Question Answering (Q&A) community forums provide an open and collaborative environment where users can post subjective questions and answers based on their life experiences or knowledge of specific domains. User interaction in online forums generates thousands of comments on different subjects, resulting in a massive amount of rich text every year. Posts in Q\&A online communities contain rich information that helps NLP scientists and developers propose approaches to help questioners find answers to their questions based on previously submitted similar questions. However, community questions are short and require additional textual information for automatic matching with the relevant long answers. Explicit information about the question is crucial for accurately predicting the ranked list of answer candidates. Community answer retrieval aims to answer new user questions by leveraging answers from previous users' questions. Automated Learning-to-Rank (LTR) models for Community Question Answering (CQA) require additional explicit information to appropriately expand questions, enabling the identification of an accurate ranked list of answer candidates based on their relevance to the question. This doctoral thesis proposes an investigation into the importance of different explicit information for expanding user questions in transformer-based ranking models. User tags and question descriptions are the explicit information I observed in this study to expand user question information. A key contribution of this study is a novel automated tag classification approach for identifying domain-specific tags in question descriptions, which enables a comparison of ranking model performance by incorporating both user- and automatically selected tags as inputs. The proposed automatic tag classification helps users find adequate tags to summarise a long question description into a tag set. This study also presents new annotated datasets containing thousands of user questions in two domains (personal finances and home improvements). This thesis concludes that incorporating explicit user information for question expansion enhances the predictive performance of LTR models in sorting the answer candidate list by relevance. Among the models employed in this study, those using question descriptions as additional information to expand user questions outperform other input configurations, achieving precision rates exceeding 90\% for rank-aware measures such as Mean Reciprocal Rank (MRR) and Mean Average Precision (MAP). The experiments further reveal that including automatically selected tags and user tags as part of the input yields comparable performance across all LTR CQA models. This study helps retrieval systems answer new questions based on corresponding comments used to answer old questions.

Overweight and obesity affect billions of adults. Limiting the scope to nutrition, help must offer diet guidance that fits easily into daily life. Health-aware meal plan recommendations must integrate user preferences, nutrition goals, and everyday context, like time and date. However, existing recipe data sets and recommender systems lack general health guidelines, they miss capturing user-recipe interactions, or do not provide explanations or other persuasion methods to improve on the recommendations effectiveness. This thesis closes these gaps and evaluatesproposed systems for health-aware meal planning. We give an extensive background on nutrition and persuasion, an overview of LLMs, and introduce related work in terms of data sets and recommendation approaches. Then, our contribution can be split into three parts. Our first contribution is HUMMUS, a large, linked food graph with approximately 500 000 recipes, 300 000 users, 1.9 million interactions, nutrition scores, and semantic links to FoodOn (a food product hierarchy) and FoodData (nutritional ingredient and food product data). It enables reasoning over ingredient classes and nutrient checks. Its size, presence of interactions, and diverse nutrition information improve usability when compared to related data sets. We discuss sparsity, preprocessing, and filtering options that support different experiments. Our second contribution is a study of food-centric behaviour. We collected questionnaire data and meal logs with an online app to explore correlations between food skills (FS), cooking skills (CS), and intake context. The sample includes 78 participants. Results show that FS and CS correlate. These findings and the app help to explore the domain of food behaviour and to improve further studies and recommender systems. Our third contribution is a framework for a daily meal plan recommendation, including explanations. We design and implement two approaches that accept natural language input.Our first approach, a KBQA system, recommends daily meal plans, supports multiple nutrient constraints, and produces path-grounded explanations. It expands its baseline PFoodReq by those features, but also inherits scalability issues and enforces soft constraints. We then propose the FoodRAG architecture, an LLM-based, agentic, modular pipeline that performs recipe retrieval, validation, ingredient substitution, and step-wise explanation. Despite a higher computational cost than the baselines, its extensive adaptability produces more targeted meal-plan recommendations. Both systems are evaluated and compared to baseline recommendation approaches (collaborative and content-based). Persuasion is addressed by integrating fitting gamification methods, providing explanation generation as well as generic UI aids. As collaborative methods underperform on sparse data like HUMMUS, we evaluate simple content-based baselines, the KBQA system, and three FoodRAG variants. The Text-to-andas retriever (an LLM with natural language input questions to generate Pandas queries used on the FoodRAG data set) gives the strongest recipe retrieval results. Keywords: Health-aware recommendation, Meal planning, Knowledge graphs, Recipe data set, Explainability, Persuasion, KBQA, RAG.

