System self-contextualizability is the system ability to autonomously adapt its behavior to the uncontrollable relevant context to keep its objectives satisfied. Self-contextualizable system must have alternative behaviors each fitting to a set of contexts. We propose to start considering context at the level of requirements engineering, adopting Tropos goal model to express requirements and complementing it with our proposed context analysis. We define variation points on goal model where a context-based decision might need to be taken, and propose constructs to analyze context. While goal analysis provides constructs to hierarchically analyze goals and discover alternative sets of tasks to be executed to satisfy a goal, our proposed context analysis provides constructs to hierarchically analyze context and discover alternative sets of facts to be monitored to verify a context.
Self-contextualizable software is able to monitor context and to autonomously adapt to it in order to keep its design objectives satisfied. This mainly means that the developed software has a space of alternative behaviors each fitting to a specific context. Consequently, and to allow for a systematic behavior adaptation to context, we need to specify the relation between each behavior and the context where such behavior is adoptable.
Context modeling, (e.g. [1]), concerns finding modeling constructs to represent context, but there is still a gap between the context models and the software behavior models, i.e. the relation between context and its use is missed. We need to reduce such gap and to allow for answering questions like: "how do we decide the relevant context?", "why do we need context?" and "how does context influence software and user behavior adaptation?". Salifu et al. [2] investigate the use of problem descriptions to represent and analyze variability in context-aware software; the work recognizes the link between requirements and context as a basic step to design context aware systems.
Goal analysis (i* [3], Tropos [4], and KAOS [5]), provides a way of analyzing high level goals to discover and represent alternative sets of tasks that can be executed to achieve such goals. Goal model is used to represent the rationale of both humans and software systems, and is a mainstream technique for requirements engineering and helps for representing software design alternatives. These features are important for self-contextualizable software that must have alternatives and a rationale that reflects users and software adaptation to context in order to adopt a useful execution course [6]. From a goal perspective, a self-contextualiazable software is assigned a set of goals and has to keep them satisfied in a way tailored to the context. However, there is still a gap between goal analysis that discovers the space of alternatives to reach a goal, and the context where each alternative is adoptable.
Customizing goal models to fit to user skills and preferences was studied in [7,8]. The selection between goal satisfaction alternatives is based on one dimension of context, i.e. user skills, related to the executable tasks of the goal hierarchy, and on user preferences which are expressed over softgoals. Lapouchnian et al. [9] propose techniques to design autonomic software based on an extended goal modeling framework, but the relation with the context is not focused on. Liaskos et al [10], study the variability modeling under the requirements engineering perspective and propose a classification of the intentional variability when Or-decomposing a goal. Differently, we focus on context variability, i.e. the unintentional variability, which influences the applicability and appropriateness of each goal satisfaction alternative.
In this paper, we extend our previous goal modeling framework [11][12][13] to support the self-contextualizability at the goal level. We use goal analysis in conjunction with our proposed context analysis to build self-contextualizable goal models. We provide constructs to hierarchically analyze context at the variation points of Tropos goal model. The proposed hierarchial context analysis will help us to identify the data the system has to monitor to verify each context, i.e. it helps to identify the monitoring requirements. In what follows, we describe our vision and proposed framework.
Self-contextualization, from a goal oriented perspective, is the autonomous selection between goal satisfaction alternatives to fit to the variable context. The basic idea is to keep the goals satisfied in different contexts through an adaptable course of execution. Here we give four principles for self-contextualizability from a goal perspective: Context effects: context influences the set of satisfiable goals, the need to satisfy each goal, the possible goal satisfaction alternatives, and the quality of each alternative. Context is not adaptable: context does not adapt to users and software; rather, users and software have to adapt to context to satisfy, and/or to better satisfy, their goals. Users contextualizability comes first: defining software self-contextualizability has to be based on how its users behave in different contexts. We need to analyze users goals to discover a space of satisfaction alternatives, and to analyze how context influences users in deciding and choosing among different alternatives for satisfying their goals. Self-contextulaizable software has to reflect users adaptation to the context first, and then adapt itself to context. Contextualization times: software contextualization can be done while deploying the software, that supports different alternatives, in a certain environment to adapt it to some contextual characteristics that never change, i.e. some alternatives are not usable and redundant in that environment. Other variable characteristics necessitate keeping different alternatives and adopting one of them at runtime based on which context is valid. In both cases, the systematic contextualization needs specifying the relation between goal alternatives and context.
