The rapid growth of textual data in the digital age presents unique challenges in domain-specific text classification, particularly the scarcity of labeled data for many applications, due to expensive cost of manual labeling work. In this thesis, we explore the applicability of prompt-learning method, which is well-known for being suitable in few-shot scenarios and much less data-consuming, as an emerging alternative to traditional fine-tuning methods, for domain-specific text classification in the context of customer-agent interactions in the retail sector. Specifically, we implemented the entire prompt-learning pipeline for the classification task, and, our investigation encompasses various strategies of prompt-learning, including fixed-prompt language model tuning strategy and tuning-free prompting strategy, along with an examination of language model selection, few-shot sampling strategy, prompt template design, and verbalizer design. In this manner, we assessed the overall performance of the prompt-learning method in the classification task. Through a systematic evaluation, we demonstrate that with the fixed-prompt language model tuning strategy, based on relatively smaller language models (e.g. T5-base with around 220M parameters), prompt-learning can achieve competitive performance (close to 75% accuracy) even with limited labeled data (up to merely 15% of full data). And besides, with the tuning-free prompting strategy, based on a regular-size language model (e.g. FLAN-T5-large with around 770M parameters), the performance can be up to around 30% accuracy with detailed prompt templates and zero-shot setting (no extra training data involved). These results can offer valuable insights for researchers and practitioners working with domain-specific textual data, prompt-learning and few-shot / zero-shot learning. The findings of this thesis highlight the potential of prompt-learning as a practical solution for classification problems across diverse domains and set the stage for future research in this area.

Cybersecurity is still one of the main challenges of the digital era for organizations and individuals alike. Threat modeling is an important tool for building systems that are reliable and secure. The research question for this study is to create a domain specific language (DSL) with the Meta Attack Language (MAL), STRIDE and DREAD. One of the main challenges is to choose a DSL that is suitable for threat modeling. The purpose of the study is to provide people with threat modeling with additional tools that can be used in attack simulations. MAL is a meta language used for creating DSL that can be used for attack simulations. An example of a MAL project that usually serves as a template for other DSL is coreLang, which models the general IT infrastructure. STRIDE is a model used in threat modeling to enumerate and categorization of cyberthreats. DREAD is a model used for risk assessment that scores each threat by a value between one and ten. The proposed method for answering the research question is the Design Research Science Method (DRSM), which is often used for creating artifacts. Evaluation of the results is done with tests written in Java using the Junit framework. The result of the study is the creation of strideLang that maps attack steps in coreLang (MAL implementation of the general IT infrastructure DSL) to STRIDE and DREAD models. The primary source of error in the investigation is the risk assessment with DREAD, which can be somewhat inaccurate depending on what specific DSL is used. It would have been valuable if the study incorporated feedback from domain experts specifically with risk assessment. The nature of the STRIDE and DREAD models is that the models are very subjective in practice. However, this study does provide insights in how a DSL can be created based on DREAD and STRIDE. Future work might investigate a different DSL, incorporate tools such as SecuriCAD and compare different threat models. / Cybersäkerhet är fortfarande en av de främsta utmaningarna i den digitala eran för såväl organisationer som individer. Hotmodellering är ett viktigt verktyg för att bygga tillförlitliga och säkra system. Huvudmålet för denna studie är att skapa ett domänspecifikt språk (DSL) med Meta Attack Language (MAL), STRIDE och DREAD. En av de främsta utmaningarna för att nå målet med studien är att hitta ett domänspecifikt språk som är lämpligt för denna typ av hotmodellering. Syftet med studien är att förse personer som arbetar med hotmodellering med ytterligare verktyg för att kunna använda i sina attacksimuleringar. MAL är ett metaspråk som används för att skapa domän-specifika språk och utföra attacksimuleringar. Ett exempel på ett MAL projekt som oftast används som en mall för att skapa nya domänspecifika och modellerar den generella IT infrastrukturen. STRIDE modellen används för att lista och kategorisera digitala hot. DREAD brukar användas tillsammans med STRIDE och används för att risk bedöma digitala hot genom att