Chalmers Open Digital Repository

Välkommen till Chalmers öppna digitala arkiv!

Här hittar du:

Studentarbeten utgivna på lärosätet, såväl kandidatarbeten som examensarbeten på grund- och masternivå
Digitala specialsamlingar, som t ex Chalmers modellkammare
Utvalda projektrapporter

Forskningspublikationer, rapporter och avhandlingar hittar du i research.chalmers.se

Enheter i Chalmers ODR

Välj en enhet för att se alla samlingar.

Visar 1 - 2 av 2

Senast publicerade

Explainable AI for Automatic Document Classification in Regulated Finance
(2026) Stenhammar, Zachris; Alavala, Praveen
The increasing volume of digital documents in regulated financial environments has created significant challenges related to information security, regulatory compliance, and efficient information management. Financial institutions routinely process sensitive information, including internal business data, customer records, and regulatory documents, where incorrect handling or classification may result in legal, financial, and reputational consequences. Despite the importance of information classification, the process is often performed manually, making it inconsistent, time-consuming, and difficult to scale. These challenges motivate the need for automated and trustworthy document classification systems that can support regulated organizations while maintaining transparency and accountability. This thesis investigates the use of transformer-based language models for automatic document classification in regulated financial environments, with a particular focus on explainability and auditability. The study explores how contextual and semantic information within documents can be used to distinguish between different information sensitivity levels, including Public, Internal, Confidential, and Strictly Confidential classifications. To address privacy and regulatory constraints, the work utilizes a synthetic and semi-controlled dataset .generated using a controlled template-based synthetic document generation methodology with constrained vocabulary, document structures, and contextual patterns designed to reflect the structural and linguistic characteristics of financial documents while avoiding the use of sensitive real-world data. The proposed system is based on a fine-tuned transformer architecture combined with explainable artificial intelligence (XAI) techniques. Attention-based explanations and Integrated Gradients feature attribution methods are integrated into the classification pipeline to provide insight into the model’s decision making process. The explainability analysis investigates whether the generated explanations align with meaningful contextual indicators associated with document sensitivity and whether they can support transparency, trust, and compliance requirements within regulated financial settings. The experimental results demonstrate that transformer-based models can effectively learn contextual patterns related to information sensitivity within the controlled dataset while also providing interpretable explanations of classification decisions. The study further analyzes explanation consistency, confidence behaviour, robustness against external documents, and potential shortcut learning effects. Since both the training and evaluation data were generated using the same controlled template-based document generation methodology, the results should be interpreted within the context of this experimental setting. Although separate documents were used for training and evaluation, come from the same dataset share similar linguistic and structural characteristics. Therefore, further evaluation using independent datasets is required to assess the generalizability of the proposed approach. This work contributes to the growing field of explainable AI in regulated industries by demonstrating how modern natural language processing techniques can be combined with explainability methods to support secure, transparent, and trustworthy information classification in financial organizations, while also highlighting the importance of independent evaluation when using controlled and synthetic data.
Accessibility in Motion - Developing context-specific accessibility recommendations and re-designing the Thule mobile application
(2026) Akin, Enes; Blom, Benjamin
Mobile applications are increasingly used in dynamic environments, yet existing accessibility frameworks, such as WCAG 2.2, primarily address static use cases. This Master’s thesis investigates the limitations of current guidelines when faced with Situationally Induced Impairments and Disabilities (SIIDs), including screen glare and reduced dexterity from physical multitasking. In collaboration with Thule Sweden AB, the study employs a Research through Design methodology within a Double Diamond framework. By combining contextual inquiry, general and targeted surveys, and heuristic evaluations, the research identifies critical pain points in active mobile usage. The project culminates in the formulation of 14 context-specific accessibility recommendations and the iterative development of a high-fidelity mobile prototype. This proposed final design features an "Active Mode", an interface variation that re moves non-essential visuals to prioritize cognitive clarity. The findings demonstrate that digital accessibility in high-motion environments relies heavily on functional minimalism, ergonomic adaptability, and error tolerance, ensuring usability supersedes traditional aesthetics.
Design and Evaluation of a Digital Educational Tool Benefiting from Educational Tools in a Distraction-Free Environment
(2026) AlAllaf, Noor
