TAQE: Tweet Retrieval-Based Infrastructure Damage Assessment During Disasters

Abstract

Twitter is an active communication channel for the spreading of updated information in emergency situations. Retrieving specific information related to infrastructure damage offers the situational views to the concerned authorities, who can take necessary action to disburse help. However, such usages of Twitter demand significant accuracy of the retrieved information. Previous techniques on IR have not been able to capture the semantic variations satisfactorily in the tweets, due to low content quality and vocabulary gap, and consequently have failed to yield considerable performance. This has left ample scope for further improvement in this area of research. There are two major contributions of our work: 1) developing a relevant tweet retrieval framework that provides information about infrastructure damage and 2) assignment of a relative damage score to the affected regions so that the severity of the damage can be assessed. Our proposed technique involves a novel split-query-based mechanism with topic aligned query expansion (TAQE) to retrieve relevant tweets that are subsequently used for measuring the infrastructure damage across different locations. We report empirical results on multiple-crisis-related data sets to establish the efficacy of our approach to these events at different locations. Empirical validation of our proposed approach on manually annotated ground-truth data reveals considerably better performance metrics in terms of precision, recall, Bpref, and MAP over several state-of-the-art techniques.