Application of Content Analysis for Natural Disasters

Effective communication requires identifying the target audience, selecting effective channels to deliver information, monitoring the outcomes of the message, and gathering feedback from the targeted audience.^{Reference Schiavo42} Several studies focused on the use of social media as a tool for delivering emergency warnings and disseminating information related to natural disaster response (Table 1). Content analysis was 1 of the techniques implemented to study information dissemination and early emergency warnings. This research technique can help detect that important topics that should be disseminated during the response to a natural disaster are being discussed by social media platforms users. Analysis of Twitter accounts during the Boulder, Colorado, floods showed that only 6% of the analyzed tweets were health-related posts regarding drinking water safety, floodwater exposure, cleaning, and hygiene, despite the existence of water advisory warning.^{Reference Sutton, League, Sellnow and Sellnow26} Authorities during the South Carolina floods of 2015 used Twitter to bring attention to reducing health threats, devastation to structures, resource distribution, appreciation, and fund-raising.^{Reference Brandt, Turner-McGrievy and Friedman32} However, social media engagement from official agencies using multiple platforms is required for an effective conversation and integration of communities.^{Reference Tang, Zhang and Xu27}

Application of Social Network Analysis for Natural Disasters

SNA sought to understand and observe how the connections between individuals affect their relationships or behaviors and how information disseminates among the users of the platform.^{Reference Kim, Bae and Hastak8,Reference Kim and Hastak25,Reference Valente43} In 4 articles, SNA showed citizens were consistently the most active Twitter users during disasters of different natures.^{Reference Wang, Ye and Tsou30–Reference Ramirez Plascencia and Ramirez Plascencia33} Twitter was identified as a broadcasting tool to disseminate information related to weather conditions, avoiding threats, providing warnings, information about policies, and correcting misinformation during several natural disasters.^{Reference Albris29,Reference Grasso and Crisci31,Reference Brandt, Turner-McGrievy and Friedman32,Reference Yi and Kuri34,Reference Kaufhold and Reuter35,Reference Huang and Xiao9,Reference Stephenson, Vaganay and Coon41} SNA of Weibo during the 2012 Yiliang earthquake allowed researchers to detect smaller networks focused on specific topics, such as personal information, caution and advice, casualties and damage, donations, and request for help.^{Reference Li, Zhang and Tian36} Similar results were reported during the 2013 floods in Germany where Facebook and Twitter were used to organize help, volunteers, and share updates.^{Reference Kaufhold and Reuter35}

Research found that SNA can be included in communication efforts during the response to natural disasters. SNA can help communication specialists learn what features of social media can be included in messages, and the time they can reach a higher number of users. SNA for Hurricane Sandy revealed information could be disseminated more extensively by using the retweet function of Twitter, but most of the retweets happened an hour after sharing the original content, which exemplifies the need of using the correct channels for communication.^{Reference Wang and Zhuang10} Social media posting frequencies need to be considered for analyses. During Typhoon Haiyan and Storm Cindy tweets peaked on the day of the landfall.^{Reference Kim, Bae and Hastak8,Reference David, Ong and Legara37} Limited tweets were collected up to 48 h after landfall of the typhoon, and during Storm Cindy, Twitter interest disappeared 10 days after the event.^{Reference Kim, Bae and Hastak8,Reference David, Ong and Legara37} It was also observed that when other simultaneous national events occurred, interest in natural disasters decreased.^{Reference Kim and Hastak25}

SNA permits the recognition of changes in the information that social media users are sharing or seeking. Before a natural disaster takes place, social media users are more inclined to post messages related to information about the event and disaster relief.^{Reference Li, Zhang and Tian36,Reference David, Ong and Legara37} SNA also showed that, when no severe weather warning is active, social media activity is low.^{Reference Grasso and Crisci31} The results from these publications emphasize the usefulness of SNA to detect when platforms users are active, and when a warning or disaster response information can be shared to capture the attention of a broader number of users. SNA can help with the construction of messages tailored to answer the questions of affected individuals. These messages should include an imperative and declaratory style, which is essential for health promotion during disasters, and a hashtag to categorize messages under a topic.^{Reference Sutton, League, Sellnow and Sellnow26,Reference David, Ong and Legara37} Consistency in hashtag use can help with information dissemination and identification of those in need.^{Reference Grasso and Crisci31,Reference Cooper, Yeager and Burkle38} An essential part of the health communication cycle tailoring messages for the intended audience and adopting appropriate channels for delivery.^{Reference Schiavo42} SNA allowed the recognition of the preferences in social media platforms of individuals affected by natural disasters. During the 2015 Hurricane Patricia in Mexico, researchers observed emergency response agencies used Twitter to provide official messages, but citizens shared the tweets on Facebook where they went viral.^{Reference Ramirez Plascencia and Ramirez Plascencia33}

