Capturing information needs in disaster situations by using temporal and spatial offset learning (TSOL)

This paper presents a framework for identifying the offline information needs of persons in disaster situations by analyzing online behavioral logs and utilizing the users' location and search history. Two main challenges are addressed: accurately identifying persons most affected by the situation from noisy location data and distinguishing event-related search queries from unrelated ones. To tackle these challenges, we propose a machine-learning method, called temporal and spatial offset learning (TSOL), that incorporates both temporal and spatial distinctiveness. TSOL assigns heavier weights to these dimensions, in order to offset complexities and uncertainties surrounding user information. We validated the effectiveness of TSOL through experiments in actual disaster situations. The proposed framework and TSOL offer a promising approach to capturing and analyzing the information needs of individuals affected by disasters. The captured information needs in disaster situations have often been reported on TV in Japan as a support of those affected by disasters. • We propose a framework for identifying the information needs of people affected by disasters in Japan by extracting information from the online behavioral logs of users and utilizing the users' location and search history. • We developed the TSOL method embodying the framework and validated its effectiveness through offline experiments. • We built a practical TSOL application by using location information and search queries obtained from users of the Yahoo! JAPAN app, which boasts the largest user base in Japan.