Tweeting in Disaster Area: An Analysis of Tweets during 2016 Ma j or Floods in Indonesia

Social media allows people in the disaster area to communicate disaster information, to the people outside the disaster area, more quickly and accurately. Unfortunately, there are limited researches that examine the use of Twiter by people in the disaster sites. This study aims to explore the use of Twiter by users in the disaster-afected areas. We use the feature of twiter geolocation, to separate information from inside and outside the disaster site. This research gives depiction about communication behavior of people in the afected disaster area, through social media. The result showed that people in disaster location use twiter to give irst-hand report, coordinate rescue efort, provide help and express grief. In addition, by focusing on the afected area, Twiter used by lay people is usually found rather than other users. From the segment of time, the researcher inds a number of tweets that will increase each day. Users will share more information the days after rather, than the day of disaster. In practical term, this research explores the used of social media by the victims of disaster, which can encourage efective communication to people or group outside the location; theoretically, this research gives more detail understanding about shared information from the people in the disaster place.


Introduction
and Lennart, 2014). Shklovski, Palen, & Suton (2008) adds that disaster also causes the damage of communication infrastructure and information, so that it decreases the availability of information related to the disaster.
The emergence of social media pledges the better disaster communication. An individual can communicate two ways to the society about the disaster (Fraustino, Liu, and Yan, 2012). Alexander (2014) identiies seven beneits of social media in disaster, such as giving people the opportunity to give moral and material supports, monitoring the situation of the disaster, integrating social media data into disaster management, collaborating to help disaster victims, making social cohesion and promoting therapeutic initiatives, fundraising and research.
Social media in disaster communication is able to fulill deiciency of traditional media, especially in the function, as the platform gives up-to-date information (Takahashi, Tandoc and Carmichael, 2015). People who are in the disaster area can communicate current condition to other people outside the disaster area.  find that people outside the disaster location use more social media rather than the people in the disaster area. We believe this makes most of the researchers about social media usage in disaster management, to focus on the usage of social media without regarding where the information is from (Smith, 2010;Acar and Muraki, 2011;Muralidharan, Dillistone and Shin, 2011;Bruns et al., 2012;Murthy and Longwell, 2013). Thus, the research is needed to observe the people's behaviors in the disaster area through social media.
This research aimed to explore the use of social media in the disaster area. This research used geolocation twitter feature to obtain tweets from a speciic area. This research is useful in practice and theoretic. In practical term, it explores the used of social media by the victims of the disaster, which encourages efective communication to the people or group outside the location. Theoretically, this research gives more detail understanding about the shared information, from the people in the disaster location.
For more detail, this research will categorize the use of social media by impacted disaster people. This research will be group into a framework, which is made by (Takahashi, Tandoc and Carmichael, 2015). This research also groups the user into some groups, such as news organization, individual, government, NGO, and others. Lastly, this research will also observe whether there is communication patern during or after the disaster happened.

Research Question
Social media is one of the platforms which enable two ways communication. People in the disaster location, can give up-todate information about the disaster condition to people outside the location (Takahashi, Tandoc and Carmichael, 2015). There are some researchers which atempt to explore the diferent of using social media outside afected area and inside the area, which are Binder, 2012;Miyabe, Miura, & Aramaki (2012); Takahashi et al., (2015); Vieweg, Hughes, Starbird, & Palen, (2010).    Miyabe et al., (2012) observed Twiter usage after earthquake 2011 in Japan.
Tweets were collected just after the disaster happened. The researcher analyzed, and grouped into various factors, in which location was one of the factors. The result of that research showed that people in the disaster area tent to use Twiter to communicate with other users, through a reply-based tweet. In contrast, users outside the afected area tried to share the disaster information by using retweet.
Furthermore, Vieweg et al., (2010) analyzed microblog post in Oklahoma Grassfires, which happened in April 2009, and the Red River Floods which happened in March and April 2009. The research focused on the disaster communication by people who were in the area of the disaster. The result of that research showed that Twiter users tend to share the same information such as evacuation, sheltering, animal management and damage and injury report.
The last is Takahashi et al., (2015) who tried to give depiction about the tweet, which was shared by the users in and outside Philippines during Typhoon Haiyan. The result of that research showed that people who were in Philippines tend to share the information with secondhand reporting type, coordinating relief and memorializing whereas people outside Philippines tend to share information about secondhand reporting, memorializing and coordinating relief. Both of the types are diferent in the percentage.
Geographic aspect becomes an important aspect to observe the behavior of Twiter users when the disaster happens (Takahashi, Tandoc and Carmichael, 2015). Unfortunately, four indings above have diferent results related to Twiter users in the disaster area, using Twiter to communicate disaster. It then encourages this research to try to analyze Twiter usage in a disaster communication, with proposing three questions as follow 1) What is information which is shared by Twiter users in disaster location?, 2) Who are Twiter users in disaster location?, 3): Is there any diferent information which is shared by Twiter users during and after the disaster?

