Contribution of Social Media Analytics to Disaster Response Effectiveness: A Systematic Review of the Literature

: Disasters are sudden and catastrophic events with fatal consequences. Time-sensitive information collection from disaster zones is crucial for improved and data-driven disaster response. However, information collection from disaster zones in a prompt way is not easy or even possible. Human-centric information provided by citizen sensors through social media platforms create an opportunity for prompt information collection from disaster zones. There is, nevertheless, limited scholarly work that provides a comprehensive review on the potential of social media analytics for disaster response. This study utilizes a systematic literature review with PRISMA protocol to investigate the potential of social media analytics for enhanced disaster response. The ﬁndings of the systematic review of the literature pieces (n = 102) disclosed that (a) social media analytics in the disaster management research domain is an emerging ﬁeld of research and practice; (b) the central focus on the research domain is on the utilization of social media data for disaster response to natural hazards, but the social media data-driven disaster response to human-made disasters is an increasing research focus; (c) human-centric information intelligence provided by social media analytics in disaster response mainly concentrates on collective intelligence, location awareness, and situation awareness, and (d) there is limited scholarly research investigating near-real-time transport network management aftermath disasters. The ﬁndings inform authorities’ decision-making processes as near-real time disaster response management depending on social media analytics is a critical element of securing sustainable cities and communities.


Introduction
Catastrophic events occur following either natural disasters or man-made disasters in a disaster zone. Disasters cause not only economic losses to societies but also loss of lives in many cases. Over the past several years, the frequency of destructive disasters has dramatically increased, causing a huge amount of damage all around the globe [1]. This increasing frequency of disasters brings great challenges to the humankind [2], and these challenges, unfortunately, are expected to continue due to the changing global environment that triggers the disasters [3].
Disaster response, which refers to all actions taken by all parties during the aftermath of a disaster to alleviate the detrimental effects of the disaster [3], requires systematic efforts to analyse the consequences of the actions. Therefore, informed decision-making depending on data captured from a disaster zone to take corresponding actions is a must for disaster response. In other words, disaster response is a time-sensitive process that requires an immediate collection of the data from dispersed data resources in the disaster zone. However, often such an immediate collection of the data is not easy or even possible [4].

Challenges in Disaster Response
Disasters are sudden and catastrophic events causing a number of severe disruptions to society, and everyone across the globe has a risk to be exposed to a range of disasters. Disaster management is a process that starts with preparing the societies for a disaster before the disaster occurs, and it continues after the occurrence of the disaster to alleviate the detrimental effects of the disaster. Disaster management is an event cycle chain that consists of the following four phases, namely preparation, mitigation, response, and recovery [9]. Disaster response refers to the actions taken by all parties during the aftermath of a disaster to reduce the impacts of the disaster on humankind [3,9]. Although each phase is a crucial element in the disaster management cycle, this study focuses on disaster response because of its highly intensive and time-sensitive nature.
Disasters, especially natural hazards, usually bring not only a catastrophic event but also cascading effects following the main catastrophic event [10]. Therefore, disaster response is an intensive and complex process that needs to alleviate the adverse impacts of multiple cascading effects occurring in a specific location within a very limited time. Since prompt actions must be taken in disaster response, the information on the cascading effects must be gathered immediately for the informed decision-making, which is the first step of disaster response [10,11]. However, the information on the cascading effects of a disaster that occurred in each spatial boundary within a very short period is hard to gather, which is perceived as the main challenge in disaster response [12,13].
The difficulties in information sharing to unfold the cascading effects of a disaster directly affect the efficiency of disaster response. Disaster response starts with the informed decision-making that requires gathering the information from a disaster zone; and the first part of gathering this information is disseminating it from the disaster affected area. Therefore, this information dissemination must be at the maximum speed [12]. However, Sustainability 2023, 15, 8860 3 of 50 disasters are damaging and destructive to conventional information dissemination tools including emergency speed dial numbers and centralized emergency call facilities [11,13].
In addition to the absence of conventional communication tools in many cases, disasters usually affect a group of dispersed individuals corresponding to different locations, which hinders gathering the information from the multiple locations at different times to spread the information to the emergency services, authorities, and volunteers for disaster response. Consequently, there is an urgent need for a new technology that can gather the information about cascading effects happening in different locations and the information on victims that are in multiple locations, replacing the conventional communication tools. Therefore, the challenges in spreading this information promptly to the multiple parties can be achieved.
Disaster response is a collaborative effort in which multiple parties carry out individual different tasks within spatial boundaries and certain time periods [12]. Therefore, outsourcing the individual tasks of the parties in disaster response is location-dependent, and this outsourcing is always done under strict time constraints [12]. The assignment of these diverse individual tasks in specific locations within a very limited timeframe requires constant crisis communication. Although disaster response follows previously prepared solid crisis communication plans in theory, there are challenges and obstacles that disturb the crisis communication in a disaster zone because of the highly complex nature of cascading effects of the disaster [13]. To achieve the challenge of organizing constant and prompt crisis communication in disaster response, collective intelligence is required to provide collaboration and group efforts for disaster response in the disaster zones.
