4.1 Contextualisation for Information Gathering
The specific term “situation awareness”, most commonly used by the Human-Computer Interaction community (Endsley 1995), is defined as “the perception of the elements in the environment within a volume of time and space, the comprehension of their meaning, and the projection of their status in the near future”. Furthermore, a situation can provide answers to some but not all questions induced by the decision problem. Therefore, situation awareness plays a central role in cognition, which spans the entire spectrum of cognitive activities from situation analysis (perception + comprehension) to near-future forecast. To efficiently support humans in their situation understanding in such decision-making processes and particularly in the induced cognitive overload context, there is a crucial need to provide in time only the critical information required. In such situation, the decision makers have to deal with a huge amount of heterogeneous information that comes from different sources and thus is often redundant, if not conflicting.
It is therefore necessary to model and specify the context and situation to provide support functions for cognitive overload mitigation such that the stakeholders can easily exchange, share, and re-use their knowledge. Situation understanding is not only a matter of knowing the objects that are present in the external world, but also to make sense of the situation assessment achieved by specifying and representing the relations amongst it. To design such support systems, one of the major challenges is to construct a situation model from which crisis management operations can be planned and updated. A situation model is a representation of the situation that will serve as a reference for each participating user; it allows a complete cognitive understanding of the situation (Mattioli, Museux, Hemaissia, and Laudy 2007). It includes the representation of all events, actions, actors, past experiences, and items which explain, impact or are impacted by the situation (note that in this chapter, actors are the individuals involved in the DICE mitigation). Therefore, the situation model provides a cognitive representation of the situation context.
To build a situation model, we use ontology concepts as they are used to capture knowledge about some domain of interest. Matheus, Baclawski, and Kokar (2003a) advocate the use of formal ontologies to describe the fundamental events, objects, and relations of a situation’s domain and logical rules. This is in order to define ways of fusing information and identifying higher order relations relevant to the situation (Matheus, Baclawski, and Kolar 2003b). Smirnov, Pashkin, Shirov, and Levashova (2007) also propose to include ontology for mega disaster management.
Therefore, the situation model contains a knowledge-level view of DICE using some specific ontology, such as Crisis Response Management (CRM) ontology. The model is created and updated by a constant flow of events and reports collected from the operational space. These events include both human and artificial intelligence. Because of the large amount of raw data being collected, the event stream needs to be processed and correlated to produce situational events, i.e., events at the domain level. This reduction and inference step is performed by an information-fusion stage. Thanks to the OASIS consortium, most of the relevant situation items (class, property or individual) can be described through the definition of the Tactical Situation Object (TSO) (OASIS Project 2006), which contains at least the following information.
Identification information: it shall be identified in an unambiguous way. It shall also describe who the originator of the information is and when the information was created.
Description of the case: the TSO is one solution to provide to other entities its own view of the case such as the type of the case, its extent, the number of casualties, the consequences on the environment, its criticality, etc.
Description of the context: the CRCT is interested to know which resources are already used, which resources are available (including the human resources), their availability, their position, and their capabilities.
Description of the actions: it is also very important to inform the others of the activities which are in progress or which are foreseen, so that the coordination is efficient. Information on the past, future, and ongoing tasks must be available (their status, the teams, and resources that are engaged on a task, their planning, etc.).
Following OASIS guidelines, the Interactive Collaborative Information Systems (ICIS) project (2003) provides crisis response management ontology composed of three modules: situational, contextual, and decisional (Figs. 3, 4 and 5). The situational information provides a view of the crisis case, such as, for example, the type of crisis, its extent, number of casualties, environmental impact, and criticality. The contextual information addresses the status of resources, which includes list of resources, their availability, position, and capabilities. The decisional information concerns the tasks that are ongoing. This comprises their status, the team, and resources involved and their planning (Mattioli et al. 2007).
Based on this situation model, the information extraction consists of the analysis of data coming from different sources so that these data become meaningful. The objective is to extract information from the available data. The techniques we use are text mining, audio mining, and image processing. The notion of text mining has been widely spread by the search engines though these do not allow the performance of sense making on the data. Text mining, as used here, seeks to find relations between blocks of text data contained in a text to provide information. Audio mining allows the extraction of information from an audio stream (speech or noise analysis). In particular, speech analysis is reinforced today by the background sounds analysis that can provide much additional information to optimise the sense-making processing. Information extraction methods, either in text, audio or images, usually use ontology to structure the input data, and supervised classification methods to analyse and extract relations amongst entities. Thus, this is very much correlated with the above modelling phase. The capability of the method to extract relationships amongst entities and model the crisis situation context is a key factor of the information extraction phase.