Methods
This research consists of several stages. The irst stage in this study is data collection.
We employ scrapping technic from the site of htps://twiter.com/search-advanced to gather the data. The scrapping process is done by utilizing Python Tweepy libraries htp://tweepy. readthedocs.io/en/v3.5.0/api.html#tweepy-apitwiter-api-wrapper and Twiter API. Scrapping tweets are automatically done to retrieve tweet data with a speciied length of time. Before the scrapping process begins, this system will check the consumer key and access token, so that the system can retrieve data from Twiter, if the token data was listed in the Twiter system.
The system will then store the tweet data into the database. Tweet data will automatically be saved in the MySQL database and exported to a .csv ile.
Scrapping from the site of htps://twiter. com/search-advanced allows the researcher to gather the data in the speciic area. Twiter introduces geolocation feature to collect the tweet in particular area. We then used geolocation feature to give depiction more detail about shared information during and after the disaster. We then do a content analysis of the content shared on Twiter. By using the case of Flood in Garut Regency, Sumedang Regency, and Bandung City, we managed to collect 1083 tweets around the disaster area.
To launch our search, using some keywords like banjir (flood), bantuan (help), bencana (disaster), korban (victim), #banjir (#flood), #prayforSumedang #prayforBandung and #prayfor Garut. The system we created, allows the system to collect data in the form of date, time, account, tweet, bio, and location. In addition, we are assisted by Microsoft Excel 2013 to group the data. In more detail, 480 tweets from this research were gathered from the loods in Garut Regency, from 21 September to 29 September 2017. Furthermore, in the case of the loods in Sumedang District, we found 120 tweets from the date of 21-27 September 2017. Finally, in Bandung, we gathered 370 from 24-27 October 2017.
The collected tweet is then categorized into categories created by Takahashi et al. (2014). The Categories include Situation report disaster, from the personal perspective, secondhand reporting, request for help, coordination of rescue efort, providing mental cancelling, criticizing the government, express the hope and sympathy, discussing the cause of the disaster, and (re)connect community members. The detailed information of each category can be seen in Table 2. The data then presented into the form of percentage to ind out what information is shared most.
By using twiter biodata, we categorize users into several categories, such as community, government, media, celebrities, and journalists. This category is used to see the actors who use the most social media in the disaster area. We present twiter user data in a percentage form. It is intended to see the signiicance of the number of actors and the categories of information they share.
In addition, the authors also focus on the information shared by the community in the disaster location. Researchers remove tweets from government, media, celebrities, and journalists. The purpose of this separation is to observe the behavior of the communication of the community in the disaster areas.

Source: Research results
Finally, using the date and time of tweets, researchers categorize information into three categories including; during the disaster, one day after the disaster, and the next day. This is done with the aim to see the difference in communication behavior of the disasters during the disaster and the following days.

Disaster Proile
Researchers take tweet data in three looded locations namely, Flood in Bandung, Flood in Sumedang and Banjir in Garut regency. The selection of the three disaster sites are based on the severity of the disaster. Here is the proile of each disaster.