Software testing plays a critical role in ensuring software quality, yet it is frequently undervalued and underutilized by both practitioners and learners. A persistent lack of motivation, driven by perceptions of testing as tedious and less creative, hampers the adoption of rigorous testing practices in both industry and education. This thesis investigates the potential of gamification and serious games to address these motivational and pedagogical challenges, aiming to enhance engagement, learning outcomes, and testing behavior. Grounded in a mixed-methods research design, this work explores the application of game design principles to the domain of software testing through tool creations, empirical studies, and experimental interventions. It identifies key technical, organizational, and motivational barriers to effective testing based on large-scale survey data and qualitative insights. Drawing on these findings, the research develops and evaluates a series of game-based and gamified learning experiences across secondary, higher education, and professional development contexts. The results demonstrate that both gamification and serious games can positively impact how learners and developers engage with software testing, leading to improved motivation, conceptual understanding, and testing practices. Furthermore, this thesis offers theoretical and practical contributions by outlining design considerations, implementation strategies, and empirical evidence for integrating playful learning and feedback mechanisms into software testing education and practice. By reimagining software testing as an interactive and intrinsically rewarding activity, this work bridges the gap between technical rigor and learner engagement, contributing to a more sustainable and effective approach to testing within the broader field of software engineering.

Database systems are at the core of modern applications, providing essential services for storing, querying, and managing data. However, they do not remain static; they continuously evolve in response to external changes. Two primary external factors drive this evolution. On the one hand, database vendors release new versions of the database software, introducing new features, bug fixes, and security patches, resulting in the ongoing evolution of the database binary. On the other hand, developers modify the database schema, which defines the structure of the stored data, to accommodate new data structures required by client applications. This cumulative dissertation investigates the challenges of binary evolution and schema evolution in database systems. These two types of evolution represent disruptive changes at both ends of the stack of a database system from the perspective of externally driven changes. We examine various families of database systems, including relational, NoSQL document and key-value stores, as well as distributed variants. The state-of-the-art approach to address the binary evolution of a database system involves a restart: The database system is shut down, the binary replaced, and restarted. While this approach may be acceptable for full feature upgrades, it poses significant overhead when fixing bugs or patching vulnerabilities. Each restart leads to service interruptions and the loss of the in-memory state. Moreover, such downtimes are particularly problematic for cloud providers offering managed Database-as-a-Service. They cannot predict when a downtime is acceptable for customers, making it challenging to promptly patch critical vulnerabilities without disrupting service. To overcome these issues, we investigate live patching as a radically novel approach to binary evolution. Live patching allows code changes to be applied in-memory to the running application without necessitating a restart. We provide the first systematic evaluation of live patching for database systems. Using the relational database management system MariaDB as a case study, our comprehensive evaluation of single-instance, multi-threaded database systems demonstrates that live patching is a viable alternative to conventional restarts. We examine the interplay between various query workloads and live patching methods, analyzing their impact on transaction throughput and tail latencies. Additionally, we explore how the size of the in-memory state impacts live patching, using the in-memory key-value store Redis, and evaluate real-world patches applied to both MariaDB and Redis. Our findings provide insights into database-specific factors that influence live patching. These insights enable informed decisions about the latency overhead clients may experience during live patching. While single-instance database systems suffer complete service outages during restarts, database clusters are designed to mitigate such disruptions. When a node is taken offline, its workload can be redistributed to other nodes. The state-of-the-art approach to patch a database cluster is a rolling update, where each node is sequentially shut down, patched, and restarted, with other nodes taking over the workload of the node undergoing maintenance. Ideally, this patching process should (1) incur no downtime, (2) avoid noticeable performance degradations such as latency spikes, (3) maintain stable resource consumption, and (4) be executed promptly. Unfortunately, conventional rolling updates often fail to meet these goals. To address these limitations, we extend live patching to database clusters, demonstrating its feasibility and applicability using the distributed in-memory key-value store Redis Cluster as a case study. In a systematic evaluation, we demonstrate that our live patching approach satisfies all four aforementioned requirements, whereas conventional rolling updates do not. To enable seamless live patching, we develop and evaluate two patch distribution mechanisms. Both automate patch distribution across the cluster, with one prioritizing distribution speed and the other designed to accommodate the elasticity of the cluster. Finally, we offer general guidelines for integrating live patching and automatic patch distribution into distributed database systems, demonstrating their successful application using a primary-replica setup of the relational database management system PostgreSQL. To provide a holistic change management approach for database systems, it is crucial to address not only binary evolution but also the evolution of the database schema. The database schema defines the structure of stored data and serves as a contract between the database system and the applications that interact with it. When it evolves, it raises a critical question: Are the schema changes backward compatible, or do they introduce breaking changes for client applications? Unlike relational database management systems, NoSQL database systems do not follow a unified schema standard; instead, schemas are tailored to the specific data format being stored. For semi-structured data, JavaScript Object Notation (JSON) is widely used, with JSON Schema serving as the de-facto schema language for JSON data. To address the challenge of schema evolution for JSON Schema-based database schemas, we explore the concept of JSON Schema containment. JSON Schema containment can be used to decide whether the language defined by one JSON Schema is a subset of another. Our evaluation of existing JSON Schema containment checkers, applied to a diverse set of real-world schemas, reveals significant limitations. These tools often lack support for advanced JSON Schema features or yield inconsistent results, and therefore lack in consensus. Recognizing the absence of comprehensive schema management solutions for JSON-based NoSQL document stores that use JSON Schema for data validation, we propose a novel DevOps workflow implemented through our prototype tool, Josch. Josch is geared for MongoDB and integrates third-party JSON Schema extraction and containment checking solutions to assist developers and DevOps teams in efficiently managing schema changes. It enables the extraction of JSON Schemas from existing JSON documents when an explicit schema is absent and employs containment checking to ensure that schema changes remain backward compatible, thereby preventing breaking changes for client applications. In summary, our research provides a comprehensive approach to managing externally driven changes in database systems. By unifying our research on live patching for binary evolution with our solutions for schema evolution, we enable seamless updates to the database binary without disrupting services and protect client applications from breaking changes due to schema modifications.

