Managing social media uncertainty to support the decision making process during Emergencies

Abstract

Recent emergencies have shown the positive impact of using social media and social networks for communicating and exchanging information. Citizens and authorities can make safer decisions during emergencies based on the real-time information available on social media. Decision-making starts with information gathering and social media provides the opportunity to inform multiple citizens at once. However, message and source uncertainty can place emergency stakeholders in a risky position, as it is not always possible to know if messages are accurate, rumours or even malicious. Current approaches for social media information verification focuses on technical resources like analytical packages. Little research has been developed to provide citizens and field workers with tools to evaluate social media information. This paper presents research in progress for developing a verification framework – for all emergency stakeholders – to support their decision-making process by managing social media uncertainty during emergencies.

Keywords:

• Social media
• decision making process
• verification
• emergencies stakeholders
• emergency management

Introduction

Decision Making is ‘a mechanism for making choices at each step of the problem-solving process’ (FEMA, 2005, p. 2-1). It results in the selection of an action among several alternative possibilities. This is a critical skill during emergencies when it is paramount to make timely sound decisions to reduce human and material losses (Kowalski-Trakofler, Vaught, & Scharf, 2003).

The cost of emergencies and disasters can be really high. The 2015 storms in Ireland, Desmond and Frank, had an estimated cost of €100 million for repairing the roads after the floods (Pope, 2016). As of December 2015, US$1.62 billion was mobilised by the World Bank Group to finance the Ebola response and recovery efforts (Chavez, 2015). In the twentieth century until 2014, it is estimated that hurricanes cost an average of US$5 billion in property loss each year in the USA and have accounted for the deaths of 587 U.S. residents (Hurricane Research Division, 2014).

Decisions focus on reducing the impact of a disaster as well as saving lives. During emergencies, most decisions are the products of complex interactions (Patterson, Weil, & Patel, 2010) with sometimes faulty or incomplete information (Kowalski-Trakofler et al., 2003; Tapia & Moore, 2014). There are multiple models used to explain the decision-making process. Traditional theories include prescriptive and descriptive models of analytical decision-making, which are based on the evaluation of options in the decision-making process (Elliott, 2005). However, in dynamic, complex and uncertain scenarios where group decision may be required, the analytical approach may not be suitable. As an alternative, the naturalistic decision-making (NDM) school introduced the concept of ‘experience’. NDM theories focus on ‘how people use experience to make decisions in naturalistic environments (e.g. under time pressure, shifting conditions, with unclear goals, degraded information and within team interactions)’ (Elliott, 2005, p. 8). However, not all emergencies stakeholders are experts in emergencies. Therefore, communication channels enable the connection between non-experts decision makers with more informed decision makers.

A relatively new information communications technology (ICT) used in emergency management is social media. Social media is a group of internet-based applications that allow citizens and authorities alike to create, post and exchange content (Kaplan & Haenlein, 2010). These applications have multiple benefits. For example, they have been used to help identify and rescue disaster victims (Merchant, Elmer, & Lurie, 2011). A more extended use of social media is as a mass communication channel, in order to inform large numbers of stakeholders at once (Ki & Nekmat, 2014; Muralidharan et al., 2011; Stříteský, Stránská, & Drábik, 2015; Utz, Schultz, & Glocka, 2013). Additionally, it provides a unique opportunity to collect and make use of large amounts of real-time data (Big Data), which is paramount in decision support systems and any emergency management decision-making process (Gaynor, Seltzer, & Moulton, 2005; Kamel Boulos et al., 2011; U.S. Department of Homeland Security, 2014). However, data accessibility is not enough to improve key decision-making processes.

Data quality truly impacts stakeholder decision-making (Chengalur-Smith, Ballou, & Pazer, 1999). In 2013 during the ‘Big Data and Disaster Management JST / NSF Joint Workshop’, veracity¹ was declared as one of the biggest challenges for the correct integration of big data within emergency management. Ideally, verification should be achieved in real-time to support decisions made during emergencies (Gaynor et al., 2005; Kamel Boulos et al., 2011). Due to limitations on resource availability, the importance of real-time data verification increases after the emergency takes place; and therefore during the response phase (Computing Community Consortium, 2012; Radisch & Jacobzone, 2010).