Flood in Garut
The rain that fell within 4 hours on September 20, 2016, in Garut regency has caused overflow the Cimanuk River and Cikamuri River. Heavy rain caused lash loods in the Garut regency. The lood that came with the mud lunged at 22:00 pm, so that some people were already sleeping. This lash lood hit 6 districts of Kota Garut, Bayongbong, Karangpawitan, Taraging Kidul, Taragong Kaler, and Banyuresmi. In addition to the high rainfall, lood loods occur because the Cimanuk river basin is in critical condition. This is exacerbated by a narrowing and silting of the river (BBC, 2016).
The lood also hit people's houses on the banks of the river. The status of the emergency response took efect until September 27th, 2016.
As of that date, 19 people were still missing. In addition, there were still many public facilities that had were not functioning. damage to homes and public facilities. In more detail, more than 1700 houses were damaged, from minor to moderate damage. Public facilities damaged were hospitals, schools and places of worship. More than 40 schools were damaged, causing about two thousand students, not in school. In addition to this, lood damage also resulted in 15 facilities worship damaged.
Reported by BNPB (2016), the landslide was also piled houses in Cimareme, Pasanggrahan Baru Village, Sumedang Selatan Subdistrict. Two people were found dead, and a person was suspected to be buried by avalanches. While in the Village, Baginda, South Sumedang, landslide piled up one house residents, one person was successfully evacuated in a state of death. Meanwhile, the 1.6-meter-high floods soaked six houses in Puyuh Mountain Village, Sukajaya Village, Sumedang Selatan District. As a result, 6 families were evacuated from their homes. In addition, 1500 people were sufering, and 1220 people were evacuated.

Flood in Bandung
On October 24, 2016, Bandung was hit by rain for two hours, from 11.30 to 13.30. The rain caused floods that reached more than one meter high. The worst flood point was in Pasteur region. The high and rapid low of the loods caused some vehicles to drift away (Tempo, 2016).
Based on the BNPB report (2016), loods caused hundreds of homes to be submerged. Some houses were damaged by eroded loods on the banks of Cilimus River. Flooding also damaged one educational facility. This lood also caused a person died. Moreover, a health facility was also reported broken by this lood

