Sentiment Analysis and Ridesharing Models of Disaster Response
MetadataShow full metadata
Hurricanes, wildfires, floods, and earthquakes: these natural disasters remind us that humans are small and vulnerable. They destroy thousands of lives and properties, and the frequency of natural disasters is remarkably increased year to year. To reduce the negative impact of natural disasters, all the countries and regions around the world attach great importance to disaster response. Unfortunately, most natural disasters are inevitable, and resources like foods, medicines, and vehicles are always limited when natural disasters occur. Motivated by improving the efficiency and fairness, this thesis focuses on the preparedness and response stages of emergency management for better disaster response. It consists of two sections: analyzing the public’s attitudes towards the disaster response through social media and introducing the idea of ridesharing into disaster evacuations.
Analyzing attitudes hidden in social media data, known as sentiment analysis, is crucial for government or relief agencies to improve disaster response efficiency, but it has not received sufficient attention. Therefore, this section implements Python in collecting Twitter data. Then, the public perception is assessed quantitatively by these opinioned texts, which contain information like the demand for targeted relief supplies, satisfactions of disaster response, and fear of the public. A natural disaster dataset with sentiment labels is created that contains 49,816 records about natural disasters in the United States. Second, this section proposes eight machine learning models for sentiment classification, which are the most popular models used in classification problems. Third, the comparison of these models is conducted via various metrics. This section also discusses the optimization method of these models from the perspective of model parameters and input data structures. Finally, a set of real-world instances is studied from the standpoint of analyzing public opinion changes during different natural disasters and understanding the relationship between similar types of natural disaster and time series.
In terms of disaster evacuation, timely evacuation is crucial to disaster response, as people can avoid suffering and loss of lives when a disaster happens. With the popularity of the sharing economy, ridesharing has grown in recent years, which has the advantage of reducing congestion, saving travel time, and optimizing transportation mode. Thus, we propose integrating the concept of ridesharing into evacuation. Participants involved in the ridesharing evacuation plan are divided into two groups based on drivers (volunteers who are willing to offer ridesharing services) and rider (victims who do not have a private vehicle for evacuation). Then, a mixed-integer programming model is proposed, in which individuals who have vehicles can choose either evacuate to a gathering location directly or provide a ride to carless individuals along the way when a disaster occurs. Furthermore, variants of the previous model are developed, which consider different vehicle capacities and the split evacuation route. In this research, a real-world case study based on Houston, Texas is used to validate these proposed models. A series of instances are designed to compare the evacuation efficiency using three indicators, evacuation percentage (EP), average evacuation percentage (AEP) and average travel distance (ATD).