Abstract
Hazard-intense applications of cyber-physical systems (CPSs), such as the evacuation of a building on fire, require personalized informing on the basis of a real-time assessment of dynamic context. In this paper, a context-dependent message construction mechanism (CD-MCM) is proposed with the objectives (i) to inform people about the emergence and development of a situation unsafe for them, and (ii) to instruct them what they have to do according to an adaptively computed action plan. To achieve this, the concepts of ‘situation’ and ‘impact indicator’ have been introduced in order to facilitate the computation of personalized action plans and to send messages about the level of danger and the requested actions. In both activities, the inferred implications of situations are used as the basis of informing the involved people. The messages are adapted to actual situations. In addition, the concept of ‘relevance indicator’ was utilized to assess the significance of the standing situations for the concerned people in a quasi-real-time manner. The level of danger was evaluated for each person by totaling the values of the situation-related relevance indicators. This was also used to select the proper message templates from the predefined alternatives in the process of message construction. The personalized messages were generated based on the chosen message template and the various message components describing concerned situations or providing instructions. The CD-MCM was validated in a simulated indoor fire evacuation guiding application. In the practical evaluation of the quality of the generated messages, a sample of test people was involved. The results of the evaluation show that the messages generated by the proposed CD-MCM lead to more effective messaging about the personal context and the expected actions than the messages constructed by using static context information only. The reason is that the proposed template-based message construction mechanism facilitates the appropriateness as well as the articulation of the contents of context-sensitive personalized messages.