Reinforcement learning (RL) is an important class of machine learning techniques, in which intelligent agents optimize their behavior by observing and evaluating the outcomes of their repeated interactions with their environment. A key to successfully engineering such agents is to provide them with the opportunity to engage in a large number of such interactions safely and at a low cost. This is often achieved through developing simulators of such interactions, in which the agents can be trained while also different training strategies and parameters are explored. However, specifying and implementing such simulators can be a complex endeavor requiring a systematic process for capturing and analyzing both the goals and actions of the agents and the characteristics of the target environment. We propose a framework for model-driven goal-oriented development of RL simulation environments. The framework utilizes a set of extensions to a standard goal modeling notation that allows concise modeling of a large number of ways by which an intelligent agent can interact with its environment. Through subsequent formalization, the model is used by a specially constructed simulation engine to simulate agent behavior, such that off-the-shelf RL algorithms can use it as a training environment. We present the extension of the goal modeling language and its semantics, and show how models built with it can be made executable. Further, we introduce a model development and transformation toolset to allow efficient generation and execution of simulator specifications. Through experimental studies we show how the proposed language and tool can be used for assisting the RL engineering process.

Interviewing is known to be one of the most common requirements elicitation techniques. Interviews are driven by a series of questions asked for the purpose of receiving responses that can help understanding the domain and the needs of stakeholders. However, what constitutes a successful choice and ordering of questions continues to be more of an art than a systematic process. We review literature from a broad range of disciplines in which interviewing is widely applied, in order to identify a set of categories for characterizing interview questions. The resulting typology aims at offering an initial coding language for qualitatively analyzing interview content. Such coding language can be further validated for its reliability to enable standardization and community-wide reuse. We offer examples of how such an instrument would help researchers develop and evaluate both descriptive and normative theories of interviewing.

[Background]: Conceptual modeling languages have been widely studied in requirements engineering as tools for capturing, representing and reasoning about domain problems. One of these languages, goal models, has been proposed for representing the structure of stakeholder intentions. Like most other conceptual modeling languages, goal models are visualized using box-and-line diagrammatic notations. But is this box-and-line approach the best way for visualizing goals and relationships thereof? [Aims]: Through a series of experimental studies we have recently endeavored to find out. In this presentation, we describe features of our alternative visualization proposals and present experiences gained from our attempts to empirically evaluate them. [Results]: Central to what we learned is the usefulness of distinguishing between language visualization and intended language semantics and of measuring the degree by which the former serves correct recognition of the latter. [Conclusions]: Our experience from these studies could be useful for those interested in experimentally-driven conceptual modeling language design.

[Background]: Developing conceptual models is an integral part of the requirements engineering (RE) process. Goal models are requirements engineering conceptual models that allow diagrammatic representation of stakeholder intentions and how they affect each other. A specific goal modeling language construct, the contribution of goal satisfaction of one goal to another, plays a central role in supporting decision problem exploration within goal models. We report on an experiment whose aim was to measure the user perception of the meaning of the aforementioned modeling construct. A set of contributions under different scenarios were given to experimental participants who were asked what they thought the effect of the contribution was. We found that participants are not always in agreement either within themselves or with the designers' intentions on the meaning of the language. The results call for possible adaptations to the way goal modeling languages are used.

Goal models allow efficient representation of stakeholder goals and alternative ways by which these can be satisfied. Preferences over goals in the goal model are then used to specify criteria for selecting alternatives that fit specific contexts, situations and strategies. Given such preferences, automated reasoning tools allow for efficient exploration of such alternatives. Nevertheless, to be amenable to such automated processing, goals and preferences need to be specified in a formal language, making automated processing inaccessible to the very bearers of goals and preferences, i.e., the stakeholders. We combine natural language processing techniques to allow specification of preferences through natural language statements. The natural language statement is first matched through regular expressions to distinguish between the preference component and the goal component. The former is then mapped to a preferential strength measure, while the latter is used to identify the relevant goal in the goal model through statistical semantic similarity techniques. The result constitutes a formal representation that can be used for alternatives analysis. In this way, stakeholders can access advanced goal reasoning techniques through simple natural language preference expressions, facilitating their decision making in various requirements analysis contexts. An experimental evaluation with human participants shows that the proposed system is of substantial precision and that a mapping from natural preferential verbalizations to predefined preferential strength labels is possible through sampling from crowds.

As modern software systems grow in size and complexity so do their configuration possibilities. Users are easy to be confused and overwhelmed by the amount of choices they need to make in order to fit their systems to their exact needs. We propose a method to construct adaptive configuration elicitation dialogs through utilizing crowd wisdom. A set of configuration preferences in the form of association rules is first mined from a crowd configuration data set. Possible configuration elicitation dialogs are then modeled through a Markov Decision Process (MDP). Association rules are used to inform the model about configuration decisions that can be automatically inferred from knowledge already elicited earlier in the dialog. This way, an MDP solver can search for elicitation strategies which maximize the expected amount of automated decisions, reducing thereby elicitation effort and increasing user confidence of the result. The method is applied to the privacy configuration of Facebook.