Identification of the target population is essential for successful message dissemination.^{Reference Schiavo42} One way of sharing information with a large number of users is by identifying social media influencers. Influencers are opinion leaders with accounts that have large numbers of followers, messages, and likes, that have the power to influence followers.^{Reference Juliadi and Ardani44} Social media data analysis was used to identify influencers in the affected area before and after events.^{Reference Cooper, Yeager and Burkle38} Four articles identified individuals as the most prominent users of social media;^{Reference Wang, Ye and Tsou30–Reference Ramirez Plascencia and Ramirez Plascencia33} 3 identified news or media, nonprofit organizations, and government agencies as key accounts for information dissemination during natural disasters.^{Reference Wang and Zhuang10,Reference Kim and Hastak25,Reference Pohl, Bouchachia and Hellwagner40} These influential accounts can be used as a channel to share relevant information and emergency warnings to the affected population considering the trends of information changes in social media. However, identifying these accounts before the event takes place can present a challenge for emergency management officials, and a time-consuming one if they are identified during the response to a natural disaster.

Recognition of Relevant Social Media Data During Natural Disasters as a Tool to Identify Physical, Medical, Functional, and Emotional Needs

Extracted social media data can be overwhelming for emergency responders to analyze in a fast and effective manner, particularly because so many posts may be irrelevant. Twenty-eight articles analyzed relevant information on social media and how it helped to identify the needs of the affected communities (Table 2). An analysis of Hurricane Matthew and Twitter data identified that only 15.04% of the analyzed posts were related to the disaster, and 4.33% were related to damage during the event.^{Reference Yuan and Liu47} When focusing on social media posting behavior and the area where the event took place, research generally indicated an inverse relationship between distance from hurricanes and storms to areas at risk and relevant tweets to the events.^{Reference de Albuquerque, Herfort and Brenning48,Reference Kryvasheyeu, Chen and Obradovich49} However, for earthquakes, an opposite Twitter behavior was observed, with a higher number of social media posts generated in areas where the earthquake was felt with less intensity or not felt at all.^{Reference Comunello, Parisi and Lauciani50,Reference Zahra, Ostermann and Purves51}

Analysis of these data may be further clouded by the fact that, during natural disasters, posting behavior can deviate from baseline behavior, leading to fewer posts originating in geographical areas usually identified with a high volume of social media posts.^{Reference Zou, Lam and Cai52} To help decrease the number of unrelated social media data used in situation analysis, researchers suggested interviewing the responders and creating specific categories to achieve their goals.^{Reference Kiatpanont, Tanlamai and Chongstitvatana11} Including categories of interest for the agencies responding to natural disaster events allowed researchers to reduce Twitter data for the Thailand floods of 2011 to 30% of the original sample size.^{Reference Kiatpanont, Tanlamai and Chongstitvatana11}

Content and Sentiment Analysis as a Tool to Identify Physical, Medical, Functional, and Emotional Needs After Natural Disasters

Content analysis during a disaster of social media posts categorized information as damage reports, injuries, transportation conditions, power outages, evacuations, and concerns for rescue during severe weather events, typhoons, earthquakes, and wildfires.^{Reference Deng, Liu and Zhang20,Reference Wang, Ye and Tsou30,Reference David, Ong and Legara37} Social media researchers have developed methodologies to use online data to identify the physical, medical, functional, and emotional needs after a disaster by comparing the content of messages shared during the preparedness phase and after the event has taken place.^{Reference Yuan and Liu47,Reference Kryvasheyeu, Chen and Obradovich49} During the Nepal earthquake 2015, Twitter data analysis revealed that deaths and killing were the most mentioned categories 15 h after the earthquake.^{Reference Andrews, Gibson and Domdouzis56} Facebook, YouTube, and Twitter facilitated communication efforts by connecting those in need of help with responders, sharing photos, and sharing ideas to save water during natural disasters.^{Reference Tang, Zhang and Xu27,Reference Albris29,Reference Tim, Pan and Ractham53} To identify geographical areas affected by Hurricane Matthew, damage-related tweets served as a better predictor than disaster-related tweets.^{Reference Yuan and Liu47} During Hurricane Sandy in 2012, a positive correlation was observed between the per-capita number of tweets and disaster-inflicted monetary damage calculated from insurance claims after landfall with a Pearson correlation coefficient of 0.6.^{Reference Kryvasheyeu, Chen and Obradovich49}

