Statistically Modeling the Effectiveness of Disaster Information in Social Media

Abstract

Twitter has increasingly become an important source of information during disasters. Authorities have responded by providing related information in Twitter. The same information channel can also be used to deliver disaster preparation information to increase the disaster readiness of the general public. Retweeting is the key mechanism to facilitate this information diffusion process. Understanding of factors that affect twitter users' retweet decision would help the authority to adopt an optimal strategy in choosing the content, style, key words, initial targeted users, time and frequency. This helps optimizing the communications of disaster messages given the unique characteristics of the Twitter medium. As a result, it will speed up the information propagation to save more lives. In this paper, we present the analysis of user's retweeting behavior by studying the factors that may affect this decision, including context influences, network influences and time decaying factors. We aim to build a fine-grained predictive model for retweeting. Specifically, given a tweet, we would like to predict the retweeting decision of each user within a targeted network. We use logistic regression to formulate the problem into a retweeting probability conditioned on the incoming tweet and targeted users. We use this model to examine message spread, because disaster messages do not supersede other communication in the Twitter medium (unlike the emergency alert system announcements over traditional mediums such as television and radio), resulting in a need to 'earn' visibility (e.g., through a high following or reTweeting). We also analyze how time decay would affect user's retweet decision, which in turn affect the information spread and speed. Simulation results illustrate that our model has preferable recall and precision for retweet predicting, and can forecast the trend of information diffusion in the network.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 40 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.