The geographic information of the cascading effects, following the main catastrophic event, and the victims are necessary information for disaster response. In many cases, disaster response teams need to be on site in different geographic locations at different times. The evacuation of the victims, saving their lives by taking timely actions, and the distribution of basic needs including food, water, and medication in disaster zones are the components of disaster response that require promptly gathered location information from the disaster zones. However, the wider the disaster zone is, the harder it is to gather the location information from the multiple sources in the disaster zone. Consequently, the disaster response teams need a data pipeline that can provide the teams with a constant geographic location awareness of cascading effects, the victims, and their needs so that the disaster response teams can reach the locations with the possible minimum disaster response time.
Disaster zones are severely damaged and risky areas, in which damage and risk assessment need to be conducted immediately [3]. A prompt damage assessment provides authorities with the information about the severity of a disaster; hence, the extent of the cascading effects of the disaster can be understood in a better and quicker way. Furthermore, this understanding enables the disaster response teams in the disaster zone to evaluate the risks surrounding them during the disaster response. As a result, the safety of the evacuation from the disaster zone and the continuity of the actions to save the lives and to distribute the basic needs in the disaster zone can be secured.
In addition to the disaster response teams, the victims in the disaster zone need to know the possible risk items in the disaster zone so that they can protect themselves in a better way and to assist the disaster response teams in the disaster zone. However, knowing the risk items in highly risky areas requires the authorities to conduct near-realtime assessment of the constantly changing situations in the disaster zone. This constant assessment of the changing situations in the disaster zone can only be possible by constant transfer of the near-real-time information on what has happened and what might happen from the disaster zone to outside of the disaster zone [1,4].

Materials and Methods
To understand the potential of utilizing social media analytics for the disaster response, a systematic literature review was undertaken to investigate (a) how the social media has evolved from being a multi-way communication tool to a crowdsourcing tool for disaster response and (b) how the social media analytics help the authorities and the community with informed decision-making to address the challenges in the disaster response. The systematic literature review mainly followed three stages, namely (a) planning, (b) conducting, and (c) reporting proposed by [36], and the conducting stage of the methodology further adopted the approach proposed by [37] following the PRISMA diagram as shown in Figure 1. social media data in regard to disaster and emergencies were included [38]. If the data type in the referred study was not clearly explained, the record was excluded.
The articles were categorized according to the challenges to overcome in disaster response utilizing social media data in the studies by identifying the research aims, the research methodologies, and the research outcomes.  Full text articles re-screened and assessed for eligibility (n=122)

Included
Records identified through the data base search.
(Title or Publication Title Contains the term/s (("disaster risk reduction" OR "disaster response" OR "emergency response") AND ("social media")) OR (Abstract/Summary Contains the term/s (("disaster risk reduction" OR "disaster response" OR "emergency response") AND ("social media")) Publication date range: January 2012 to January 2023  Firstly, in the planning stage, Google Scholar was used to have a general understanding of the challenges in disaster response and the potential of social media platforms as a crowdsourcing tool to overcome the challenges. Then, the planning stage involved developing the research aim, the research questions, a list of keywords, and the creation of the search string to conduct the literature selection. The research aim and the research question were framed to generate insight into understanding the potential of social media to overcome the challenges in disaster response relating to collective intelligence, location awareness, and situation awareness.
Then, a list of keyword sets was created to conduct a thematic search. While the study aimed at filling the research gap for disaster response in the disaster management cycle, the keyword of 'disaster risk reduction' was also added to the search string to extend the screening process to the studies that are related to any systematic effort to reduce the risks in a disaster zone after the disaster, which is also an integral part of the disaster response [8,12]. The keyword 'emergency response' was also added to the search string to extend the screening process over the studies that consider any systematic effort in emergency situations that occurred after a disaster in addition to the direct refences to the outcomes of disaster response actions taken after the disaster. The keyword "social media" was selected to exclude other crowdsourcing tools in order to create a better understanding of the role of social media in disaster response.
Secondly, the search string was organized on Boolean search line as follows: Title or Publication Title Contains the term/s: (("disaster risk reduction" OR "disaster response" OR "emergency response") AND ("social media")) <OR> Abstract/Summary Contains the term/s: (("disaster risk reduction" OR "disaster response" OR "emergency response") AND ("social media")) in order to conduct the initial thematic search. The initial thematic search was started with the identification of the references using the search string on a university's online library engine that covers 396 databases, including Directory of Open Access Journals, Science Direct, Scopus, Web of Science, and Wiley Online Library, which were used to complete the thematic search.
