Machine learning (ML) is used increasingly in real-world applications. In this paper, we describe our ongoing endeavor to define characteristics and challenges unique to Requirements Engineering (RE) for ML-based systems. As a first step, we interviewed four data scientists to understand how ML experts approach elicitation, specification, and assurance of requirements and expectations. The results show that changes in the development paradigm, i.e., from coding to training, also demands changes in RE. We conclude that development of ML systems demands requirements engineers to: (1) understand ML performance measures to state good functional requirements, (2) be aware of new quality requirements such as explainability, freedom from discrimination, or specific legal requirements, and (3) integrate ML specifics in the RE process. Our study provides a first contribution towards an RE methodology for ML systems.

Requirements Engineering for Machine Learning: Perspectives from Data Scientists

Context: User stories have been widely accepted as artifacts to capture the user requirements in agile software development. They are short pieces of texts in a semi-structured format that express requirements. Natural language processing (NLP) techniques offer a potential advantage in user story applications. Objective: Conduct a systematic literature review to capture the current state-of-the-art of NLP research on user stories. Method: The search strategy is used to obtain relevant papers from SCOPUS, ScienceDirect, IEEE Xplore, ACM Digital Library, SpringerLink, and Google Scholar. Inclusion and exclusion criteria are applied to filter the search results. We also use the forward and backward snowballing techniques to obtain more comprehensive results. Results: The search results identified 718 papers published between January 2009 to December 2020. After applying the inclusion/exclusion criteria and the snowballing technique, we identified 38 primary studies that discuss NLP techniques in user stories. Most studies used NLP techniques to extract aspects of who, what, and why from user stories. The purpose of NLP studies in user stories is broad, ranging from discovering defects, generating software artifacts, identifying the key abstraction of user stories, and tracing links between model and user stories. Conclusion: NLP can help system analysts manage user stories. Implementing NLP in user stories has many opportunities and challenges. Considering the exploration of NLP techniques and rigorous evaluation methods is required to obtain quality research. As with NLP research in general, the ability to understand a sentence’s context continues to be a challenge.

/pdf/user-stories-and-natural-language-processing-a-systematic-2atf5mds3e.pdf

User Stories and Natural Language Processing: A Systematic Literature Review

In 2006, Geoffrey Hinton proposed the concept of training ''Deep Neural Networks (DNNs)'' and an improved model training method to break the bottleneck of neural network development. More recently, the introduction of AlphaGo in 2016 demonstrated the powerful learning ability of deep learning and its enormous potential. Deep learning has been increasingly used to develop state-of-the-art software engineering (SE) research tools due to its ability to boost performance for various SE tasks. There are many factors, e.g., deep learning model selection, internal structure differences, and model optimization techniques, that may have an impact on the performance of DNNs applied in SE. Few works to date focus on summarizing, classifying, and analyzing the application of deep learning techniques in SE. To fill this gap, we performed a survey to analyse the relevant studies published since 2006. We first provide an example to illustrate how deep learning techniques are used in SE. We then summarize and classify different deep learning techniques used in SE. We analyzed key optimization technologies used in these deep learning models, and finally describe a range of key research topics using DNNs in SE. Based on our findings, we present a set of current challenges remaining to be investigated and outline a proposed research road map highlighting key opportunities for future work.

A Survey on Deep Learning for Software Engineering

Natural Language Processing (NLP) is widely used to support the automation of different Requirements Engineering (RE) tasks. Most of the proposed approaches start with various NLP steps that analyze requirements statements, extract their linguistic information, and convert them to easy-to-process representations, such as lists of features or embedding-based vector representations. These NLP-based representations are usually used at a later stage as inputs for machine learning techniques or rule-based methods. Thus, requirements representations play a major role in determining the accuracy of different approaches. In this paper, we conducted a survey in the form of a systematic literature mapping (classification) to find out (1) what are the representations used in RE tasks literature, (2) what is the main focus of these works, (3) what are the main research directions in this domain, and (4) what are the gaps and potential future directions. After compiling an initial pool of 2,227 papers, and applying a set of inclusion/exclusion criteria, we obtained a final pool containing 104 relevant papers. Our survey shows that the research direction has changed from the use of lexical and syntactic features to the use of advanced embedding techniques, especially in the last two years. Using advanced embedding representations has proved its effectiveness in most RE tasks (such as requirement analysis, extracting requirements from reviews and forums, and semantic-level quality tasks). However, representations that are based on lexical and syntactic features are still more appropriate for other RE tasks (such as modeling and syntax-level quality tasks) since they provide the required information for the rules and regular expressions used when handling these tasks. In addition, we identify four gaps in the existing literature, why they matter, and how future research can begin to address them.