betygsätta hoten med ett värde mellan ett och tio. Den valda metoden för att lösa forskningsfrågan är Design Research Science Method (DSRM), som används oftast i samband med skapandet av artefakter. Evaluering av resultatet gjordes med tester skrivna i Java med ramverket JUnit. Studien resulterade med skapande av strideLang som mappar attack steg i coreLang till STRIDE och DREAD modellerna. Den främsta felkällan i denna studie är riskbedömningen med DREAD eftersom noggrannheten på riskbedömningen kan variera från specifika domän i IT infrastrukturen. Det hade varit värdefullt om studien integrera domänexperters bedömning i studien främst för DREAD bedömningen. STRIDE och DREAD modellerna är subjektiva vilket betyder att olika experter kan komma till olika slutsatser för samma hot. Däremot så kan studien förse med intressanta insikter om hur ett domän-specifikt språk kan skapas baserat på DREAD och STRIDE modellerna. Framtida studier kan undersöka en mer specifik domän inom IT infrastrukturen, integrera verktyg som SecuriCAD och jämföra olika modeller som används inom hotmodelleringen

Software tools are being used by experts in a variety of domains. There are numerous software modeling environments tailored to a specific domain expertise. However, there is no consistent approach to generically synthesize a product line of such modeling environments that also take into account the user interaction and experience adapted to the domain. The focus of my thesis is the proposal of a solution to explicitly model user interfaces and interaction of modeling environments so that they can be tailored to the habits and preferences of domain experts. We extend current model-driven engineering techniques that synthesize graphical modeling environments to also take interaction models into account. The formal semantics of our language framework is based on statecharts. We define a development process for generating such modeling environments to maximize reuse through a novel statechart refinement technique. / Les outils logiciels sont utilisés par des experts dans une variété de domaines. Il existe de nombreux environnements de modélisation logicielle adaptés á une expertise spécifique. Cependant, il n’existe pas d’approche cohérente pour synthétiser génériquement une ligne de produits de tels environnements de modélisation qui prennent également en compte l’interaction et l’expérience utilisateur adaptées au domaine. L’objectif de ma thése est la proposition d’une solution pour modéliser explicitement les interfaces utilisateur et l’interaction des environnements de modélisation afin qu’ils puissent étre adaptés aux habitudes et aux préférences des experts du domaine. Nous étendons les techniques d’ingénierie actuelles pilotées par un modéle qui synthétisent des environnements de modélisation graphique pour prendre également en compte les modèles d’interaction. La sémantique formelle de notre cadre linguistique est basée sur des statecharts. Nous définissons un processus de développement pour générer de tels environnements de modélisation afin de maximiser la réutilisation à travers une nouveau technique de raffinement de statecharts.

En ingénierie dirigée par les modèles, les langages spécifiques au domaine (DSL) offrent des notations adaptées à un domaine précis pour représenter ses différents concepts. De nombreux outils permettent la définition de DSLs en explicitant les relations entre un concept et ses représentations. En fonction de la sémantique du domaine, l’ingénieur du langage peut choisir entre des notations textuelles ou graphiques. Les langages de modélisation graphique nécessitent une gestion de la position, la taille et la disposition des éléments visuels afin de maximiser leur expressivité visuelle. La plupart des éditeurs de modélisation manquent de support automatique pour gérer ces propriétés de la syntaxe concrète. Les éditeurs projectionnels permettent aux utilisateurs de se concentrer sur la conception de leur modèle en limitant les modifications de la syntaxe concrète. Cependant, bien qu’ils offrent de multiples notations, ces éditeurs ne permettent pas la création de langage graphique. Dans ce mémoire, nous proposons une nouvelle approche pour concevoir des éditeurs graphiques et projectionnels. Nous avons créé une extension d’un éditeur projectionnel orienté vers le web, Gentleman, qui nous a permis d’extraire différentes exigences. Au cours du mémoire, nous décrivons leurs impacts sur les projections et proposons des lignes directrices ainsi que des exemples d’implémentation. Comme l’édition projectionnelle demande une gestion spécifique de l’interaction, nous présentons différentes approches pour interagir avec les représentations graphiques utilisant les nouvelles informations disponibles dans les projections. Étant donné que la plupart des exigences se concentrent sur la disposition des projections, nous avons défini plusieurs algorithmes simples de disposition qui couvrent une large gamme de structures pouvant être retrouvées dans