An Information Filtering and Management Model for Twitter Traffic to Assist Crises Response Coordination

TitleAn Information Filtering and Management Model for Twitter Traffic to Assist Crises Response Coordination
Publication TypeJournal Article
Year of Publication2014
AuthorsHemant Purohit, Andrew Hampton, Shreyansh Bhatt, Valerie Shalin, Amit Sheth, John Flach
JournalJournal of Computer Supported Cooperative Work (Special Issue on Crisis Informatics and Collaboration)
Volume23
Issue4
Pagination513-545
Abstract

Disasters such as Hurricane Sandy in 2012 result in extensive social media traffic, using networking platforms such as Twitter, as citizens report on their situations, identify needs and attempt to distribute resources. We address the challenge of finding relevant, actionable tweets from this large volume with an information filtering model. Driven primarily by concern for coordination, the initial domain independent analysis incorporates psycholinguistic theory to filter for potential messages of cooperation. The subsequent domain dependent analysis leverages a lightweight, language-driven, disaster-related domain model to extract resource references (e.g., food, shelter, etc.) in its first phase. Using a lexicon of verbs concerning the transfer of property, combined with simple syntactic frames, a second phase of domain dependent analysis assists in the identification of a particular kind of tacit cooperation, in the declarations of resource needs and availability. The results populate an annotated information repository to support the presentation of organized, actionable information nuggets regarding resource needs and availability at varying levels of abstraction. Computationally grounding the abstractions in raw data enables complex querying ability for who-what-where in coordination. Initial evaluation of the annotations relative to human judgment shows fair to good agreement. In addition to the potential benefits to the formal emergency response community of a filtered and organized corpus, the results serve as a benchmark for evaluating more computationally intensive efforts and characterizing the patterns of language behavior for coordination during a disaster.