The Use of NLP-Based Text Representation Techniques to Support Requirement Engineering Tasks: A Systematic Mapping Review

The development and evaluation of In-Vehicle Information Systems (IVISs) is strongly based on insights from qualitative studies conducted in artificial contexts (e.g., driving simulators or lab experiments). However, the growing complexity of the systems and the uncertainty about the context in which they are used, create a need to augment qualitative data with quantitative data, collected during real-world driving. In contrast to many digital companies that are already successfully using data-driven methods, Original Equipment Manufacturers (OEMs) are not yet succeeding in releasing the potentials such methods offer. We aim to understand what prevents automotive OEMs from applying data-driven methods, what needs practitioners formulate, and how collecting and analyzing usage data from vehicles can enhance UX activities. We adopted a Multiphase Mixed Methods approach comprising two interview studies with more than 15 UX practitioners and two action research studies conducted with two different OEMs. From the four studies, we synthesize the needs of UX designers, extract limitations within the domain that hinder the application of data-driven methods, elaborate on unleveraged potentials, and formulate recommendations to improve the usage of vehicle data. We conclude that, in addition to modernizing the legal, technical, and organizational infrastructure, UX and Data Science must be brought closer together by reducing silo mentality and increasing interdisciplinary collaboration. New tools and methods need to be developed and UX experts must be empowered to make data-based evidence an integral part of the UX design process.

Automotive UX design and data-driven development: Narrowing the gap to support practitioners

Simulink is an example of a successful application of the paradigm of model-based development into industrial practice. Numerous companies create and maintain Simulink projects for modeling software-intensive embedded systems, aiming at early validation and automated code generation. However, Simulink projects are not as easily available as code-based ones, which profit from large publicly accessible open-source repositories, thus curbing empirical research. In this paper, we investigate a set of 1734 freely available Simulink models from 194 projects and analyze their suitability for empirical research. We analyze the projects considering (1) their development context, (2) their complexity in terms of size and organization within projects, and (3) their evolution over time. Our results show that there are both limitations and potentials for empirical research. On the one hand, some application domains dominate the development context, and there is a large number of models that can be considered toy examples of limited practical relevance. These often stem from an academic context, consist of only a few Simulink blocks, and are no longer (or have never been) under active development or maintenance. On the other hand, we found that a subset of the analyzed models is of considerable size and complexity. There are models comprising several thousands of blocks, some of them highly modularized by hierarchically organized Simulink subsystems. Likewise, some of the models expose an active maintenance span of several years, which indicates that they are used as primary development artifacts throughout a project’s lifecycle. According to a discussion of our results with a domain expert, many models can be considered mature enough for quality analysis purposes, and they expose characteristics that can be considered representative for industry-scale models. Thus, we are confident that a subset of the models is suitable for empirical research. More generally, using a publicly available model corpus or a dedicated subset enables researchers to replicate findings, publish subsequent studies, and use them for validation purposes. We publish our dataset for the sake of replicating our results and fostering future empirical research.

/pdf/characteristics-potentials-and-limitations-of-open-source-uh2c2fbe7d.pdf

Characteristics, potentials, and limitations of open-source Simulink projects for empirical research

Internet of Things (IoT) applications promise to make many aspects of our lives more efficient and adaptive through the use of distributed sensing and computing nodes. A central aspect of such applications is their complex communication behavior that is heavily influenced by the physical environment of the system. To continuously improve IoT applications, a staging environment is needed that can provide operating conditions representative of deployments in the actual production environments – similar to what is common practice in cloud application development today. Towards such a staging environment, we present Marvis, a framework that orchestrates hybrid testbeds, co-simulated domain environments, and a central network simulation for testing distributed IoT applications. Our preliminary results include an open source prototype and a demonstration of a Vehicle-to-everything (V2X) communication scenario.