In view of the significant advancement in digital technology in the last decade, January 2012 was selected as the milestone for the identification process. Initially, a total of 867 references were detected. Then, only peer-reviewed journal articles, the articles written in the English language, and the articles that were available online were selected, excluding edited or authored books, conference proceedings, journal editorials, articles written in languages other than English, government and industry reports, and non-academic research. After this filtration, the number of the records was reduced to 421 references. Then, these references were imported to a reference manager software (EndNote X9) to double-check the duplicated records, and there was no duplicated record detected.
These articles were then "eye-balled" to ensure that the records were consistent with the thematic search, and the titles and abstracts were assessed against the research aim. This resulted in 183 articles left after the screening. The full texts of these articles were then read to determine the relevance of the selected articles to the research aim. After the first round of full-text screening, the results were reduced to 122 articles. Finally, these records were checked to see whether the type of the data captured by social media platforms, i.e., image, textual, and/or video, was clearly indicated in the study. After another round of full text screening, the total number of the records were reduced to 102 to categorize and analyse. At this point, it should be noted that only the articles that limited their data sources to sole social media data, excluding web-harvesting data, were included into the study. At this step, articles utilizing datasets such as CrisisMMD that contain only social media data in regard to disaster and emergencies were included [38]. If the data type in the referred study was not clearly explained, the record was excluded.
The articles were categorized according to the challenges to overcome in disaster response utilizing social media data in the studies by identifying the research aims, the research methodologies, and the research outcomes.
For each article, first, the aim of the article was identified, then each analysis conducted in the study was listed, separately; after that, the outcome of each analysis conducted was identified. The decision was made by investigating which challenges are separately targeted and which challenges are simultaneously targeted in the study. As the final step, categorization was made based on the combinations if there was a combination of separate challenges targeted in the same study. In this context, the research outcome was

General Observations
The selected articles, firstly, were analysed with regard to the research type, social media data platforms used in the study as the crowdsourcing tools, and the type of data captured by social media platforms in their methodologies.
The research type refers to the method used in the study to test the methodology adopted in the study. At this point, it should be noted that analytic and experimental studies shown in Table 1 utilized an analytical approach for the study aim and then tested the analytical approach using an experimental method in the same study. The experimental methods were usually virtual setups or scenarios created in the virtual laboratory environments to test the analytical approach used in the study. In other words, the five studies that made an 'active' effort to test their analytical approach were classified as "analytical and experimental" (n = 4.90%). In contrast to 'active' effort, six studies (n = 5.88%) adopted a 'passive' approach without testing their analytical approach using either an experimental method or a case study, and passive studies used only historic data. The dominant group in the selected literature used case studies to test their methodologies (n = 89.21%). 1 Some articles utilize several social media platforms. 2 Some articles use more than one type data as the input.
As shown in Table 1, Twitter was the most common used social media platform as the data source in the selected literature, and it was followed by the Sina Weibo social media platform which is called "Chinese Twitter" by [39]. It should be noted that some articles used more than one social media platform as the data source and more than one type of data as the input. The four different data types captured using social media platforms in the selected articles, i.e., textual, image, video, and numerical, referred to data types that were used as input in the methodologies of the selected studies. Textual data type was the most common data type used by the selected articles. Images and videos are the other data types used by the selected articles, and these visual data are starting to be utilized more often in recently conducted studies thanks to the new deep learning and machine learning algorithms [40]. Numerical data type refers to the data type used in the studies that gather only the number of tweets or posts without considering the content of the data.
It would be a strict limitation on this study that to include only the articles which studied disaster response to natural disasters in the literature, excluding the articles in the literature that focused on disaster response to man-made disasters. Although social media data have been excessively utilized for disaster response to natural hazards, the second group focusing on disaster response to man-made disasters represents almost a quarter of the selected literature; hence, they cannot be neglected. In the selected literature, the number of the studies focused on disaster response to natural hazards is placed at the top of the list with 78 articles (n = 76.47%), followed by social security events with 9 articles (n = 8.82%). Public health events and accidents are studied in seven articles (n = 6.86%) and in five articles (n = 4.90%), respectively. In the selected literature, three studies focused on disaster response to multiple disaster themes in the studies. Ref. [34] focused on 101 major disasters, under the identified disaster themes, that happened in China between 2010 and 2017. Ref. [41] focused on the utilization of Twitter accounts of Red Cross and Red Crescent National Societies for disaster response to different disaster themes and [42] focused on Queensland Fire and Emergency Services' Twitter and Facebook accounts for disaster response to different disaster themes. The rest of the 99 studies focused on disaster response to one type of disaster theme either using a case study (or multiple case studies) or analytical and experimental approach. Figure 2 summarizes the number of articles that used the case study method to analyse different types of disaster themes. Only 2 case studies, out of 91 case studies (n = 2.20%), focused on the disaster response to accidents; 9 case studies (n = 9.90%) focused on the disaster response to public health events; another 9 case studies (n = 9.90%) focused on the disaster response to social security events; and the remaining 71 case studies (n = 78%) focused on the disaster response to different natural hazards. The exponential increase in the number of natural hazard-case studies in the last decade started after Hurricane Sandy, which happened in 2012, and another drastic increase in the number of the natural hazard-case studies happened after Hurricane Harvey occurred in 2017. The first case study within the selected timeline in this study that focused on the disaster response to a natural hazard was [43], and after two years, [44] was published focusing on the potential of social media use for the disaster response to a social security event, which clearly shows that the great potential of social media analytics for disaster response to various disaster themes other than natural hazards was explored very shortly. COVID-19 is another reason for the increased number of case studies in the public health events. The simultaneous increasing trend in the case studies cross the four disaster themes after 2020 mainly stems from the increasing popularity of deep learning models in social media analytics that improves the analysis of the unstructured data captured by the social media platforms for disaster response [45].