un éditeur graphique. Enfin, afin d’évaluer cette approche, nous avons exploré la génération de trois éditeurs graphiques et projectionnels pour différents domaines: les machines d’états, les diagrammes de séquences et les partitions de musique. / In model-driven engineering, domain specific-languages (DSL) provide tailored notations towards a specific problem domain to represent its different concepts. Multiple tools allow the definition of DSL by specifying the relations between a concept and its representations. Depending on the semantics of the domain, the language engineer can choose between textual or graphical notations. Graphical modeling languages require proper management of position, size, and layout to maximize their visual expressiveness. Most modeling editors lack automated support to manage these graphical concrete syntax properties. It is a time-consuming effort that affects the understandability of the model. Projectional editors prevent end-users from modifying the concrete syntax so they can focus on the modeling task. However, while they offer multiple notations, these editors lack support for graphical languages. During this thesis, we propose a new approach to design graphical and projectional editors. We created an extension of a web-oriented projectional editor, Gentleman, that allowed us to extract different requirements. During the thesis, we describe their impact on the projections and propose guidelines and examples of implementation. Because projectional editing requires specific management of the interaction, we present multiple approaches to interact with the graphical representations, using the new information available in the graphics. Since most of the requirements were focusing on the disposition of the projection, we define multiple simple layout algorithms that cover a large range of structures that can be found in a graphical editor. Finally, we explore the generation of three graphical and projectional editors for different domains: statecharts, sequence diagrams, and music sheet.

Attack simulations provide a viable means to test the cyber security of a system. The simulations trace the steps taken by the attacker to compromise sensitive assets within the system. In addition to this, they can also estimate the time taken by the attacker for the same, measuring from the initial step up to the final. One common approach to implement such simulations is the use of attack graph, which trace the various dependencies of every step and their connection to one another in a formal way. To further facilitate attack simulations and to reduce the effort of creating new attack graphs for each system of a given type, domain-specific languages are employed. Another advantage of utilizing such a language is that they organize the common attack logics of the domain in a systematic way, allowing for both ease of use and reuse of models. MAL (the Meta Attack Language) has been proposed by Johnson et al. to serve as a framework to develop domain-specific languages [1]. My work is based upon the same. This thesis report presents AWSLang, which can be used to design IT system models in context to the AWS (Amazon Web Services) environment and analyse their weaknesses. The domain specifics of the language are inspired from and based on existing literature. A Systematic Literature Review (SLR) is performed to identify possible attacks against the elements in an AWS environment. These attacks are then used as groundwork to write test cases and validate the specification. / Attacksimuleringar är ett användbart sätt att testa cybersäkerheten i ett system. Simuleringarna spårar de steg som angriparen tog för att försvaga säkerheten av känsliga tillgångar inom systemet. Utöver detta kan de uppskatta hur länge attacken varade, mätt från första till sista steget. Ett gemensamt tillvägagångssätt för att implementera sådana simuleringar är användningen av attackgrafer, som spårar olika beroenden av varje steg och deras koppling till varandra på ett formellt sätt.För att ytterligare underlätta attacksimuleringar och minska ansträngningen att skapa nya attackgrafer för varje system av en given typ, används domänspecifika språk. En annan fördel med att använda ett sådant språk är att det organiserar domänens gemensamma attacklogiker på ett systematiskt sätt, vilket möjliggör både en enkel användning och återanvändning av modeller. MAL (Meta Attack Language) har föreslagits av Johnson et al. att fungera som ramverk för utvecklingen av domänspecifika språk [1]. Mitt arbete är baserat på detsamma.I denna uppsats presenteras AWSLang, som kan användas för att utforma IT-systemmodeller i kontexten av AWS-miljön (Amazon Web Services) och analysera deras svagheter. Språkets domänspecifikationer är inspirerade av och baserade på befintlig litteratur. En systematisk litteraturöversikt görs för att identifiera möjliga attacker mot elementen i en AWS-miljö. Dessa attacker används sedan som grund för att skriva testfall och validera specifikationen.