Following the information extraction, the construction of a coherent overall situation picture will require some fusion of heterogeneous and redundant information techniques. Most of the time, because information is coming from a wide range of sources including both human and signal sources, the pieces of information must be combined and correlated in accordance with
The objectives of the crisis response (for example, for early crisis response the objectives can be immediate life saving, search and rescue, medical aid),
The information reliability and relevance, and
The information temporality (obsolescence and update).
The processing and correlating of all available information is performed by an information-fusion stage. The information items, stored as graph structures, are combined using fusion heuristics. The fusion algorithm is based on a graph homomorphism algorithm where we look for compatible subgraphs (i.e., fusible). The compatibility is processed thanks to the fusion heuristics – called fusion strategies – that contain both a part to test the compatibility of two observations and a part to compute the fused observation (Laudy and Ganascia 2008a,b; Laudy, Ganascia, and Sedogbo 2007).
This early overall situation picture then has to be consolidated with some situation understanding process and sense-making analysis. This is to provide a meaningful situation awareness, the basis on which the final decision-making step can be performed. These two steps are presented hereafter.
4.2 Contextualisation to Get a Meaningful Situation Awareness
Situation understanding is used to manage, in real time, crisis situations of critical infrastructures (such as production line breakdown, network intrusion or attack of public area). For example, an alert can be raised only if responsible actors have a clear understanding of the global situation. The situation understanding has to be as complete as possible and has to consider the information from different points of view, from the smallest standpoints to wider perspectives. For example, the detection of weak signals amongst a huge amount of information is of major importance for information intelligence activities. This implies a capability of analysing the situation down to its smallest component using a well-established global situation analysis. This should enable the detection of small pieces of information that – put back into the context – are of prior interest.
The situation understanding process is a cognitive and analytical process by which we can take some information and synthesise it with previously held information. It is then used in a new context, or used in a different way to that envisioned when first collected. The difference between situation understanding and information gathering is the difference between learning and memorising as compared to this ability to transform information through experience and know-how. Thus, the situation understanding depends not only on the information content but also on its contextualisation.
It should be pointed out that context also impacts the frequency at which information used to get meaningful situation awareness is constantly updated. As already mentioned in the introduction, real-time here refers to the notion of on time and can range from a few seconds to many days depending on the DICE type (example of bomb explosion vs. pandemic situations). The need for real-time information for situation awareness is indeed very dependent on the context, as shown in the following example which is very illustrative, though not directly linked to DICE management.
When enhancing air traffic management (ATM) procedures and tools, new concepts have to be validated through real-time simulation exercises. During an air traffic simulation session, pilots were asked to deal with a new ATM procedure that involved an automatic adaptation of the aircraft speed with regard to the preceding aircraft speed and position. The modified simulator displayed the value of the aircraft speed taking into account the new requirements in real time. However, after a few training exercises, this speed value displayed in real time had to be averaged and smoothed so as to display a stabilised value of the speed. This was because pilots were unable to get a comprehensible situation awareness from a constantly changing speed. It is therefore important to adapt the notion of real time to the use, context, and decision need. Thus, real-time here implies the notion of on time, or as referred to (Nelson 2004) time to decision.
This outlines the crucial role of contextualisation in the previous information- gathering step to building a contextualised situation understanding, and emphasises that this information gathering requires real-time updates. Complex event processing (CEP) is widely used for situation analysis and understanding. It allows the identification of relevant information to provide answers, in real time, to the crisis managers’ needs. It consists of the detection of complex events, the correlation, and analysis of hierarchical events and events relations (such as causality or temporality) (Museux, Mattioli, Laudy, and Soubaras 2006).
With a keen, updated, and contextualised situation awareness, a coherent and efficient crisis management plan to mitigate and resolve the DICE can be envisaged and elaborated.
4.3 Dual Impact of the Context on Decision Making
Decision-making support is used in the response planning and implementation phase to optimise resource allocation and scheduling. This aims to first meet specific relief objectives and later to fulfil recovery and development goals. Because of the high combinatory complexity of resource allocation and scheduling problems, we use a hybrid method, combining an optimisation algorithm with a meta-heuristic algorithm. This enables us to split the problem into subproblems. To be efficient and operational, the selected decision has to correctly tackle the operational constraints in the optimisation problem. These constraints then are to be included into the Situation Model. The final decision also takes into account some potential conflicting goals. This last issue can be addressed by the use of some negotiations in the optimisation algorithm (Hemaissia et al. 2006).