This paper presents research in progress that focuses on the data quality concept. It presents the development of a framework to manage social media uncertainty to support the decision-making process during emergencies. Section 1 provides an overview of the challenges of using social media as an information and communication tool. It identifies two key challenges associated with data verification and information quality. Section 2 explores current approaches to verify social media information with a description of the difficulties of implementing these approaches for different stakeholders. Section 3 proposes three research questions to support the development of the information verification framework. Section 4 introduces the planned activities for the proposed research and how this are aligned with design research methodology. The final section is the summary and conclusions of this paper.

Challenges of using social media as information communication technology

Social Media has been used in multiple events including hurricanes, such as Sandy, Isaac, Katrina, Hugo and Rita (Belardo & Harrald, 1992; Huang & Xiao, 2015; Kapucu & Garayev, 2011; Tinker, 2013) and flooding, like in Queensland and Saxony in 2013 (Ehnis & Bunker, 2012; Peters & Paulo, 2015). However, its use is not without its considerable challenges.

Five potential policy issues for the use of social media in the context of emergencies and disasters are: 1) administrative cost considerations, 2) privacy issues, 3) technological limitations, 4) accurate information and 5) malicious use of social media during disasters (Lindsay, 2011). These last two challenges are shared by all emergency stakeholders: citizens and organisations alike during the decision-making process. For example, misinformation was an issue in the Boston Marathon bombing in 2013, where the wrong suspects were pointed out in Social Media (Golgowski, 2013; Hughes, Palen, & Peterson, 2014; Petrecca, 2013). This impacted on citizens and emergency responders alike by disrupting suspects’ lives and allocating resources to monitor the wrong suspects.

Inaccuracy of information and rumours

Inaccurate information and rumours have been reported as a challenge by multiple researchers and emergency responders (Castillo, Mendoza, & Poblete, 2011; Mendoza, Poblete, & Castillo, 2010; Oh, Agrawal, & Rao, 2013; Procter, Vis, & Voss, 2013; Terpstra et al., 2012). Misinformation can spread really fast in social media platforms (Lukasik, Cohn, & Bontcheva, 2015; Takayasu et al., 2015). This challenge impacts on the allocation of resources by first responders and citizens’ decisions before, during and after emergencies.

For example, in March 2011 after the Japanese earthquake and tsunami, tweets for assistance were ‘retweeted’ after the victims had been rescued (Acar & Muraki, 2011). In another example during the Japanese earthquake, the tweet:

Please spread: To those people who live close to the east shore of Tokyo Bay! Due to the explosion of oil tanks, harmful chemical materials may fall with rain soon. Bring your umbrella and rain coat with you to protect your skin from the dangerous rain!!

However, this was a rumour without scientific basis (Chen et al., 2015; Takayasu et al., 2015). Other examples can be found after Hurricane Sandy in October 2012, where rumours circulated on social networks about paid volunteer opportunities and reimbursements for survivors (Jacobs & Tuohy, 2012).

Malicious use of social media

Malicious use of social media can create additional instability during terrorist attacks, by congregating people in a location and having a second attack at that location (Gao, Barbier, & Goolsby, 2011; Lindsay, 2011; Weaver, Boyle, & Besaleva, 2012). An example of this is the use of social media during the Mumbai attacks in 2008, with the following message was posted during the incidents (Oh, Agrawal, & Rao, 2011):

RT @celebcorps remember when tweeting details that it is CONFIRMED terrorists have satphone (satellite phone – authors added) access to net sources (1:50 AM Nov 27th, 2008 from Ubiquity)

It is possible to observe that these two challenges are highly related to social media data quality. Thus, the next section evaluates potential solutions proposed to overcome this data verification problem.

Social media information verification approaches

In the literature, social media information verification has been mainly explored from a journalist perspective (cf. Heravi & Mcginnis, 2013); however, in recent years emergency researchers have started to develop different methods to filter, classify, analyse and verify this data stream. There are two main approaches to data verification (Table 1), differentiated based on: 1) Intrinsic properties and 2) Extrinsic properties.

Table 1. Intrinsic and Extrinsic Social Media Information Verification Approaches (Literature Review).