The 102 selected articles were published in 68 different journals in multiple research domains, reflecting the applicability of the current social media platforms in disaster response. Table 2 lists the journals with a minimum of two articles published between 2012 and 2023. The International Journal of Disaster Risk Reduction secures a place at the top of the list with 12 articles followed by the International Journal of Geo-information with 5 articles. Furthermore, 48 journals published in different databases that comprise almost 45% of the selected literature contain at least one article that used social media as the data source with purposes of disaster response in different domains. mainly stems from the increasing popularity of deep learning models in social media analytics that improves the analysis of the unstructured data captured by the social media platforms for disaster response [45]. The 102 selected articles were published in 68 different journals in multiple research domains, reflecting the applicability of the current social media platforms in disaster response. Table 2 lists the journals with a minimum of two articles published between 2012 and 2023. The International Journal of Disaster Risk Reduction secures a place at the top of the list with 12 articles followed by the International Journal of Geo-information with 5 articles. Furthermore, 48 journals published in different databases that comprise almost 45% of the selected literature contain at least one article that used social media as the data source with purposes of disaster response in different domains.

Attributes of Social Media in Disaster Response
In this study, attributes of social media in disaster response were identified according to the challenges that were addressed in disaster response by utilizing social media data. Then, the attributes were categorized with regards to the aim and the outcome of the individual research article so as to provide a better understanding in the role of social media data used for overcoming challenges in disaster response rather than focusing on only the role of social media platforms as the information and communication tools for disaster response. Based on the identified patterns of the aim of the social media data utilization in the selected articles, it is concluded that the selected literature utilized social media data to overcome the challenges of collective intelligence, location awareness, and situation awareness in disaster response. These patterns align with the previously identified attributes of the conventional crowdsourcing tools (including open-source mapping platforms and map-mashups) for disaster response in the literature [16]. Appendix A Table A1 lists selected studies.
A few studies conducted analysis covering different attributes, while most of the selected literature focuses on one attribute. A study, for example [46], focused on only the identification of the location of the victims by using social media data according to the pattern of a hurricane, which refers to location awareness, while another study, for example [47], suggested the simultaneous evolution of the effects of the pattern and the severity of the hurricane, which refers to situation awareness. Because the latter case tends to create a disaster severity mapping methodology using social media data in which the location information is coupled with the damage severity assessment, the study extends its attributes beyond to location awareness [47,48]. Three attributes have seven possible combinations, and the distribution of the selected articles on each combination is shown in Figure 3. In total, 78 articles (n = 77%) focused on only one attribute whereas 24 articles (n = 23%) combined separate attributes. Table 3 lists the articles depending on the attributes and the combinations of the attributes according to disaster themes. It should be noted that the references focusing on more than one case study or scenario were categorized under multiple disaster themes in Table 3. Natural hazards are the dominant disaster theme in the literature regardless of whether the study considers only one attribute or a combination of the attributes.

Collective Intelligence
The utilization of social media to overcome the challenge of collective intelligence in disaster response is at the forefront of the selected literature. This is not surprising because the aim of any crowdsourcing tool is to address the challenges and obstacles in sharing information between different parties, and social media as a crowdsourcing tool is not an exception [13]. The easiest and quickest way to organize constant and prompt crisis communication in disaster response is to provide the volunteer mass information exchange between the parties using the social media tools [28]. Such high interest in providing a collective intelligence for disaster response, especially for disaster response to natural hazards, shows the necessity to organize large number of people through social media platforms to provide a collaborative disaster response. The literature [49] advocates that both the local volunteers in a disaster zone and digital volunteers can function as boundary spanners; hence, authorities can link the information from the disaster zone to external sources of information as a function of crisis communication. Furthermore, this communication helps the communities to express environmental and economic concerns as well as public frustration towards the authorities [49]. A recent study [50] proposed a study to explore whether demographic dimensions in a disaster-affected community can be understood using social media analytics, and the study concluded that collective intelligence provided by social media data enables the authorities to identify the more sensitive groups in the disaster-affected community. Furthermore, collective effort made by the disaster-affected public to disseminate the information allows the authorities to better understand the public's on-site needs after the disaster [51]. Similarly, the authorities can lead the disaster-affected public in a quick and prompt way by disseminating the required information to organize the disaster response on-site.
