Status : Verified
Personal Name Lagria, Raymond Freth A.
Resource Title A lookup-based decision support system for classification and prioritization of disaster-related tweets for disaster response
Date Issued 07 January 2019
Abstract Disasters have been around for generations and have unsettled the normal functions and way of life of countries and communities around the world. The Philippines is one of the most badly hit by disasters every year and due to less coping capabilities the country has constantly been ranked in the top 3 of the World Risk Index.
As part of the efforts to lessen the impacts of disasters is by improving technology-based capabilities and one of these tech-based platform is Twitter. Twitter is a social media platform where users post messages called tweets. Twitter has been experimental when it comes to disasters as current literature discusses Twitter as instrument for text mining frameworks used in Disaster Risk Reduction Management. The applications of Twitter ranges from content analysis, sentiment analysis and disaster events prediction.
The problem that this research tackles is that there is no existing multilingual text mining framework that can classify and prioritize Filipino-authored disaster related tweets for faster disaster response by local disaster responders. There is also a need to identify standard disaster information class labels and improve the accuracy of disaster related tweets classification models in terms of correctly predicted class labels.
The proposed framework describes four stages composed of different modules namely the Pre-processing Stage, Topics Modelling, Classification Module and the Prioritization Module which introduces the Tweet Prioritization Score (TPS). The end result of the proposed framework produces a set of prioritized tweets that would call the attention of and be used by disaster managers for planning and performing immediate disaster response.
To test the proposed framework, case studies were done on five sample tweets datasets – three of which are real datasets extracted during noteworthy typhoons in the past two years (Typhoon Basyang 2018, Typhoong Domeng 2018 and Typhoon Urduja 2017) and the other two datasets are scripted datasets (Typhoon Ekis and Typhoon Arya) which were used to verify the functional features of the framework.
Results of the Classification Module found that Random Forests produced the highest average accuracy of 69.22% for the direct multiclass classification and average accuracy of 78.60% for transformation to binary classification. The prioritization scoring module yielded assuring results where tweets with high TPS were verified and validated to produce tweets worthy of being urgent. The final output gave assuring results to which local disaster managers can utilize during planning and responding to disasters.
Degree Course MS Industrial Engineering
Language English
Keyword textmining, datamining, classification, decisionsupportsystems, dss, informationsystems
Material Type Thesis/Dissertation
Preliminary Pages
282.57 Kb