Zack [42] distinguished these four termsaccording to two dimensions: the nature of what is being processed and the consti-tution of the processing problem.The nature of what is being processed is either information or frames of ref-erence. With information, we mean “observations that have been cognitively pro-cessed and punctuated into coherent messages” [42]. Frames of reference [4, p.108], on the other hand, are the interpretative frames which provide the context forcreating and understanding information. There can be situations in which there is alack of information or a frame of reference, or too much information or too manyframes of reference to process.

Ta b l e 3DERMIS design premises [29]

For our research design, we drew on Walsham [33] and Klein and Myers [13],who provide comprehensive guidelines on how to conduct interpretive case studyresearch in the IS domain.

The problems of managing information and managing frames of reference are“tightly linked in a mutually interacting loop” and require “managing informationand the systems that provide it” [42]. IS have been generally designed to overcomethe information problems from Table 1. Most IS are aimed at either storing and re-trieving information to reduce uncertainty, such as database management systemsand document repositories, or at analyzing and processing large amounts of infor-mation to reduce complexity, such as decision support systems [31]. However, aswe have previously discussed, information related strategies are not always helpfulin coping with a variety of potential meanings.Problems of interpretation and the creation and management of frames of refer-ence, which aids Sensemaking, have generally not been taken into account whendesigning IS. Most IS currently seem tointend the opposite because they aim atreplacing or suppressing the possibility tomake sense of situations.

there is scarce research on how IS can support informa-tion processing challenges—specifically related to Sensemaking—in crisis manage-ment [14]

Sensemaking is about contextual rationality, built out of vaguequestions, muddy answers, and negotiated agreements that attempt to reduce ambi-guity and equivocality. The genesis of Sensemaking is a lack of fit between whatwe expect and what we encounter [40]. With Sensemaking, one does not look at thequestion of “which course of action should we choose?”, but instead at an earlierpoint in time where users are unsure whether there is even a decision to be made,with questions such as “what is going on here, and should I even be asking this ques-tion just now?” [40]. This shows that Sensemaking is used to overcome situationsof ambiguity. When there are too many interpretations of an event, people engagein Sensemaking too, to reduce equivocality.

Decision making is traditionally viewed as a sequential process of problem classifi-cation and definition, alternative generation, alternative evaluation, and selection ofthe best course of action [26]. This process is about strategic rationality, aimed atreducing uncertainty [6, 36]. Uncertainty can be reduced through objective analysisbecause it consists of clear questions for which answers exist [5, 40]. Complex-ity can also be reduced by objective analysis, as it requires restricting or reducingfactual information and associated linkages [42]

The central problem requiring Sensemaking ismostly that there are too many potential meanings, and so acquiring informationcan sometimes help but often is not needed. Instead, triangulating information [34],socializing and exchanging different points of view [20], and thinking back of pre-vious experiences to place the current situation into context, as the retrospectionproperty showed us, are a few strategies that are likely to be more successful forSensemaking.

Just as the information processing challenges from Table 1 are not mutually ex-clusive, Sensemaking and decision making cannot be separated, but instead operatesimultaneously. Meaning must be established and then sufficiently negotiated priorto acting on information [42]: Sensemaking shapes events into decisions, and deci-sion making clarifies what is happening [40].

Weick et al. [41, p. 419] formulate a gripping conclusion on what the sevenSensemaking properties are all about: “Taken together these properties suggest thatincreased skill at Sensemaking should occur when people are socialized to makedo, be resilient, treat constraints as self-imposed, strive for plausibility, keep show-ing up, use retrospect to get a sense of direction, and articulate descriptions thatenergize. These are micro-level actions. They are small actions. But they are smallactions with large consequences.”

The seven different properties of Sensemaking can be captured by the acronym SIRCOPE: Social context, Identity construction, Retrospection, Cue extraction, Ongo-ing projects, Plausibility, and Enactment [17–21, 37–39]

Crisis environments are characterized by various types of information problemsthat complicate the response, such as inaccurate, late, superficial, irrelevant, unreli-able, and conflicting information [30, 32]. This poses difficulties for actors to makesense of what is going on and to take appropriate action. Such issues of informationprocessing are a major challenge for the field of crisis management, both concep-tually and empirically [19].