In addition to structural and environmental damages, natural disasters also created a change in the emotional state of the individuals expressed through social media.^{Reference Wu and Cui18,Reference Albris29,Reference David, Ong and Legara37,Reference Zou, Lam and Cai52,Reference Gul, Shah and Ahad54,Reference Ragini, Anand and Bhaskar55,63} Sentiment analysis methods studied individuals’ feelings, emotions, expressions, and trends about a particular topic or natural disaster.^{Reference Ungvarsky12} For example, sentiment analysis detected negative terms in Twitter content like “death” when focusing on a food category and “crisis” when analyzing medical emergencies.^{Reference Ragini, Anand and Bhaskar55} Analyzing the sentiment of social media messages helped to identify the areas in need, but the identification of where response efforts should be deployed first was a challenge. To evaluate the changes in the sentiment of Weibo users after the Ya’an earthquake in 2013, researchers implemented a 4-quadrant distribution to help with faster identification of areas in need to plan and deliver an adequate response, where areas with a high negative sentiment needed to be addressed immediately.^{Reference Bai and Yu46}

Mapping Tools for Social Media Data to Identify Physical, Medical, Functional, and Emotional Needs After Natural Disasters

Several teams used social media data analysis in innovative approaches by constructing maps for damage identification, actions needed, reports of deaths, earthquake detection, mobility detection, and inundation maps.^{Reference Resch, Uslander and Havas21–Reference Avvenuti, Cresci, Del Vigna and Tesconi23,Reference Pohl, Bouchachia and Hellwagner40,Reference Andrews, Gibson and Domdouzis56–Reference Wang and Taylor62} Mapping social media information can help government agencies or emergency responders to attain a spatiotemporal view to offer an efficient response.^{Reference Xu, Zhang and Sugumaran57} Maps using social media data included spatial variables that can also help visualize the collected data at different geographical levels.^{Reference Resch, Uslander and Havas21–Reference Avvenuti, Cresci, Del Vigna and Tesconi23,Reference Pohl, Bouchachia and Hellwagner40,Reference Xu, Zhang and Sugumaran57–Reference Li, Wang and Emrich61}

Mapping strategies using social media data have gone beyond identifying a specific location. Researchers have successfully developed mapping systems as a situational awareness and damage reports tool after earthquakes and flooding events.^{Reference Avvenuti, Cresci, Del Vigna and Tesconi23} Researchers harnessed social media images to identify affected areas, estimate depth of the flood, and damages during the event.^{Reference Cervone, Sava and Huang22,Reference Fohringer, Dransch and Kreibich60,Reference Li, Wang and Emrich61} While investigating Twitter data to build rapid flooding maps during the 2015 South Carolina floods, researchers were able to identify flooded areas previously unknown in official reports.^{Reference Li, Wang and Emrich61} Mapping social media data was helpful for emergency responders to decrease the time needed for a response, such as in the case of the Chennai floods where Twitter data analysis allowed the calculation of faster routes to connect those in need with those offering help.^{Reference Nath, Priya and Rene Robin45}

Despite the advances reached by developing mapping methods using social media data analyses, several challenges were reported by researchers when testing their frameworks. Overestimation of damage in areas affected by earthquakes and low accuracy for earthquake detection for events with a Richter magnitude of 3.5 or lower were reported.^{Reference Resch, Uslander and Havas21,Reference Avvenuti, Cresci, Del Vigna and Tesconi23,Reference Avvenuti, Cresci and La Polla59} The reviewed articles also presented challenges when analyzing flooding events. When using images from Twitter, Flickr, and other spatial data, depth of inundation was overestimated, and the presence of clouds on images affected the results in overestimation or underestimation of structural damage,^{Reference Cervone, Sava and Huang22,Reference Fohringer, Dransch and Kreibich60} suggesting the need for methods that overcome the limitations caused by our environment.