Findings
The first research question in this paper focuses on the use of social media by users in the disaster area. In the three case studies presented, the data show that the three most frequently shared information tweeted by Twiter users in the disaster area include: disaster event reports, based on user perspective, coordination of rescue efforts, and condolences. For the category of disaster report from the user's perspective, the highest percentage was found in the City of Bandung with 77% tweet used in this category, followed by Sumedang Regency with 36% percentage. The lowest percentage is occupied by Garut Regency which is only 1% adrift of Sumedang Regency, which is 36%.
The second most used social media function in the disaster area is coordinating rescue eforts. The results showed that 36% tweets were collected in Garut Regency used to coordinate rescue eforts. In Sumedang District, we found 34% of tweets were used to perform this function. Lastly, in the City of Bandung, there are only 3% tweets from netizens who use this function.
In the category of the expression of condolence that ranks third. In Sumedang District we found 17% of the total number of tweets gathered discussed about this function. Meanwhile, in Garut 14% tweet is talking about this information. Finally, in Bandung, this function is only used 4%.
Second-hand reporting in three case studies is not used singly. Overall, only 3% of netizens in these three regions use twitter to provide secondhand reports. In addition, requests for assistance are also not widely used by netizens in the disaster area. Only 3% of netizens in disaster areas use this function. Furthermore, the function of providing counseling and health assistance is used by 3% in all three areas. People in the disaster areas rarely use twiter to criticize disaster management eforts by the government.
This is due to the findings of the research, showing that only 1% of users in Bandung are talking about this function. In the other two cities, this category is not found. The last is the use of the function of social media to discuss the cause of the disaster. This function is used 2% by people in Bandung. Meanwhile, this function is not found in the case of loods in Sumedang and Garut regencies.
To sum it up, Table 3 explains that generally, there are three information which are often shared by netizens in the afected area, such as disaster report from users' perspective, coordinating relief effort, and express the sympathy. The highest level, is the disaster report from users' perspective, which reaches 49% from the whole tweets collected. Generally, tweets in this category are about weather report, and up-to-date condition in the disaster area. The second place is coordinating relief report, with the percentage of 24% from the total data. In this category, netizens used these tweets to collect and distribute help from the people. The last is expressing sympathy, with the percentage of 11%. In this category, netizens usually pray and hope that the disaster ends. In addition, some other social media functions that are used, but not very signiicantly include, secondhand reporting, assistance requests, providing counseling and medical assistance, criticizing the government, and discussing the causes of disasters.
By seeing twitter bio, the writers try to identify kind of users. So far, ive types of users participate in disaster. The users are as follow: society, government, media, celebrity, and journalist. The result of the research shows that users are dominated by local people, with the percentage of 96%. Other than that, there are media with 3%. Medias who participate are local medias of radio or local news organization.  Table 4 shows the users who participated in the use of Twiter in the disaster area. In all the three case studies, we found that users were dominated by the lay people. In Sumedang District, we found that the overall Twiter users in the disaster area were lay people, while in the city of Bandung 95% of the user is the lay people. Then, in the District of Garut, we found that 84% of the users were lay people.
We also ind the participation of 'local media,' such as local radio and local newspapers.
In the city of Bandung, we found 1% of users are the media, while in Garut district, 5% of users is the media. Other actors exist, but the numbers are not significant; among others, the government, celebrity, and journalists. We also ind the absence of some users in social media. In the district of Sumedang, we did not ind the presence of government, media, celebrity, and journalists in the disaster area.
Meanwhile, in Bandung, we did not ind any journalists. Finally, In Garut district we did not ind the presence of government, celebrity, and journalist in the disaster area. This study also shows the behavior of social media, use by lay people in disaster areas, that is presented in Table 5. We found that in Sumedang District, the function of social media is more widely used by clay people to report disaster events from the user's perspective. This function is found in 36% tweets. Diferent 2% with the previous function, coordinating relief efort ranks second, found as much as 34%. The last is the expression of sadness that ranks last. This category is found in 17% tweets of the entire tweet.
In the City of Bandung, most of the lay people use Twiter to provide reports on disaster situations from their perspective. A total of 71% of tweets we found, which were used for this category. Furthermore, 3% of users use Twiter to express sympathy and pray for disaster not to recur, and victims are given safety. Lastly, 1% tweet is used to coordinate rescue eforts. In Garut District, the majority of information shared by users is a disaster report from a user's perspective. A total of 11% of tweets were used for the category of the expression of condolences. The last category is the coordination of rescue eforts found as much as 2% of the total tweets. This is because the numbers are not signiicant, the writer just engages the lay people to look for information they shared during the disaster. The result of the research showed that the overall society lay persons in the disaster area, sharing disaster report from users' perspective with the percentage of 43%. The second place is coordinating relief efort with the percentage of 23%. The last is expressing sympathy with the percentage of 11%. Table 6, shows the use of Twiter during the disaster, the day after the disaster occurred and the next day. The irst day of the loods in Bandung, 17% tweets contains information about the report of the disaster from the user's perspective. The number increased to 37% on the second day and fell to 23% on the next day. The use of Twiter to coordinate rescue eforts was only found on the irst day of the disaster, and was not found on other days. Lastly, the use of Twiter to express sympathy, on the irst and second day is used 1%. The total number increases to 2% in the next day.
In Sumedang District, Twitter on the first day is more used to provide disaster reports and express sympathy, used 3% for each category. In this case, we found 1% of tweets used to coordinate rescue eforts. On the second day, tweet to give disaster report from user's perspective is equal to tweet to express sympathy, used 3% for each category. Then, ind that tweet to coordinate the rescue eforts used as much as 5%. On the next day, tweets for coordinating the eforts of the rescue of the victims, and reporting the disaster from the perspective of users have the same number of 30%. The rest is the tweet to express the sympathy used 9%.
The irst day of the loods in Sumedang district, the users use Twiter to report disaster situations from the users' perspective. On the same day, we did not ind the use of Twiter to coordinate rescue eforts, or express sympathy.
The next day, the use of Twitter for each category increased. On the second day, the use of Twiter to report disaster events from the perspective of users found 10%. Then, we found 3% tweets are tweeted to express condolences. In addition, the use of Twiter to coordinate rescue eforts used 3%. The number of tweets also increased from the second day to the next day. The use of Twiter for disaster reports from the user's perspective was found 41% of the total tweets. Then, the use of Twiter to express the condolences is found as much as 20%. Finally, the use of Twiter for the rescue coordination efort was found 18%.
By considering the time of Twiter usage, writers try to identify if there is different information, which has been shared each day.
The result of the research shows in the irst day of the disaster; users tend to use Twiter to report the incident of the disaster. Then, the information increases the day after. Twiter usage for coordinating relief efort in the irst day, the percentage is not signiicant, but it will increase for the day after. Likewise, the