Cryptographic keys are fundamental components for ensuring security in digital systems. To ensure reliable key generation and management, various technical concepts have been developed, primarily based on dedicated hardware components such as Trusted Platform Modules (TPMs). However, many modern systems, especially small resource-constrained devices, typically lack hardware support for secure key generation and management. To address these limitations, Physical Unclonable Functions (PUFs) have proven to be an effective solution for key generation, device authentication, and identification tasks. PUFs leverage inherent variations in hardware components to produce unique, device-specific keys. For a well-designed PUF, these keys can be reproduced reliably on the same device but are practically impossible to clone. Various types of PUFs exist, including those that exploit slight delay differences in circuits with symmetric paths. Others rely on physical characteristics of components already present in the computing system, such as SRAM or DRAM. However, many of these constructions rely on technologies that could be replaced by emerging ones in the future. Such a replacement may involve a transition from traditional memory technologies, such as SRAM, DRAM, and flash memory, to emerging Non-Volatile Memories (NVMs), including Ferroelectric RAM (FRAM), Magnetoresistive RAM (MRAM), and Resistive RAM (ReRAM). These new technologies, in turn, necessitate innovative hardware security solutions for generating intrinsic hardware fingerprints, ensuring security for next-generation embedded devices. Furthermore, the integration of nanomaterials, such as carbon nanotubes, into processor architectures and the adoption of reconfigurable hardware platforms like Field-Programmable Gate Arrays (FPGAs) require the development of specifically tailored cybersecurity solutions. This dissertation aims to develop hardware-based security mechanisms for these types of devices by designing new PUF constructions and demonstrating their practical applications. One focus lies on PUFs extracted from nanomaterials and emerging circuit elements, particularly memristive devices and Carbon NanoTube Field-Effect Transistors (CNT-FETs). For memristive devices, which form the basis of ReRAM memory, this work analyzes methods ranging from simple binary quantization to advanced techniques exploiting device-specific response patterns. In the case of CNT-FETs, custom-fabricated wafers are developed to construct PUFs with optimal properties, such as high robustness, uniformity, and entropy, even under varying environmental conditions. These conditions include fluctuations in ambient temperature. Based on an analysis of fundamental system components, this work evaluates the feasibility of deriving PUFs from fully integrated circuits. A specific focus is placed on emerging non-volatile memory technologies, assessing their potential for PUF applications. To achieve PUF behavior in these memory devices, techniques such as intentional timing manipulation, induced bit flips through row hammering, and variations in supply voltage are examined. These resulting bit flips can be exploited as PUF responses. Additionally, transforming raw PUF responses into cryptographically usable keys and integrating specific PUFs into practical applications are core components of this work. The demonstrated practical applications include an innovative architecture for encrypting and binding data to non-volatile memory modules, implemented on Multiprocessor System-on-Chips (MPSoCs) incorporating FPGAs. This architecture enables the storage of confidential data on non-volatile memory while simultaneously using the same module as a PUF, without requiring separate memory partitions solely for the PUF functionality. Finally, practical applications of hardware fingerprints in the automotive sector are demonstrated, including an FPGA-based implementation to maintain security while preserving the temporal determinism of time-critical messages. These goals are met through the use of hardware-implemented cryptographic algorithms coupled with an FPGA-based ring oscillator PUF. To summarize, this work presents new types of PUF implementations, starting with nanomaterials and emerging circuit elements, extending to PUFs derived from integrated circuits, and demonstrates innovative solutions for their integration into MPSoC-based architectures.