Masteruppsatser

Permanent URI for this collectionhttps://gupea-staging.ub.gu.se/handle/2077/22012

Browse

Now showing 1 - 20 of 313

Balancing Fun and Function: Exploring Game Design Features for Long-Term Engagement in Anxiety-Relief Serious Games
(2025-10-09) Schröter, Pia; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Anxiety disorders are one of the most common psychological disorders, yet many people who are experiencing anxiety-related symptoms often remain undetected and untreated. To address this issue, more accessible and engaging solutions, such as serious games, are needed. This thesis explores how psychological practices for treating anxiety can be integrated with game design features that promote replayability without compromising therapeutic value. Various established therapeutic methods and game design features have been explored in order to create an effective mobile prototype. The final prototype featured three mini-games based on breathing exercises, relaxation-based exercises, and attention bias modification. It also incorporated additional features such as an in-game store, a streak system, a scoreboard and more to enhance user engagement. A playtest with 22 people was conducted to evaluate the prototype’s effectiveness in reducing anxiety and promoting continued use. Results indicated a reduction in self-reported anxiety and stress levels, and high user interest in replaying the prototype. Feedback from the testers highlighted the importance of customisation, as responses to the mini-games varied depending on individual preferences. The findings suggest that customisation and a diverse range of therapeutic content are key to supporting both anxiety relief and user engagement. Further research is needed to assess the prototype’s long-term impact.
Characteristic Stripe Pattern Masks Creation for Driving Maneuvers Identification Using Synthetic Data
(2025-10-09) Myroshnychenko, Svitlana; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Previous research has demonstrated that encoding n-dimensional driving data using space-filling curves reveals visual patterns that we call CSPs, which repeat among trajectories of the same type (e.g., roundabout passings, turns, braking, etc.). Unlike traditional methods, which rely on a limited number of real data samples to manually create binary CSP-masks for event identification, our approach systematically creates CSP-masks based on synthetic data. With this, we can explore in a structured way combination of factors such as speed, acceleration, and more that may help to determine the performance of software components, which process multi-dimensional, time-series data for pattern identification, to support testing of software components. To systematically create these masks for event identification, we compile the CSPs for each type of maneuver and cluster occurring stripes (same as clustering the indices).
Machine Learning on the football field: Predicting match performance through GPS data
(2025-10-09) Alvarado Pachón, Natalia; Schelling, Laura; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This master thesis evaluates machine learning models for predicting football match performance from GPS-based practice data. The growing use of GPS-based wearable tracking in professional football underscores the need for approaches that transform large datasets into practical insights for teams, while also contributing new methods and strategies to the scientific literature. Random Forests (RF), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN) are tested on targets representing aerobic capacity (endurance in sustained activity), anaerobic capacity (ability to perform high-intensity efforts without oxygen), and explosiveness (short bursts of speed and power), using both non-overlapping rolling and adaptive feature windows. Three prediction strategies are compared: row-to-row, where each input window is paired with its corresponding target window; all-input-to-row, where the entire input sequence is used to predict each target row independently; and all-inputplus- previous-row to-row, which extends the second strategy by incorporating the previous target as an additional input. The results show that all models outperform linear regression in the last two strategies, with RF performing best for aerobic and anaerobic metrics and CNN and RNN for explosiveness. RF also provides feature importance scores, indicating that linear acceleration from the practice day immediately preceding the match is the strongest predictor. Angular velocity and angular jerk from the fourth and third practice days before the match also emerge as key factors, suggesting that strenuous training loads in the days leading up to competition may play a decisive role in match performance. CNN and RNN, in contrast, function as black-box models and do not directly provide interpretable insights into the relative importance of input features. Regarding windowing techniques, adaptive windowing reduces the Root Mean Squared Error (RMSE), which highlights a possible gain from moving into this approach for sport analysis. These findings offer practical insights for sports training and show how machine learning can turn wearable sensor data into useful performance metrics.
Multi-Turn Contextual Sentiment Analysis of IT Support Dialogues: A BERT-based Approach
(2025-10-09) Klasan, Aljosa; Hagberg, Jonatan; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This thesis investigates whether modern natural language processing (NLP) techniques can extract reliable sentiment insights, within the textual content of ITsupport dialogues. Declining survey response rates and limited resources for manually assessing user satisfaction motivate the use of transformed-based models as a scalable alternative. This study explores utterance-level sentiment classification that explicitly incorporates dialogue context. This is done by evaluating the performance of a pre-trained BERT model, prior to and post fine-tuning on multi-turn, domain-specific e-mail conversations. Furthermore, the study explores automatic anonymisation of sensitive entities within the data, by integrating named entity recognition to an anonymisation pipeline. The results show that fine-tuning significantly improved contextual understanding and sentiment classification performance within the IT-support domain, while still showing encouraging results on single-turn context levels. However, its performance dropped on out-of-domain data, showing moderate generalisability levels. Additionally the evaluation on out-of-domain data, suggests label noise in the training data of the base-model. The anonymisation pipeline successfully masked most personal-, locational-, and organisational names, but showed a tagging error rate of over 40%. This exposes potential legal risk if applied without human oversight. The study demonstrates the potential of dialogue-aware fine-tuning for sentiment analysis while underscoring the importance of domain-specific, high-quality, human-annotated data and robust anonymisation to ensure safe, transferable applications.