Discussion
The research provides an empiric depiction of social media usage, from netizens in the afected disaster area. This research used category which was made by Takahashi et al., (2015). This research also tried to observe characteristic based on users' category (society, government, journalist, celebrity, and media) and the time of the usage (during the disaster and the days after). The result of the research showed that Twiter users in the afected disaster area, tend to use Twiter to report the accident of the disaster from users' perspective, coordinating relief efort and expressed sympathy. The result of the research is in line with the function of social media, which enable the users to create content and connect with other users (Blank and Reisdorf, 2012). Users can use the feature they have, to document and share information about the disaster, even when professional media and journalist are not in that event (Meikle and Redden, 2011). This research also conirms the inding of Alexander (2014), that social media can be used to monitor disaster situation. The inding in this research is not the same with the research of Takahashi et al., (2015). That researcher found that users in the disaster location tend to share secondhand information. It is because of Takahashi et al., (2015) used location in proile, to categorize disaster location; whereas this research used geolocation feature of Twiter.
Besides the sharing of the disaster information, Twitter users in the disaster location use Twiter to coordinate relief efort, which content of tweet to receive and distribute help, to the disaster victims. The result of the research is in line with the inding of Takahashi et al., (2015), which found that Twiter usage for coordinating relief effort was in second place after the secondhand report. Moreover, this result is also in line with Hughes & Palen, (2009) who found that users use social media to help recovery activities. The most shared information by users in the disaster area is memorialized about prayer and condolences, and hope that victims are given fortitude. Alexander, (2014) reveals that social media can be used to make people sympathy in certain things. People who are in disaster location feel that they are supported to face the future (Taylor et al., 2012).

Source: Research results
disaster situation, which then proves that users through their gadget make and consume the content (Meikle and Redden, 2011). From the aspect of time, this research finds that tweet for people in the disaster location; the numbers are increasing. It is suitable with the inding of Amanda L Hughes, Palen, Suton, Liu, & Vieweg (2008) who found that social media will be used after the disaster, as a way to help disaster victims. This research is not the same as Binder (2012) indings which observed that Twitter usage is not related to time. In contrast, this research inds that numbers of tweets will increase from time to time.

Conclusion
This research gives depiction about behavior communication for people in the disaster-afected area, through social media.
By using a framework made by Takahashi et al. (2015), this research inds that people in the disaster location use twiter to give irst-hand report, coordinate rescue efort and give help and express the grief. Then, the result of this research shows that by using geolocation, the depiction of Twiter usage is clearer than for bordering the location. By focusing on the affected area, Twitter used by lay people is usually found rather than other users. From the segment of time, the researcher inds a number of tweets that will increase each day. Users will share more information the days after rather than the day of disaster. The researcher is certain, that by observing the tweets which are from the disaster-afected area, it will result in a clearer information by not bordering data location.
This research also has some limitations. Researcher only focuses on 3 locations. The further researcher may be able to add location to get more data. Second, researcher neglects that local language that maybe it is still used on twiter. Yet, the researcher tries to use keywords to represent disaster communication in the afected area. Then, researcher can engage local language to collect more data.