Multivariate embedding based causaltiy detection with short time series


Existing causal inference methods for social media usually rely on limited explicit causal context, preassume certain user interaction model, or neglect the nonlinear nature of social interaction, which could lead to bias estimations of causality.

Besides, they often require sufficiently long time series to achieve reasonable results. Here we propose to take advantage of multivariate embedding to perform causality detection in social media. Experimental results show the efficacy of the proposed approach in causality detection and user behavior prediction in social media.