PROJECT TITLE :

Feature Constrained Multi-Task Learning Models for Spatiotemporal Event Forecasting - 2017

ABSTRACT:

Spatial event forecasting from social media is potentially extremely useful but suffers from important challenges, like the dynamic patterns of options (keywords) and geographic heterogeneity (e.g., spatial correlations, imbalanced samples, and completely different populations in several locations). Most existing approaches (e.g., LASSO regression, dynamic question growth, and burst detection) address some, but not all, of these challenges. Here, we propose a novel multi-task learning framework that aims to concurrently address all the challenges concerned. Specifically, given a collection of locations (e.g., cities), forecasting models are designed for all the locations simultaneously by extracting and utilizing applicable shared info that effectively will increase the sample size for every location, so improving the forecasting performance. The new model combines each static features derived from a predefined vocabulary by domain experts and dynamic features generated from dynamic question growth in a multi-task feature learning framework. Different ways to balance homogeneity and diversity between static and dynamic terms also are investigated. And, efficient algorithms based mostly on Iterative Cluster Laborious Thresholding are developed to realize economical and effective model coaching and prediction. Extensive experimental evaluations on Twitter knowledge from civil unrest and influenza outbreak datasets demonstrate the effectiveness and potency of our proposed approach.