This thesis explores assisting students in having better learning experiences by benefiting from digital educational tools while not getting digitally distracted. Digital distractions come from entertainment platforms like social media, streaming services, games, and notifications that pull focus from studying and get the students unfocused. The project starts by evaluating the strengths and limitations of existing digital educational tools in the market and in schools. The process also involves looking into research conducted on using these tools for learning and educational purposes. The aim then is to design and evaluate prototypes of possible ideas based on user feedback, and interaction design and human-computer interaction principles. The objective is to deliver a final solution in form of a descriptive concept and prototype to effectively assist students in their educational journey. The project follows a user-centered design approach, mainly the triple diamond design method. The project starts with requirement gathering, prototyping, and iterative evaluation involving the target group, students. The methods used also include sending out surveys for pilot studies and conducting A/B testing to compare and evaluate the concepts. The research resulted in the development and evaluation of multiple prototypes, with findings indicating a user preference for a software-based "Study Space" solution integrated into users’ personal devices: phones, pads, and laptops. This outcome provides a clear direction for creating focused digital learning environments that leverage educational technology effectively.
Defining AI Roles in Requirements Elicitation and Analysis
(2026) Li, Yuhan; Ren, Keyu
Artificial intelligence is increasingly used to support Requirements Engineering (RE), especially in text-intensive activities such as summarizing stakeholder input, extracting candidate requirements, and organizing requirements-related information. Yet in early RE, where requirements elicitation and requirements analysis depend on contextual interpretation, incomplete information, and stakeholder negotiation, it remains unclear how responsibilities should be allocated between human practitioners and AI systems. This thesis investigates how human–AI roles are configured in requirements elicitation and requirements analysis, and how practitioners evaluate the benefits, risks, and trust conditions associated with these configurations. A sequential exploratory mixed-methods design was used, combining six semi-structured interviews with soft ware engineering practitioners and a follow-up survey with 67 valid responses. Interview data were analyzed thematically and used to inform the survey design, while survey data were analyzed descriptively. The findings show that AI involvement in early RE is task-contingent rather than general. Practitioners were most willing to use AI as an assistant or preliminary processor in bounded, information-heavy, and verifiable tasks, including summarizing, extracting, organizing, and structuring requirements-related material. By contrast, tasks involving stakeholder interaction, interpretation of implicit needs, prioritization, trade-off decisions, and accountability remained strongly human-led. AI-assisted configurations were perceived as the most useful, whereas AI-only configurations were rare and associated with the highest perceived risk. Trust in AI was therefore conditional rather than general: it increased when outputs were easy to verify and tasks were low in decision criticality, and decreased when work depended on contextual judgment, sensitive information, or consequential decisions. Based on exploratory mixed-methods evidence, this thesis proposes a practitioner informed role-boundary framework at the task level for AI involvement in early RE. The framework clarifies where AI can productively support elicitation and analysis work and where human judgment should remain dominant. Overall, the study argues that AI should be integrated into early RE as a calibrated collaborative support tool rather than as a substitute for human practitioners.
Further Application of Progressive Verification
(2026) Heljeberg, Mia; Nyberg, Arvid
A digital signature is a fundamental cryptographic primitive that provides authenticity and integrity by allowing anyone with a public key to verify that a message was produced by a signer with a corresponding secret key. Such verifications typically produce a binary output only when the process is finished. In contrast, progressive verification (PV) performs verification in smaller incremental steps, gradually building confidence in the signature’s validity over the course of the process. Progressive verification offers several key advantages for post-quantum cryptographic (PQC) schemes on resource constrained devices as it allows for early rejection of invalid inputs and supports adjustable soundness (allowing for a trade-off between security and efficiency). Furthermore, PV can shrink the public key size which addresses a common challenge of PQC schemes. This thesis explores the design and applicability of PV on post-quantum secure digital signature schemes currently involved in the NIST PQC standardisation process. The approach utilises a compiler framework developed by Boschini et al [1] which transforms matrix-vector based (Mv-style) verifications into progressive ones. We explore whether this approach extends to further multivariate quadratic (MQ) schemes as well as to code based schemes. In addition, we investigate whether the compiler can be applied to zero-knowledge proofs, thereby addressing the broader applicability of progressive verification beyond digital signatures. By identifying the matrix-vector structure in the schemes and analysing how the compiler interacts with the verification steps, we assess correctness preservation, security aspects, and practical feasibility. Our findings show that the PV compiler applies cleanly to the MQ-based scheme Unbalanced Oil and Vinegar (UOV), enabling gradual verification without modifying the signing or key-generation algorithms. For code-based schemes, we demonstrate that PV is not applicable to the Codes and Restricted Objects Signature Scheme (CROSS), despite it containing a matrix-vector multiplication in the verification. Progressive verification was also shown to be partially applicable to the verification of a zero-knowledge proof. Overall, this thesis expands the set of post-quantum digital signature schemes known to support progressive verification and highlights design features that make a scheme compatible with PV. These insights can guide both future implementations of PV and the development of new PQC schemes intended for constrained environments.