Fig. 1 represents a partial goal model for a promotion information system that will interact with customers and sales staff, through their PDAs, for promoting products with a number of alternatives. To make it self-contextualizable, we need to explicitly represent the relation between these alternatives and context. Contexts can be related to the following variation points of Tropos goal model: We need to analyze context to discover, represent, and to agree on how it can be verified. Differently from the research in context modeling (see the survey [14]), we do not provide an otology or a modeling language for representing context, but modeling constructs to hierarchically analyze context. We specify context by a high level statement and start to refine it to get by the end alternative sets of facts that are definable on monitorable data. The truth value of facts is bottom-up propagated to define the truth value of the high level statement (see Fig. 2). We also need to keep track of the elements of environment the different goals, tasks, statements and facts are concerned with. To do this, we propose to parameterize goals, tasks, statements, and facts. Parametrization is necessary to decide and make it explicit the domain of discourse of goals, tasks, statements, and facts, i.e. the elements of the environment under discussion.
To better explain the approach, let us consider the Or-alternative " Promote by crossselling the product to customer" which concerns the elements: product, and customer.
A simplified analysis of the context where this alternative is adoptable is shown in Fig. 2.a. Moreover, by analyzing the elicited facts, the analyst can extract the conceptual data model the monitoring system has to instantiate to make facts validation possible. The analysis of the last context facts leads to the conceptual data model of Fig. 2 We provide a set of modeling constructs to analyze high-level context statements and to elicit the monitorable facts that are used to assign truth values to statements. Context, the reification of the environment that surrounds the system and influences it, can be monitored but not controlled by the system itself [15]. Under this assumption, systems cannot change the context but should adapt to it for satisfying users objectives.
The objective method to compute a fact truth value requires monitoring some characteristics and history of a set of relevant environment elements.
Statement verification could not be objectively done because the system is not able to monitor and get all the data needed to compute the truth value of a statement, or because there could be no consensus about the way of knowing the truth value of a statement. Anyhow, to handle such problem we adopt a relaxed confirmation relation between facts, which are objectively computable by definition, and statements, in order to assign truth values to statements. We call this relation "help" and define it as following:
The relation help is strongly subjective, since different stakeholders could define different help relations for the same statement. Our proposed hierarchical context analysis has the potential to make a context (i) more understandable for the stakeholders, (ii) easily modifiable as it is not given as one monolithic block, and (iii) more reusable as parts of the statement analysis hierarchy can be also used for other variation points or other stakeholders context specifications. Moreover, the analysis justifies why the monitoring system has to capture environmental data, as such data are needed to verify facts that in turn confirm or disapprove statements that are needed to make a decision at the variation points of the goal model. Specifying for each fact the related fragments of the data conceptual model is useful for purpose of tracking. For example, if for some reason, a group of stakeholders decided to drop, to alter, or to reuse one alternative, statement, or fact, we still can track which fragments in the conceptual data model could be influenced.
In this paper, we have introduced a goal-oriented framework for self-contextualizable systems. We have discussed the self-contextualizability from goal-based perspective. We have proposed the association between goal satisfaction alternatives and context. In turn, context is defined through statement analysis that elicits the alternative sets of facts the system has to validate on monitorable data so to confirm the high level statements. As future work, we will define models covering all development phases of selfcontextualizable software and a process to facilitate the development of high quality self-contextualizable software. We also want to extend the formalization and reasoning mechanisms we proposed in [11,12] to fit to the modeling framework introduced in this paper. We will work on complex case studies in order to better validate our approach. Finally, statements and facts suffer of interaction problems; therefore, supporting tools and reasoning techniques should be proposed to support the design and verification of models.
This work has been partially funded by EU Commission, through the SERENITY, and COMPAS projects, and by the PRIN program of MIUR under the MEnSA project. We would also like to thank Prof. Jaelson Brelaz de Castro for the valuable discussion we had about this work.