Data Verification using Intrinsic Properties
Approach	Description	Authors Sample
Content analysis and post behaviour	Analysis of content posted in Social Media. Content features include: length of tweet, num of words, emoticons, sentiment words, multimedia content, mentions, hashtags, URLs	Text Analysis (Canini, Suh, & Pirolli, 2011; Castillo et al., 2011; Oh et al., 2013)
		Image (Boididou, Papadopoulos, Kompatsiaris, Schifferes, & Newman, 2014; Gupta, Lamba, Kumaraguru, & Joshi, 2013)
User information analysis and behaviour	Analysis of information related to the user including num of followers, follower-friend ratio, num of times listed, user is a verified user, profile image, num of tweets, num of URL in messages	(Boididou et al., 2014) (Stringhini, Kruegel, & Vigna, 2010)
Network behaviour	Analysis of reaction from other users: likes, shares, reported as spam…	(Mendoza et al., 2010)
Data Verification using Extrinsic Properties
Approach	Description	Authors Sample
Expertise	User uses his/her expertise or authoritative sources to verify the message.	(Martin & Comm, 2014)
Crowdsourcing / human sensors	Verification is achieved by involving large groups of users to evaluate information quality	(Chamales, 2013; Popoola, Krasnoshtan, Toth, Naroditskiy, Castillo, Meier, & Rahwan, 2013; Heravi & Mcginnis, 2013; Imran, Castillo, Lucas, Meier, & Vieweg, 2014; Kamel Boulos et al., 2011; Ortmann, Limbu, Wang, & Kauppinen, 2011; Poblet, García-cuesta, & Casanovas, 2014; Schulz, Paulheim, & Probst, 2012; Weaver et al., 2012)
Linked data	Integration of other data sources and social media data in order to verify information.	(Crowley et al., 2013; Daume, Albert, & von Gadow, 2014)

The intrinsic approach analyses properties of a specific Social Media platform. Properties considered include content in multimedia format, user information, timestamp and geographic information. For example in a journalism and enterprise perspective, ‘SocialSensor’ and ‘REVEAL’ FP7 projects explored the possibility of analysing the validity of tweets or authors based on a 3 ‘C’s framework: Content, Contributor and Context analysis (Corney et al., 2014; Middleton & Gottron, 2014). In this approach interaction with other systems is for an analytical purpose and not for verification.

The extrinsic approach relies on using additional resources to verify the information. Three main approaches are identified: experts, crowds, and machines (Kamel Boulos et al., 2011). For example, in crowdsourcing the input from a third person is paramount to identify the veracity of the information.

The approaches described in Table 1 have been developed using analytical packages and machine learning algorithms. These approaches require technical resources that may not be available to all citizens during emergencies. Therefore, additional research is required to provide a framework that provides guidance to technical and non-technical emergency stakeholders.

Research objectives

The objective of this study is to create a social media message verification framework to support emergency stakeholders’ decision-making processes during the response phase of any emergency (small, medium, and large scale). It will provide a suitable environment to find the most appropriate combination for data verification in different emergency management scenarios.

This objective is operationalized through three research questions:

(1)	How have social media been used in past emergencies during the decision-making process?

Extensive research has been conducted on the positive impact of social media and online social networks on information analysis approaches and the intelligence phase of decision making, in general (Antunes & Costa, 2012; Chen, Storey, & Chiang, 2012; Crowley, Dabrowski, & Breslin, 2013; Jindal, Sharma, & Sharma, 2014; Rupnik & Kukar, 2007) and in the emergency management field (Gaynor et al., 2005; Imran & Castillo, 2015; Sullivan et al., 2013; Temnikova, Castillo, & Vieweg, 2015; Temnikova, Vieweg, & Castillo, 2015). By applying a systematic analysis of the usage of social media in past events, this study will analyse the challenges faced by emergency stakeholders using this platform for decision making.

(2)	How does the data available in social media impact on the decision-making and verification process during an emergency?