Structured Product Labels (SPLs), the package inserts that accompany drugs governed by the Food and Drugs Administration (FDA), hold information about Adverse Drug Reactions (ADRs) that exists associated with drugs post-market. This information is valuable for actors working in the field of pharmacovigilance aiming to improve the safety of drugs. One such actor is Uppsala Monitoring Centre (UMC), a non-profit conducting pharmacovigilance research. In order to access the valuable information of the package inserts, UMC have constructed an SPL mining pipeline in order to mine SPLs for ADRs. This project aims to investigate new approaches to the solution to the Scan problem, the part of the pipeline responsible for extracting mentions of ADRs. The Scan problem is solved by approaching the problem as a Named Entity Recognition task, a subtask of Natural Language Processing. By using the transformer-based deep learning model BERT, with domain-specific pre-training, an F1-score of 0.8220 was achieved. Furthermore, the chosen model was used in an iteration of Active Learning in order to efficiently extend the available data pool with the most informative examples. Active Learning improved the F1-score to 0.8337. However, the Active Learning was benchmarked against a data set extended with random examples, showing similar improved scores, therefore this application of Active Learning could not be determined to be effective in this project.

Big Data workflows are composed of multiple orchestration steps that perform different data analytics tasks. These tasks process heterogeneous data using various computing and storage resources. Due to the diversity of application domains, involved technologies, and complexity of data sets, the design and implementation of Big Data workflows require the collaboration of domain experts and technical experts. However, existing tools are too technical and cannot easily allow domain experts to participate in the process of defining and executing Big Data workflows. Moreover, the majority of existing tools are designed for specific applications such as bioinformatics, computational chemistry, and genomics. They are also based on specific technology stacks that do not provide flexible means of code reuse and maintenance. This thesis presents the design and implementation of a Big Data workflow solution based on the use of a domain-specific language (DSL) for hiding complex technical details, enabling domain experts to participate in the process definition of workflows. The workflow solution uses a combination of software container technologies and message-oriented middleware (MOM) to enable highly scalable workflow execution. The applicability of the solution is demonstrated by implementing a prototype based on a real-world data workflow. As per performed evaluations, the proposed workflow solution was evaluated to provide efficient workflow definition and scalable execution. Furthermore, the results of a set of experiments were presented, comparing the performance of the proposed approach with Argo Workflows, one of the most promising tools in the area of Big Data workflows. / Big Data-arbetsflöden består av flera orkestreringssteg som utför olika dataanalysuppgifter. Dessa uppgifter bearbetar heterogena data med hjälp av olika databehandlings- och lagringsresurser. På grund av stora variationen av tillämpningsområden, den involverade tekniken, och komplexiteten hos datamängderna, kräver utformning och implementering av Big Data-arbetsflöden samarbete mellan domänexperter och tekniska experter. Befintliga verktyg är dock för tekniska och vilket försvårar för domänexperter att delta i processen att definiera och genomföra Big Data-arbetsflöden. Dessutom är majoriteten av befintliga verktyg utformade för specifika tillämpningar, som bioinformatik, beräkningskemi och genomik. Verktygen är också baserade på specifika teknikstackar som inte erbjuder flexibla metoder för att kunna underhålla och återanvända kod. Denna avhandling ämnar att presentera design och implementering av en Big Data-arbetsflödeslösning som utnyttjar ett domänspecifikt språk (DSL) för att dölja komplexa tekniska detaljer, vilket gör det möjligt för domänexperter att delta i processdefinitionen av arbetsflöden. Arbetsflödeslösningen använder en kombination av mjukvaruutrustningsteknik och meddelande-orienterad mellanvara (MOM) för att möjliggöra en mer skalbar körning av arbetsflöden. Tillämpningslösningen demonstreras genom att implementera en prototyp baserad på ett verkligt dataflöde. Efter en granskning av de genomförda testerna modifierades den föreslagna arbetsflödeslösningen för att uppnå en effektiv arbetsflödesdefinition och skalbar körning. Dessutom presenteras resultaten av en uppsättning experiment där man jämför skalbarheten för det föreslagna tillvägagångssättet med Argo Workflows, ett av de mest lovande verktygen inom Big Data-arbetsflöden