Collective intelligence can act as the control system of disaster response after a disaster by using the text classification systems to control the distribution of the resources needed by the public in the disaster zone [53][54][55][59][60][61]63,[68][69][70]72]. If there is any unfairness in the propagation of the resources in disaster response, it can be identified thanks to the collective intelligence provided by the disaster-affected public in the disaster zone [73], and reliability of actions taken for disaster response can also be identified by the participation patterns through the collective intelligence [74][75][76][77].
The current social media platforms including Facebook and Twitter provide the community with built-in collective systems including group applications and hashtags. These built-in collective systems increase interaction, which accelerates the collective intelligence after a disaster [42]. Given that this information flow must be checked to prevent the systems from misleading collaborative information [52], a group of articles have focused on the data quality and improving information dissemination ways to address the problem related to misleading collaborative information. Social media data can provide the community with the safest information dissemination, eliminating the misleading collaborative information by classifying the messages of the messengers [55], detecting misleading information by near-real-time cross-checking the information against other sources [57], and mapping the communication networks [59].
COVID-19 has emerged as a major research topic in many research domains as it is a public health event that affects everyone on the globe without exception. A recent study [79] explored the characteristics of COVID-19 patients using data-mining methods on social media data, and [80] provides a case study to understand the characteristics of TikTok users and the effects of these characteristics on information sharing via social media with regards to COVID-19. Additionally, Refs. [81,82] conducted sentiment analysis to understand public opinion on how health agencies respond to the COVID-19 disaster. The literature shows that social media data can be utilized to not only understand the authorities' success in disaster response to COVID-19 but also to understand the affected people's feelings and characteristics.
Large-scale accidents and social security events can have disastrous consequences; hence, they are perceived as man-made disasters in the literature [34]. Ref. [49] aimed to understand the public sentiment after the Deepwater Horizon oil spill, an accident that happened in the Gulf of Mexico in 2010. The study conducted content analysis and sentiment analysis of the content and flow of the tweets to show how the flow accelerated disaster response by providing higher collective intelligence. Similarly, Refs. [44,83,85] used social media analytics to understand how the collective intelligence evolved into the public sentiment in the aftermath of mass shooting events in Kenya, India, and Brussels-Nice-Paris, respectively. In addition, Ref. [84] conducted content and sentiment analysis to evaluate the public opinion on the Syria Chemical attack in 2017. In summary, the public sentiment after a disaster can be captured by collective intelligence in not only social security events but also natural hazards including earthquakes [62], floods [64,65,71], bush fires [78], tsunamis [66], and typhoons [67].

Location Awareness
Disaster response requires accurate spatial information to locate the cascading effects of a disaster and the victims [86]. Conventional technologies to gather location-based information after a disaster, including radar satellites, aerial tools and equipment, and types of aircraft, are prone to atmospheric conditions [86]. Furthermore, these conventional technologies are complex, and they require skilled human resource to operate. These challenges lead to the need for new technologies that have the capacity to capture the human-centric location-based information under any atmospheric condition without requiring skilled human resource. Social media platforms can provide human-centric location-based information intelligence to address the challenges in gathering location-based information data after disasters to provide improved and data-driven disaster response. Furthermore, the human-centric location-based information provided by social media data is accessible to anyone, and it can be used by many parties simultaneously, which accelerates the speed of disaster response [38].
There are two ways to gather human-centric location-based information intelligence via social media data: (a) geo-located data and (b) toponyms [46]. Social media data have proven capacity to provide human-centric location-based data through application programming interfaces (APIs) of social media platforms in the form of geo-located data [87,88]. Additionally, textual information without using any form of geo-located data but using toponyms can be used as volunteer geographic information to identify the specific locations in the disaster zone. As a result, the disaster response teams can identify the specific locations of the cascading effects of the disaster, victims, and on-site needs to provide better disaster response where the geo-located data are not sufficient [46].
The near-real-time mapping of disasters' footprint through social media data with live updates improves the disaster response time [86]. While open-source mapping platforms such as Google Maps, Open Street Map (OSM), Bing Maps, Yahoo Maps, and map-mashups including Victorian Bushfire Map, Queensland Globe, and Flood Awareness Online can provide the community with near-real-time interactive maps through crowd-mapping, these interactive maps require skilled workforce to process the data and to visualize the location-based information on the interactive maps for the end-users [46,87,88]. Therefore, the propagation speed of location-based information heavily depends on not only the reaction time of the authorities to process the location data but also on the speed of gathering the location data using conventional crowdsourcing tools [87].