Social Media (e.g. Twitter) provide real time data for monitoring and analysing the impact of emergencies (Huang & Xiao, 2015; Imran & Castillo, 2014; Imran et al., 2014). This research question aims to gain a better understanding of how multiple formats available in Social Media (text, video, images, sound, GPS) and users’ information in online communities impact on the perception of message credibility and information trust. Furthermore, it aims to explore the steps that are taken to verify the information, and how processes and data available impacts on the decision making process. The answer of this question will guide the development of the verification framework and its evaluation.

(3)	How can a verification framework be implemented in online and offline decision support systems during emergencies?

Technology advances and its availability² – including dedicated sensor networks and multi-purpose sensor networks (Sullivan et al., 2013) – allow the collection of a large volume of both structured and unstructured data (Big Data) during emergencies. Information can be obtained from different sources. For example, the “Global Disaster Alert and Coordination System” managed by the European Commission Joint Research Centre provides alerts and impact estimations after major disasters through their website. Information obtained through social media can be verified using more traditional channels including first responders, official reports, weather forecast and existing systems already in place. This research question aims to analyse how the developed framework can be used during emergencies and in conjunction with existing systems.

Planned activities

Research questions will be addressed using a design science research methodology. Design science aims to devise artefacts to achieve specific goals (Simon, 1996). Hence, the goal of this research project is to develop a framework to reduce social media uncertainly during emergencies. Table 2 aligns design science methodology with each research question and describes the process for gathering and analysing data.

Table 2. Design research methodology alignment with research in progress.

Problem definition will mix information obtained from the literature review and stake-holder feedback. After each iteration the problem definition will be review and adjusted in order to incorporate the knowledge gathered. Four iterations are planned: content iteration, contributor iteration, context iteration and final iteration. These iterations have been decided by adapting the social media components proposed by Corney et al. (2014) and Middleton and Gottron (2014) to the emergency area: Content, Contributor and Context.

Each iteration will have three stages: design, build and evaluate. Design will focus on the component’s literature review. It will focus on detecting social media features related to the component analysed. In order to analyse each component to assess whether it should be included in the framework, an experiment will be created. This will be distributed using an online survey approach (Bradley, 1999; Kajewski, 1994) through social media channels and e-mail. A non-probability self-selecting sampling technique will be used. Sample profile will include social media users that are information distributors and/or seekers during emergencies. It will include citizens, community managers for fire brigades, police force and other emergency forces (for e.g. @CALFIRE_PIO, U.S. Fire (USFA), @CALFIRESANDIEGO, @kentfirerescue, @fema, @BaltimorePolice).

The survey will ask users to analyse different messages from social media according to their degree of reliability. The experiment will use a similar methodology used in Castillo et al. (2011) where users were asked to classify tweets in one of the following categories: (i) almost certainly true, (ii) likely to be false, (iii) almost certainly false, and (iv) ‘I can’t decide’. Additionally, the experiment will ask the responder to decide if they will send an emergency response to the user (i.e. ‘send emergency team’ or ‘not an emergency’). Three main elements will be analysed: Content, Contributor and Context (Figure 2). This will be complemented by providing additional intrinsic and extrinsic information to the responder (expert insights, wisdom of crowd, linked data). The objective is to find what information is more useful to assess the veracity of social media messages with technical/analytical and non-technical/analytical resources.

The experiment will provide the basis to design a verification framework which will be presented during interviews to emergency managers, IT practitioners and academia. Interviews will focus on evaluating each iteration and identifying potential implementation in existing systems and future lines of research.

Summary and conclusions

This paper describes the importance of information quality in social media for decision-making during emergencies. It explores current approaches to verify information using social media intrinsic properties and additional resources such as experts’ advice, crowdsourcing and linked data (i.e. sensor nodes). Current approaches have been designed to support emergency managers and organisations. Little research has been developed to provide citizens and field workers with tools to evaluate social media information. In order to address this gap, three research questions are proposed to support the development of the information verification framework. The final section has introduced the planned activities for the proposed research using design research methodology.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Data veracity is understood as data quality, reliable and trustworthiness.

2. e.g. social media, earthquake detection using seismometers, wireless networks, unmanned systems, embedded sensors, pattern recognition, surface reconstruction, data fusion, and scheduling algorithms (IBM Corporation Software Group, 2013; Pu & Kitsuregawa, 2013).