[pt] Pesquisas são pervasivas no mundo moderno e seu uso vai de medidas de satisfação de consumidores ao rastreamento de tendências econômicas globais. No centro do processo de pesquisa está a coleta de dados que é, usualmente, assistida por computador. O desenvolvimento de software destinado à coleta de dados em pesquisas envolve a codificação de questionários que variam de simples sequências de questões abertas à questionários complexos nos quais validações, cálculo de dados derivados, gatilhos para garantia de consistência e objetos de interesse criados dinamicamente são a regra. A especificação do questionário é parte dos metadados da pesquisa e é um fator chave na garantia da qualidade dos dados coletados e dos resultados atingidos por uma pesquisa. São os metadados da pesquisa que estabelecem a maior parte dos requisitos para os sistemas de suporte a pesquisas, incluindo requisitos para o software de coleta de dados. À medida que a pesquisa é planejada e executada, esses requisitos devem ser compreendidos, comunicados, codificados e implantados, numa sequência de atividades que demanda técnicas adequadas para que a pesquisa seja eficaz e efetiva. A Engenharia Orientada a Modelos (Model Driven Engineering) propõe estratégias que visam alcançar esse objetivo. Neste contexto, esta dissertação propõe o uso de Linguagens de Domínio Específico (Domain-specific Languages - DSLs) para modelar questionários, apresenta um protótipo e avalia DSLs como uma técnica para diminuir a distância entre especialistas de domínio e desenvolvedores de software, incentivar o reuso, eliminar a redundância e minimizar o retrabalho. / [en] Surveys are pervasive in the modern world with its usage ranging from the field of customer satisfaction measurement to global economic trends tracking. At the core of survey processes is data collection which is, usually, computer aided. The development of data collection software involves the codification of questionnaires which vary from simple straightforward questions to complex questionnaires in which validations, derived data calculus, triggers used to guarantee consistency and dynamically created objects of interest are the rule. The questionnaire specification is part of what is called survey metadata and is a key factor for collected data and survey quality. Survey metadata establishes most of the requirements for survey support systems including data collection software. As the survey process is executed, those requirements need to be translated, coded and deployed in a sequence of activities that demands strategies for being efficient and effective. Model Driven Engineering enters this picture with the concept of software crafted directly from models. In this context, this dissertation proposes the usage of a Domain Specific Language (DSL) for modeling questionnaires, presents a prototype and evaluates DSL as a strategy to reduce the gap between survey domain experts and software developers, improve reuse, eliminate redundancy and minimize rework.

Hadron Therapy (HT) exploits properties of ion radiation to gain therapeutic advantages over existing photon-based forms of external radiation therapy. However, its relative superiority and cost-effectiveness have not been proven for all clinical situations. Establishing a robust evidence base for the development of best treatment practices is one of the major challenges for the field. This thesis investigates two research infrastructures for building this essential evidence. First, the thesis develops main components of a metadata-driven software architecture for the collection of clinical information and its analysis. This architecture acknowledges the diversity in the domain and supports data interoperability by sharing information models. Their compliance to common metamodels guarantees that primary data and analysis results can be interpreted outside of the immediate production context. This is a fundamental necessity for all aspects of the evidence creation process. A metamodel of data capture forms is developed with unique properties to support data collection and documentation in this architecture. The architecture's potential to support complex analysis processes is demonstrated with the help of a novel metamodel for Markov model based simulations, as used for the synthesis of evidence in health-economic assessments. The application of both metamodels is illustrated on the example of HT. Since the biological effect of particle radiation is a major source of uncertainty in HT, in its second part, this thesis undertakes first investigations towards a new research facility for bio-medical experiments with ion beams. It examines the feasibility of upgrading LEIR, an existing accelerator at the European Organisation for Nuclear Research (CERN), with a new slow extraction and investigates transport of the extracted beam to future experiments. Possible configurations for the slow-resonant extraction process are identified, and designs for horizontal and vertical beam transport lines developed. The results of these studies indicate future research directions towards a new ion beam facility for biomedical research.