On the other hand, the propagation speed of social media provided human-centric location-based information is much faster compared to the propagation speed of locationbased information using conventional crowdsourcing tools. Moreover, Ref. [88] advocated that the biggest advantage of the utilization of social media for raising location awareness is that an individual is not required to be physically present in a disaster zone to propagate the location-based information. Indeed, virtual volunteers can disseminate the human-centric location-based information captured by social media without physically observing the disaster zone. It helps to accelerate the speed of disaster response by communicating the human-centric location-based information to a wider community that is not limited to the people who are able to use open-source mapping platforms and map-mashups.
The timeline of the geographic distribution of the distractions caused by a disaster can be tracked by the disaster response teams to differentiate the hotspots from the other locations in each timeline in the disaster zone thanks to social media analytics. Therefore, spatio-temporal data captured via social media are crucial for predicting the mobility in the disaster zones [89]. Furthermore, deep and machine learning algorithms that can process spatio-temporal data captured via social media can provide the disaster response teams with an automated mapping systems that are crucial to near-real-time mapping to project the evolution of the disaster [40]. Network intelligence is another important factor to bring the disaster response teams and disaster-affected public's on-site needs to the disaster zone, which can be projected by near-real-time mapping thanks to deep learning models that use human-centric location-based information gathered by social media [45].
The avoidance of misleading collaborative information can be achieved by not only collective intelligence but also location awareness [52]. Therefore, near-real-time mapping is crucial to provide improved and data-driven disaster response, which is a time-sensitive process, because any misleading information on the social media platforms during the disaster response can be detected and eliminated by near-real-time mapping. Volunteered geographic information in the geo-located data can provide this safeness by checking the geolocation of the information depending on the spatial properties [86]. Geo-tagged picture messages posted on the social media platforms can identify the disaster zone and the characteristics of the disaster-affected people and compare them against the information [90]. Thus, integrating Geographic Information Systems (GIS) into near-real-time mapping by using social media data will help check the accuracy of the information on the social media platforms [40].

Situation Awareness
Disaster zones are chaotic areas where the constantly changing situations cannot be foreseen by the disaster response teams because the damage assessment of constantly changing situations in these chaotic zones depends on many constantly changing fac-tors [121]. The constantly changing situations in the disaster zones are hard to assess if depending on only conventional data gathering tools including radars and satellites [96]. Furthermore, natural disasters are sudden and catastrophic events that happen in very large spatial areas within a very limited time [98]. Therefore, conventional data gathering tools or conventional crowdsourcing tools are not capable of providing the authorities and the disaster response teams with the spatio-temporal information on risk items they will face in the disaster zones [96,98,99]. Furthermore, remote sensing techniques to gather the data from disaster zones are prone to the atmospheric conditions, which hinders the prompt data gathering from the disaster zones. However, the human-centric on-site near-real-time information on the changing situations in the disaster zones gathered by social media will facilitate prompt assessment of the risk items corresponding to the changing situations in the disaster zones.
Ref. [96] proposed an approach to supplement the satellite images on a volcano with social media data with the purpose of disaster risk prediction to provide improved and data-driven disaster response to the Taal Volcano eruption that happened in the Philippines. The results of the study prove that (a) social media data can be integrated with external data for a quick and cost-efficient disaster damage assessment over a very wide spatial area and (b) situation awareness provided by hybrid data including social media data leads to faster disaster response because the disaster response teams can utilize the real-time statements to identify the ashfall area with the severity after the volcano eruption. Another study [54] compared social media as a crowdsourcing tool to a conventional crowd mapping tool, and the study concluded that social media is superior to the conventional crowd mapping tool in terms of raising situation awareness in a disaster zone because of its capability of gathering the data directly from the disaster-affected population.
In the selected literature, a number of studies tested the potential of the data captured by social media to assess the severity of a disaster. Ref. [94] proposed a new methodology to conduct near-real-time intensity assessment of the disaster-affected public after Typhoon Haiyan. In the study, an index, Normalized Affected Population Index (NAPI), to leverage social media data for the disaster severity assessment was created to provide more timely and accurate disaster information for the disaster response teams. Ref. [98] tested to create a Mercalli intensity scale by using social media data that is used to express the intensity of an earthquake's damage.
The advantage of using social media data is that it provides the authorities with nearreal-time damage assessment whereas a Mercalli intensity scale report can take days to be prepared [98]. The study concluded that the data captured by social media successfully creates a rapid situation awareness in the intensity of the earthquake. Ref. [99] created a conceptual decision-making framework depending on the social media data to create situational awareness for emergency management. The framework was tested in two different natural hazards, and the study concluded that only 2% of the social media data captured during a disaster is enough to create the conceptual decision-making framework for raising situation awareness [99].