We use the theory of Sensemaking to study exactly this: how people makesense of their environment, and how they give meaning to what is happening. Sense-making is a crucial process in crises, as the manner and thereby the success of howone deals with crucial events is determined by the grasp one has of a situation.

This distinction enlightens the reading of thegrowing social media and mass emergency lit-erature for three reasons. First, without it, thisnew literature risks undoing decades of work bysocial scientists who have dismantled the mythsof disaster, with a dominant discourse thatincludes panic and unlawful behavior by victims.But in disasters arising from natural hazards, weknow such behaviors are not typical. Massemergencies arising from criminal behavior canhave a much wider range of collective behaviorbecause the source of the hazard is unknown,unpredictable and perhaps more imminentlydangerous

Misinforma-tion arising from natural hazards or exogenousevents might be greater in kind, but less inimpact, with fewer in-common readers as it tra-verses a network that can move a little slowerthan it might in criminal mass emergency events.Because the problem-solving tends to be morediffuse in exogenous events, the same messagemight not reach enough people; in other words,the misinformation might also be thinly diffused.Misinformation in such events is more likely toage out, or not be relevant to enough locations topose a big threat—in other words, all informationin thefirst place is less likely to be categoricallycorrect or incorrect, and as such, it is hard tofindas much value in pursuing the threat of misin-formation in such situations.

Wefindendogeneityandexogeneityof haz-ards to be a meaningful distinction in socialmedia in mass emergencies research, one thatreadily clarifies for a range of researchers andreaders who are outside the social science disci-pline. Just as events that arise from exogenousand endogenous hazards differently impact legal,political, health, and other societal systems, so dothey differently impact social media behavior.8With exogenous events, the culprit is beyondreach, and unstoppable. With endogenous agents,the suspect lies within. Therefore, organizingfeatures of the communication are distinctlydifferent, because the source(s) of the problem(s),the nature of their solutions,and the ability forthe perception of the collective control of theoutcomeare different. Online participation focu-ses on in-common salient problems when theyare present; when the problems are lessin-common and must be addressed in parallel, thecrowd organizes in many smaller groupings and,often endogeneity and exogeneity of hazardspredicts this (Palen & Anderson,2016).

We make this point because we worrythat the very idea of“social media”flattens themany meanings of“crisis”and“emergency”forwhich social sciencefields have worked to pro-vide insight. For example, because Twitter orFacebook are available for use in any kind ofcrises, it is easy to make these applications thesalient concern, and ask“Is Twitter or Facebookbetter in emergency response?,”rather thanquestion how the very nature of emergencyresponse might beg for different forms of infor-mation seeking and reporting. We refer to thisflattening of communication medium and hazardas thesocial media and crisis confound.

Research has demonstrated that data fromsocial media interactions can provide situationalawareness for specific crisis-related tasks anddomains. Using natural language processing (afield of study which enables computers to ana-lyze and understand the human language),machine learning (techniques that provide com-puters with the ability to learn), and crowd-sourcing (the process of accomplishing a task bydividing it into subtasks that can be performed bya large group of people), several research groupshave developed methods and tools for detectingand monitoring epidemics through social mediadata analysis (Brennan, Sadilek, & Kautz,2013;Chen, Hossain, Butler, Ramakrishnan, & Pra-kash,2016; Munro,2011; Olteanu, Vieweg, &Castillo,2015).

ubsequentresearch has focused on developing natural lan-guage processing classifiers that analyzes text tohelp identify tweets contributing to situationalawareness (Corvey, Verma, Vieweg, Palmer, &Martin,2012; Verma et al.,2011), though ingeneral the state-of-the-art of thefield is such thatautomation behind situational awareness deriva-tion is difficult to do dependably. Ireson (2009)assessed the extent to which public forum post-ings could add to situational awareness duringthe 2007floods around Sheffield, UK and foundextractable relevant event information despite theinconsistent quality and conversational nature ofthe posts.