Crisis response requires the management of all relevant resources (infrastructures, material, human); a lack of resources can lead to the abortion of the whole response. It is thus very important to have an updated knowledge of the availability of resources and of the procedures required to use them. Besides, in order to support the decision makers, the shared situation awareness picture has to materialise the “decisional environment” and so makes it clear why emergency operation planning is necessary.
Once some allocation or scheduling of resources has been performed, and before the commitment to the crisis response tasks has been made, there is a great necessity to analyse what would be the impact of the response operations. This impact should be analysed in terms of results as well as risk involving the crisis mitigation. Between two responses, it might be wiser not to choose the more efficient one (efficient with regard to some predefined criteria such as time and/or cost) if its level of risk is very high compared to the other one. The efficiency and risk can be evaluated by the modelling of the situation evolution. The final response selection thus requires a time projection of the situation understanding process. We call it the “dual impact of the context for the crisis management decision support”. The context first constrains the resolution of the resource allocation problem (or scheduling), but it also helps discriminate amongst several resource allocation solutions.
To measure the impact of a decision to the context, some simulation tools to “play” the decision in advance are used to analyse its consequences. As an example, some recent research carried out by the multiagent community is now presented.
This research aims to define a simulation tool for road traffic crisis management and uses the context as a key function for the agent to take his/her decisions. Environment as active support of interaction (EASI) is a multiagent model of interaction conceived by the Laboratoire d’Analyse et de Modélisation des Systèmes pour l’Aide à la Décision (LAMSADE), which is based on a reification of implicit and explicit rules to enable the interaction of a multiagent system with the environment. The objective is to model the link between the perception domain and the activities of the agent, and so to allow the use by the agent of this knowledge to reach its objective. More recently, a new contextual simulation multiagent model has been studied by F. Balbo and F. Badeig. The objective is to extend the EASI contextual principle of interaction to a contextual activation of the simulation agents (Saunier, Balbo, and Badeig 2007). This model, called Environment as Active Support of Simulation (EASS), separates the simulation process from the activation process of the agents. The approach used is the study of the context and the interactions between the actors (Badeig, Balbo and Pinson 2007). A first application of this work concerns crisis management in the transport domain. The goal is to simulate the complexity of a crisis for better analysis and understanding. The simulation of different scenarios for the same crisis will help the conception of new intervention plans and the validation and adaptation of existing plans (Badeig and Balbo 2008).
The above description of the decision-making process thus shows that the situation understanding process is essential and that a decision has to be correlated to its context.
4.4 Contextualisation of the DICE Management Upgrade Process
A very important point to take full advantage of DICE management processes and to improve them is to regularly train all the actors who may have to play a role in the DICE mitigation. It must be stressed that people involved in the process during a DICE management might not be people usually in charge of the process. The training should thus be large enough to familiarise the people with a large event of practices. Training sessions aim at getting the people used to the tools and practices, often through life-sized exercises of simulated DICE events. The goal is to put people in some degraded situations already encountered by others, in order to develop individual best practices. This implies getting the actors accustomed to different contexts (Gauthey 2008).
The upgrade of the DICE management process also relies, of course, on the analysis of the feedback from past events so as to learn and capitalise from past mistakes. As stated by Cheila Duarte-Colardelle, most of the incident report forms do not allow for catching the “why” and “how” of the acts and decisions. The situation perception is ignored, which is a limitation to the understanding of some events. Dysfunction and misleads are analysed with the aim of understanding not only the “how”, but also especially the “why”. When someone makes a decision, several choices are available; the decision and the resulting action might not be understood by others. “Why has he done this?” There is always some reason that can be explained by the context (Duarte-Colardelle 2006).
Rescue plans have also been defined step-by-step and adapted with regard to some evolution of the context. These evolutions are detected and analysed to capitalise on and benefit from them. In the work of Duarte-Colardelle (2006), the author explains how to trace and analyse the dynamic progress of each case study. This method, called positive feedback (REX positif), aims no longer at learning from accidents or dysfunctions, but at detecting the good practices and at strengthening them; the notion of “key moment associated with a cycle of decision” is the most significant contribution of the method. The comparison between the decisions that make the situation worse, with “good” decisions that could have saved the situation, is the basis of the process. We agree with Sarant that this is an important contribution, provided the decision making contexts are taken into account (Sarant 2003).