The literature [39,113,114] utilized social media to create a near-real-time situation awareness tool for natural hazards, social security events, and public health events, respectively. Ref. [39] utilized social media data for real-time disaster damage assessment for not only an aftermath of a natural hazard, Typhoon Nepartak, but also the aftermath of a social security event, the Tianjin explosion. Ref. [114] created a framework for people who were at the airport during the Fort Lauderdale Hollywood airport shooting. Ref. [113] created a new dissemination pipeline as an alternative model channel to provide the society with situation awareness during Ebola. These studies concluded that the social media platforms can be successfully utilized as near-real-time situation awareness tools to inform people in the disaster zones while the disaster is happening.
Ref. [100] utilized textual data for near-real-time damage assessment by defining a text-based rapid damage assessment framework for an earthquake aligning with Ridgecrest earthquake sequences; Ref. [101] utilized location data captured by social media to prepare a novel model, which is called the spatial logistic growth model, to evaluate the spatial growth of citizen-sensor data after an earthquake. These studies prove that social media data can be combined with external damage assessment indexes or create an individual index to assess the intended objective function in terms of damage assessment to create near-real-time rapid damage assessment models.
The near-real-time damage assessment of flooding can be conducted based on different parameters using the data captured by social media, and the flooded area can be monitored after the flooding depending on the same data and the same parameters to provide a constant assessment for the constantly changing situations in the flooded areas. Ref. [104] utilized social media data to evaluate flood inundation probability, and [105] proposed a model for waterlogging using social media data. Ref. [106] harvested social media data for flood map generation, while [47] created flood severity map using social media data combining images with text. Social media data that contain two different types of data, namely images and texts, increase the accuracy of damage assessment models for flood damage assessment [109].

Findings and Insight
The study investigated the potential of social media utilization for disaster response to address the challenges in disaster response. The study offered a novel contribution to the literature by unfolding the potential of social media analytics in disaster response, extending the notion of social media being only a two-way communication tool in disaster response. The key findings of the study help improve the understanding of the current state of the literature and propose a future research agenda. The increasing academic interest in the utilization of social media for improved and data-driven disaster response in the last decade is expected to continue due to the increasing popularity of deep learning and machine learning algorithms that can analyse the unstructured data and process different data types such as texts and images simultaneously.
The utilization of social media for disaster response to natural hazards is the dominant research theme in the selected literature, while the number of studies that utilize social media analytics to address the challenges in disaster response to man-made disasters is increasing. The human-centric near-real-time information provided by social media has already become an integral part of disaster response, which is reflected in 91 case studies conducted focusing on the disasters that happened in the last decade. Considering this fact, social media has the potential of becoming a major alternative data pipeline to the conventional crowdsourcing tools for disaster response that can be used by not only government agencies, emergency authorities, and disaster response teams but also individuals in their daily lives.
Social media platforms are publicly available and usually free of charge platforms, which makes them widely adopted by individuals. This wide adaption of social media platforms leads to massive amount of information exchange, which will facilitate gathering the human-centric information from multiple parties and citizen sensors, disseminating the information to multiple parties, and storing the data in their data banks. The APIs of the social media platforms enable the researchers, data analysts, and emergency managers to harvest and to analyse the data [91]. Twitter is the most used social media platform in the selected literature because of its wide adoption by individuals and free API that provides the researchers with the free data [91].

Challenges, Opportunities, and Practice Implications
The challenges in disaster response stem from the fact that disasters are sudden, time-limited, and catastrophic events that usually affect multiple locations with different severity. In other words, disasters have spatial and temporal boundaries, and the challenges in disaster response are related to information dissemination from the disaster zones within a very limited time. In many cases, conventional data-gathering technologies including remote sensing techniques, radars and satellites, and conventional crowdsourcing tools including open-source mapping platforms and map-mashups are not capable of disseminating the information in a prompt way for data-driven disaster response. On the other hand, human-centric information gathered by social media platforms can be disseminated from multiple parties to multiple parties in a prompt way, which improves the disaster response. Furthermore, it is the promptest way to gather the information from the disaster zones and disseminate the information because the human-centric information gathered by social media is not prone to atmospheric conditions, and the dissemination of the information does not require a skilled workforce.
The utilization of social media to address the challenge of organizing a constant and prompt crisis communication for disaster response leads to collective intelligence provided by social media for disaster response. Through the near-real-time information exchange in collective intelligence during the disaster response, collective intelligence enables emergency managers and disaster response teams (a) to understand the public opinion about disaster response and the social dimensions of the disaster-affected public, (b) to eliminate the misleading collaborative information dissemination during disaster response, and (c) to monitor the propagation of the disaster response actions and the resources needed by the public in disaster zones.
