News & Events

17WCEE proceedings

Assessing Post-disaster Recovery Using Sentiment Analysis. The case of L'Aquila, Haiti, Chile and Canterbury 

 

Abstract

Memorial days of disasters represent an opportunity to evaluate the progress of recovery. Social media posts from the 10th anniversary of the earthquakes, in L’Aquila, Haiti, Chile, and Canterbury constitute a valuable source of data to mine insights into the progress of their recovery. The sentiments of affected people have the potential to highlight the successes and gaps in the post-earthquake recovery process. We propose a new method to assess post-disaster recovery through sentiment analysis of disaster-related tweets from citizens. The procedure for assessing recovery consists so far of 9 phases namely, hashtag identification, hashtag selection, data collection, data extraction, data processing, classification, feature extraction, features reprocessing and features selection. The data collection starts with social media monitoring, we identify the hashtags related to the memorial of each specific earthquake and then we collect tweets using two sources:1) manually and 2) a third party vendor from which we can collect the tweets related to the selected hashtags seven days around the anniversary of the event. The data process prepares the data to be understood by a machine learning algorithm. The classification phase combines a machine learning algorithm and supervised classification, tweets are categorized as positive, negative and neutral. Based on this classification and the general content of the tweets it is then possible to assess the relative success of the recovery based on the satisfaction of the community with the process. To detect the sentiment of tweets, we have combined supervised classification in the case of L’Aquila and linguistic features in the case of Haiti to classify the text according to the tweeter’s assessment of the recovery process. In the case of L’Aquila, whose 10 year anniversary took place on the 6th April 2019, we have obtained 4,349 original tweets between the 5th and the 10th of April 2019 with the hashtag #L'Aquila. In the case of Haiti, the 10 years anniversary took place on the 12th January 2020, we have obtained 8,157 original tweets between the 7th and the 15th of January 2020. So far we have identified more than 40 tweets regarding the anniversary of the earthquake in Chile and two tweets with respect to the anniversary of the Canterbury earthquake. The preliminary sentiment analysis of the tweets for the cases of L’Aquila and Haiti evidences that the negative polarity prevails. We do not yet have a sample big enough to determine the polarity of the twitter posts in the case of Chile and Canterbury. Nevertheless, preliminary results allowed us to demonstrate that sentiment analysis is a feasible tool to evaluate the success of a post-disaster recovery process. Affected communities in L’Aquila and Haiti are not satisfied with the progress of the post-earthquake recovery. In the case of L’Aquila, reconstruction of private buildings is more advanced than public ones and that there has been a delay in the re-opening of schools and other urban facilities in the historical city centre. The complaints about the mismanagement of financial resources and corruption are common in both cases: L’Aquila and Haiti.

Keywords: Post-disaster recovery; earthquakes; machine-learning algorithm; Twitter; sentiment analysis.

Full-text: 

 

 

Last modified: Mon, 05 Oct 2020 14:10:38 BST