. In practice, there may exist multiple STPs for an individual over a long time period. Regarding such fact, in this paper, we argue that it is more rationale to find out multiple STPs (mSTP) for each individual. While, this is a non-trivial task due to the inherent sparsity and uncertainty in social media based location data. To address this problem, we propose a social media based mSTP mining framework. The spatial–temporal hotspots of individual activities are first clustered. Bayesian theorem is then applied to find out the mSTP amongst the generated hotspots. An mSTP visualization method is also provided to further indicate the relationship amongst the mined mSTP. Experiments with collected trace from Weibo, the biggest social media in China, is conducted to evaluate the validity of the proposed method with respect to several metrics. The experiment results indicate that the our method is able to correctly mine mSTPs consistent with human common sense.