The challenge in gathering location-based information from the disaster zones by using conventional methods can be overcome through adopting human-centric locationbased information intelligence. This intelligence, location awareness, is provided by social media platforms in the form of geo-located data and toponyms. Given that geo-located data might be insufficient, as shown in some of the selected literature, toponyms have been utilized as volunteer geographic information by many researchers using various text classification techniques and deep learning algorithms. Location awareness provides emergency managers and disaster response teams with the information on the specific locations of cascading effects of a disaster, victims, and on-site needs in the disaster zone. Consequently, (a) the near-real-time mapping of disasters' footprint, (b) identification of the hotspots in the disaster zones with regards to spatial-temporal properties of a disaster, and (c) detection of collaborative misleading information on social media platforms are the sub-challenges in disaster response that can be achieved through location awareness.
The damage assessment of constantly changing situations in disaster zones is hard to conduct, which hinders the emergency managers and disaster response teams from being provided with the information about on-site risk items in the disaster zones. The human-centric on-site near-real-time information on the constantly changing situations in the disaster zones gathered by social media can provide situation awareness. The situation awareness enables the researchers (a) to combine the data captured by social media with external supplement data in order to create near-real-time damage assessment data pipelines, (b) to create near-real-time damage severity mapping by utilizing the realtime statements, (c) to create novel models for damage assessment that aligns with the existing damage assessment frameworks, and (d) to monitor the situation after a disaster, creating a near-real-time situation awareness tool that is capable of assessing the constantly changing situations in disaster zones depending on constantly changing parameters.

Research and Future Outlook
Despite the growing popularity of social media platforms in our daily lives, there is, unfortunately, no systematic implications of social media analytics in the disaster response. Despite a very wide range of applications in different disciplines including natural hazards such as flooding, earthquakes, tornados; public heath events such as COVID-19; accidents and social security events including mass shootings, no previous scholarly published work has been conducted considering all these perspectives. This research fulfills this research gap.
The fast-growing social media analytics literature is heading toward identifying the potential of deep learning and machine learning algorithms that support the social media analytics in disaster response. This potential is expected to inform disaster managers, engineers who design water and transport infrastructures, and urban planners to plan better disaster response activities considering the different perspectives of disaster response. Therefore, the decision-making processes, disaster response guidelines, infrastructure design guidelines, and urban planning processes can be assessed using social media analytics according to the previous disasters.
The study suggests that future research is need for an improved and data-driven disaster response by utilizing the data captured by social media platforms. In the selected literature, Ref. [3] created a framework to a near-real-time damage assessment for road damage scenarios, combining textual data with images; Ref. [91] utilized social media data to assess near-real-time traffic incidents that might cause an emergency, and Ref. [92] utilized social media data to evaluate near-real-time traffic infrastructures' damages. While the transport network intelligence is crucial in delivering the disaster response teams to disaster zones, evacuating the victims from disaster zones, and delivering the on-site needs to disaster zones, there is limited research focusing on near-real-time transport network intelligence management considering the transport network performance losses.
Furthermore, as the travel time changes after a disaster due to the constantly changing new conditions on the road segments and failure of the critical infrastructures in a transport network, future research on near-real-time transport network intelligence management utilizing human-captured information provided by social media is needed to improve the disaster response. For instance, in the selected literature, many studies focus on flood mapping using social media analytics after a flooding or typhoon. However, these studies failed to map the distractions in the transport network caused by the flooding that affect the accessibility index of the locations, the most needed by the public after a disaster, including hospitals. Therefore, the near-real-time transport network intelligence methods and techniques after disasters need to be improved to build more sustainable cities, improving the resilience of cities against disasters, especially natural hazards. For future research direction, this urgent need in the research domain can be filled with a design-led qualitative research approach.

Limitations and Future Directions
The study has the following limitations: (a) conference proceedings, book chapters, and white papers were excluded; (b) selected search keywords could have omitted some relevant articles; (c) authors' unconscious bias could have an influence on the findings; (d) although the paper covered disaster response and disaster risk reduction, the review did not specifically focus on these areas; (e) the methodology was a manual literature review technique and did not include techniques such as cognitive mapping and concept clustering. Despite these limitations, the research results shed light for the way forward to provide better understanding of the current literature and potential of social media in disaster response for a future research agenda.     Notes: n/a = not available as not a case study identified in the article; U.S = The United States of America; QLD = Queensland, Australia; C = Collective intelligence; L = Location awareness; S = Situation awareness CL = Collective intelligence + Location awareness; CS = Collective intelligence + Situation awareness; LS = Location awareness + Situation awareness; CLS = Collective intelligence + Location awareness + Situation awareness.