Goal models have been found to be useful for supporting the decision making process in the early requirements phase. Through measuring contribution degrees of low-level decisions to the fulfilment of high-level quality goals and combining them with priority statements, it is possible to compare alternative solutions of the requirements problem against each other. But where do contribution measures come from and what is the right way to combine them in order to do such analysis? In this paper we describe how full application of the Analytic Hierarchy Process (AHP) can be used to quantitatively assess contribution relationships in goal models based on stakeholder input and how we can reason about the result in order to make informed decisions. An exploratory experiment shows that the proposed procedure is feasible and offers evidence that the resulting goal model is useful for guiding a decision. It also shows that situation-specific characteristics of the requirements problem at hand may influence stakeholder input in a variety of ways, a phenomenon that may need to be studied further in the context of eliciting such models.

Preference-based analysis of goal variability has been found to be useful for understanding alternative ways by which actors can go about fulfilling their goals. More recently, a technique for using such preference analysis to customize software behavior has been proposed. This technique allows sets of goal alternatives that best satisfy a given preference to be provided as configuration parameters to an appropriately instrumented software system, so that the latter adapts its behavior to satisfy those alternatives. In this presentation, we show how this idea has been implemented in a prototypical tool and applied for customizing a simple ATM simulator.

Requirements can differ in their importance. As such the priorities that stakeholders associate with requirements may vary from stakeholder to stakeholder and from one situation to the next. Differing priorities, in turn, imply different design decisions for the end system. While elicitation of requirements priorities is a well studied activity, though, the modeling and reasoning side of prioritization has not enjoyed equal attention. In this paper, we address this by extending a traditional goal modeling notation to support the representation of optional and preference requirements. In our extension, optional goals are distinguished from mandatory ones. Then, quantitative prioritizations of the former are constructed and used as criteria for evaluating alternative ways to achieve the latter. A state-of-the-art preference-based planner is utilized to efficiently search for alternatives that best satisfy the given preferences. This way, analysts can acquire a better understanding of the impact of high-level stakeholder preferences to low-level design decisions.

The analysis of stakeholder requirements is a critical aspect of software engineering. A common way of specifying stakeholder requirements is in terms of a hierarchy of goals whose AND/OR decomposition captures a family of software solutions that comply with the goals. In this paper, we extend this goal modeling framework to include the specification of optional user requirements and user preferences, aggregated together into weighted formulae to be optimized. We team this with an automated reasoning tool, adapted from state of the art research in artificial intelligence planning with preferences, in order to synthesize solutions that both comply with the goals and optimize stakeholder preferences and optional requirements.

Requirements elicitation involves the construction of large sets of conceptual models. An important step in the analysis of these models is checking their consistency. Existing research largely focuses on checking consistency of individual models and of relationships between pairs of models. However, such strategy does not guarantee global consistency. In this paper, we propose a consistency checking approach that addresses this problem for homogeneous models. Given a set of models and a set of relationships between them, our approach works by first constructing a merged model and then verifying this model against the consistency constraints of interest. By keeping proper traceability information, consistency diagnostics obtained over the merge are projected back to the original models and their relationships. The paper also presents a set of reusable expressions for defining consistency constraints in conceptual modelling. We demonstrate the use of the developed expressions in the specification of consistency rules for class and ER diagrams, and i* goal models.

We introduce a variability-intensive approach to goal decomposition that is tailored to support requirements identification for highly customizable software. The approach is based on the semantic characterization of OR-decompositions of goals. We first show that each high-level goal can be associated with a set of concerns, in response to which, alternative refinements of the goal can be introduced. A text corpus relevant to the domain of discourse can be used to derive such variability concerns that are specific to the problem. In parallel, contextual facts that can vary while a goal is being fulfilled are modeled. Then, a high-variability goal model is constructed aiming at responding to the predefined variability concerns completely, while contextual factors are used to test whether it addresses all realistic background circumstances. We apply our approach in a study from the geriatric health care domain

We investigate the personalization capabilities of common personal software systems. We use a typical e-mail client as an example of such a system, and examine the configuration screens it offers to its users. We discover that each configuration value reflects each of the ways with which the user goals can be satisfied. Thus, we construct a goal model in which alternative ways for satisfying high level goals are matched with alternative system configurations. This way, automatic configuration of the system by reasoning about the overlaying goal model can be achieved. We find that the vast majority of the configuration options that refer to system functionality can be configured using this method, facilitating thereby the personalization tasks for users with no technical background, and ensuring, at the same time, consistency and meaningfulness in the configuration result.

A reverse engineering process aims at reconstructing high-level abstractions from source code. This paper presents a novel reverse engineering methodology for recovering stakeholder goal models from both structured and unstructured legacy code. The methodology consists of the following major steps: 1) Refactoring source code by extracting methods based on comments; 2) Converting the refactored code into an abstract structured program through statechart refactoring and hammock graph construction; 3) Extracting a goal model from the structured program's abstract syntax tree; 4) Identifying nonfunctional requirements and derive soft goals based on the traceability between the code and the goal model. To illustrate this requirements recovery process, we refactor stakeholder goal models from two legacy software code bases: an unstructured Web-based email in PHP (SquirrelMail) and a structured email client system in Java (Columba).

Software customization has been argued to benefit both the productivity of software engineers and end users. However, most customization methods rely on specialists to manually tweak individual applications for a specific user group. Existing software development methods also fail to acknowledge the importance of different kinds of user skills and preferences and how these might be incorporated into a customizable software design. We propose a framework for performing requirements analysis on user goals, skills, and preferences in order to generate a customizable software design. We illustrate our methodology with an email system and review an on-going case study involving users with traumatic brain injury.