Mining Unusual Search Behavior Related to Physical Events

Abhay Prakash, Parag Agrawal, Kedhar Nath Narahari, Puneet Agrawal

Abstract: Search logs of commercial search engines like Bing or Google reflect the topics of interest in humans at any given point of time. While many of these topics are predictable information needs, some are unusual and point to a curious human mind. In this paper, we propose a novel solution for mining unusual search behaviors triggered by some recent physical events using Bing search logs. Our algorithm successfully identified unusual search behaviors related to physical events during 2016. We define unusual-relatedness and use crowdsource judgment to evaluate quality of our results along with a qualitative analysis. Our results show a 9.5 times improvement over baseline.