Towards a Staging Environment for the Internet of Things

In recent years, simulations have proven to be an important means to verify the behavior of complex software systems. The different states of a system are monitored in the simulations and are compared against the requirements specification. So far, system states in natural language requirements cannot be automatically linked to signals from the simulation. However, the manual mapping between requirements and simulation is a time-consuming task. Named-entity Recognition is a sub-task from the field of automated information retrieval and is used to classify parts of natural language texts into categories. In this paper, we use a self-trained Named-entity Recognition model with Bidirectional LSTMs and CNNs to extract states from requirements specifications. We present an almost entirely automated approach and an iterative semi-automated approach to train our model. The automated and iterative approach are compared and discussed with respect to the usual manual extraction. We show that the manual extraction of states in 2,000 requirements takes nine hours. Our automated approach achieves an F1-score of 0.51 with 15 minutes of manual work and the iterative approach achieves an F1-score of 0.62 with 100 minutes of work.

/pdf/extraction-of-system-states-from-natural-language-53suxvi7wl.pdf

Extraction of System States from Natural Language Requirements

The testing of software-intensive systems is performed in different test stages each having a large number of test cases. These test cases are commonly derived from requirements. Each test stages exhibits specific demands and constraints with respect to their degree of detail and what can be tested. Therefore, specific test suites are defined for each test stage. In this paper, the focus is on the domain of embedded systems, where, among others, typical test stages are Software- and Hardware-in-the-loop. Monitoring and controlling which requirements are verified in which detail and in which test stage is a challenge for engineers. However, this information is necessary to assure a certain test coverage, to minimize redundant testing procedures, and to avoid inconsistencies between test stages. In addition, engineers are reluctant to state their requirements in terms of structured languages or models that would facilitate the relation of requirements to test executions. With our approach, we close the gap between requirements specifications and test executions. Previously, we have proposed a lightweight markup language for requirements which provides a set of annotations that can be applied to natural language requirements. The annotations are mapped to events and signals in test executions. As a result, meaningful insights from a set of test executions can be directly related to artifacts in the requirements specification. In this paper, we use the markup language to compare different test stages with one another. We annotate 443 natural language requirements of a driver assistance system with the means of our lightweight markup language. The annotations are then linked to 1300 test executions from a simulation environment and 53 test executions from test drives with human drivers. Based on the annotations, we are able to analyze how similar the test stages are and how well test stages and test cases are aligned with the requirements. Further, we highlight the general applicability of our approach through this extensive experimental evaluation. With our approach, the results of several test levels are linked to the requirements and enable the evaluation of complex test executions. By this means, practitioners can easily evaluate how well a systems performs with regards to its specification and, additionally, can reason about the expressiveness of the applied test stage.

https://link.springer.com/content/pdf/10.1007/s10664-020-09815-w.pdf

What Am I Testing and Where? Comparing Testing Procedures based on Lightweight Requirements Annotations

[Context] Simulation is a powerful tool to validate specified requirements especially for complex systems that constantly monitor and react to characteristics of their environment. The simulators for such systems are complex themselves as they simulate multiple actors with multiple interacting functions in a number of different scenarios. To validate requirements in such simulations, the requirements must be related to the simulation runs. [Problem] In practice, engineers are reluctant to state their requirements in terms of structured languages or models that would allow for a straightforward relation of requirements to simulation runs. Instead, the requirements are expressed as unstructured natural language text that is hard to assess in a set of complex simulation runs. Therefore, the feedback loop between requirements and simulation is very long or non-existent at all. [Principal idea] We aim to close the gap between requirements specifications and simulation by proposing a lightweight markup language for requirements. Our markup language provides a set of annotations on different levels that can be applied to natural language requirements. The annotations are mapped to simulation events. As a result, meaningful information from a set of simulation runs is shown directly in the requirements specification. [Contribution] Instead of forcing the engineer to write requirements in a specific way just for the purpose of relating them to a simulator, the markup language allows annotating the already specified requirements up to a level that is interesting for the engineer. We evaluate our approach by analyzing 8 original requirements of an automotive system in a set of 100 simulation runs.

/pdf/a-lightweight-multilevel-markup-language-for-connecting-3nsvjfywsy.pdf

Florian Brokhausen

Papers

Characteristics, potentials, and limitations of open-source Simulink projects for empirical research

Towards a Staging Environment for the Internet of Things

Extraction of System States from Natural Language Requirements

What Am I Testing and Where? Comparing Testing Procedures based on Lightweight Requirements Annotations

A Lightweight Multilevel Markup Language for Connecting Software Requirements and Simulations