Les divers intervenants qui décrivent, étudient et réalisent un système complexe ont besoin de points de vue adaptés à leurs préoccupations. Cependant, dans le contexte de l’Ingénierie Dirigée par les Modèles, les moyens pour définir et mettre en œuvre ces points de vue sont, soit trop rigides et inadaptées, soit totalement ad hoc. De plus, ces différents points de vue sont rarement indépendants les uns des autres. Dès lors, il faut s’attacher à identifier puis décrire les liens/les correspondances qui existent entre les points de vue pour enfin pouvoir vérifier que les réponses apportées par les différents intervenants constituent un tout cohérent.Les travaux exposés dans cette thèse permettent la définition de langages dédiés basés sur UML pour les points de vue. Pour cela, une méthode outillée qui analyse la sémantique des descriptions textuelles des concepts du domaine que l’on souhaite projeter sur UML est proposée afin de faciliter la définition de profils UML. Les résultats obtenus sur les cas d’études concrets étudiés sont encourageants et dépassent les résultats des outils existants. Pour définir des points de vue basés sur des profils UML, cette thèse propose une méthode qui permet au méthodologiste d’expliciter le point de vue voulu. Un outil génère ensuite l’outillage qui met en œuvre ce point de vue dans un environnement de modélisation ainsi que le langage dédié correspondant là où la pratique actuelle repose sur une mise en œuvre essentiellement manuelle.Pour assister l’identification des liens entre points de vue, cette thèse propose là aussi d’analyser la sémantique des descriptions textuelles des concepts des langages utilisés par les points de vue. Utilisée en complément des heuristiques syntaxiques existantes, l’approche proposée permet d’obtenir de bons résultats lorsque les terminologies des langages analysés sont éloignées. Un cadre théorique basé sur la théorie des catégories est proposé pour expliciter formellement les correspondances. Pour utiliser ce cadre, une catégorie pour les langages basés sur UML a été proposée. Afin de pouvoir également expliciter les correspondances entre les modèles issus de ces langages, la catégorie des ontologies OWL est utilisée. Une solution est proposée pour caractériser des correspondances plus complexes que la simple équivalence. Ce cadre théorique permet la définition formelle de relations complexes qui permettront de raisonner sur la cohérence de la description de l’architecture. Une fois la description de l’architecture intégrée en un tout en suivant les correspondances formalisées, la question de la cohérence est abordée. Les expérimentations faites sur un cas d’étude concret pour vérifier la cohérence à un niveau syntaxique donnent des résultats pratiques satisfaisants. Les expérimentations menées sur le même cas pour vérifier la cohérence à un niveau sémantique ne donnent pas de résultats pratiques satisfaisants. / The various stakeholders who describe study and implement a complex system require viewpoints that are dedicated to their concerns. However, in the context of Model-Driven Engineering, approaches to define and implement those viewpoints are either too rigid and inappropriate or completely ad hoc. In addition, those various viewpoints are rarely independent from each other. Therefore, we must strive to identify and describe the relationships/correspondences between the viewpoints in order to be able to verify that the parts of the solution given by the various stakeholders form a consistent whole.The work presented in this thesis provides a way to define dedicated languages based on UML for the viewpoints. For this, a method that analyzes the semantics of the textual descriptions of the concepts of the domain we want to map to UML has been implemented to facilitate the definition of UML profiles. The results we get on the concrete test cases we considered are encouraging and go beyond results of existing tools. To define a viewpoint based on some UML profiles, this thesis provides a method that lets the methodologist make explicit the viewpoint he/she wants. A tool can then generate the tooling that implements this viewpoint in a modeling environment together with the corresponding dedicated language while current practice is based on an implementation essentially manual.To assist the identification of relationships between the viewpoints, this thesis proposes again to analyze the semantics of textual descriptions of concepts of the languages used by the viewpoints. Used in combination with existing syntactic heuristics, the proposed approach provides good results when the terminologies of the languages that are analyzed are far apart. A theoretical framework based on category theory is provided to make explicit formally correspondences. To use this framework, a category for languages based on UML has been proposed. To be able to make explicit the correspondences between the models of those languages as well, the category of OWL ontologies is used. A solution is proposed to characterize correspondences that are more complex than the simple equivalence relationship. This theoretical framework provides a way to define formally complex relationships that can be used to verify the consistency of the architectural description. Once the description of the architecture has been integrated according to the formal correspondences, the issue of consistency is considered. The experiments carried out on a concrete test case to verify consistency on a syntactic perspective give satisfactory practical results. The experiments carried on the same test case to verify consistency on a semantic perspective don’t give satisfactory practical results.