Can This Sensor be Removed? Investigating ML Models for Virtual Sensors in Injection Molding
(2025-10-09) Strömer, Måns; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Injection molding relies on physical sensors to monitor critical variables such as pressure, temperature, and actuator positions. However, these sensors can be costly to install and maintain, especially in industrial environments with restricted access or calibration needs. This thesis investigates whether selected physical sensors can be replaced by data-driven virtual sensors and evaluates how different machine learning models compare in achieving this. Using real-world multivariate time series data from Tetra Pak’s injection molding process, six machine learning algorithms of varying complexity were benchmarked: Linear Regression, XGBoost, Feedforward Neural Networks (FFNN), Gated Recurrent Units (GRU), Long Short-Term Memory networks (LSTM), and Transformers. Each model was trained to reconstruct the full signal of a target sensor using a fixed-length time window from the remaining sensors. Performance was measured using range-normalized root mean squared error (RA-RMSE) to enable cross-feature comparison. Results show that GRU and Transformer architectures consistently achieved the lowest RA-RMSE values, particularly for pressure and actuator-related sensors. Temperature signals were harder to model accurately, likely due to long-term dependencies beyond the available input window. Additionally, feature importance analysis revealed trade-offs in sensor removal some sensors are highly predictable but also crucial for estimating others. The findings highlight both the feasibility of using virtual sensors and the importance of model selection. Comparative analysis shows that model architecture significantly affects prediction accuracy, guiding future decisions in sensor design and digital transformation in manufacturing.
Enhancing Patient Understanding of Medical Findings Through NLP and 3D Models
(2025-10-09) Albuszies, Lukas; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This thesis investigates patient communication in clinical settings, recognizing that effective communication is vital for ensuring patient compliance and successful treatment outcomes. Complex clinical jargon and inadequately explained pathological processes often lead to patient confusion and anxiety. Advances have been made in medical text simplification, yet personalized visualizations remain underexplored. Specifically, anatomical models can provide patients with foundational knowledge about the human body, enabling clearer explanations of their medical conditions. The challenge lies in the implicit nature of anatomical references in clinical texts, where discussion focuses mainly on pathology. Anatomical entities are connected to pathological processes, but often not mentioned explicitly, for example a report on a heart attack may mention "myocardial infarction", without mentioning the associated "heart muscle". This thesis compares existing NER methods in their capability to extract implicit entities, and introduces a novel pipeline that leverages foundational biomedical entity relationships from the Unified Medical Language System (UMLS), a compendium of controlled vocabularies that provides structured mappings of relationships between biomedical entities. Extracted entities are visualized through a user interface for a 3D anatomical model, guided by a custom multi-parameter algorithm that optimizes context and clarity. Parameters include contextual distance to surrounding structures, as well as techniques for dimming, highlighting and adjusting opacity. These visualization parameters proved effective in enhancing visual representations by emphasizing relevant structures and minimizing visual clutter. An analysis on the inclusion radius of surrounding structures revealed diminishing returns for all organs tested. A custom camera positioning algorithm was used to automatically center and orient the viewpoint based on the anatomical target’s bounds; this approach effectively improved the clarity and framing of visualizations. To assess annotation quality, large language models were employed as automated evaluators, scoring outputs on a five-point scale. A dedicated validation experiment demonstrated that these models could reliably distinguish between expert-curated and nonsensical annotations, supporting their use as scalable, reproducible evaluation tools. Results show that the proposed method outperforms baselines in annotation quality, with statistically significant and practically meaningful improvements. While opportunities for refinement remain, this research lays the foundation for broader applications in scenarios requiring the extraction and visualization of implicit biomedical entities.
Assessing Accessibility in Tabletop Games for Inclusive Museum Experiences
(2025-10-09) Stylianos Vazouras, Christos; Zhu, Xiangyu; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This research explores tabletop game accessibility in a museum setting, specifically for the "A World of Games" exhibition at Världskulturmuseet. The study began with an evaluation of the games on display, followed by a user study to identify accessibility challenges for visitors. Based on the results, we developed solutions aimed at making the games more inclusive, particularly for blind individuals and young children. One key solution was an image recognition app for the game Mekuri, designed to help players with colour blindness and cognitive impairments. A second round of testing was carried out to assess the effectiveness of these solutions. The research led to the creation of guidelines for making board games in museum exhibitions more accessible. These guidelines are meant to benefit not only future exhibitions at Världskulturmuseet but also other teams working on similar projects. While teams like Meeple Like Us have made important contributions to board game accessibility, this research focuses specifically on the unique challenges of game exhibitions, where factors such as game length, waiting activities, lighting, and space layout need to be considered. The solutions we offer aim to make the museum experience more inclusive for everyone, regardless of ability. Finally, the study outlines future directions for expanding and refining these accessibility solutions.
Towards a System-Level Functional Language: Lithium
(2025-10-08) Selander, Sebastian; Hammersberg, Samuel; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Functional programming has a rich and well-documented history. In functional programming, large problems can be described by the composition of smaller building blocks. Despite its benefits, functional programming has struggled to find its way into system-level programming. By leveraging the guarantees linear types impose, functional programming languages can be applied to system-level programming without sacrificing performance. This thesis presents Lithium, a system-level functional programming language that is based on a variant of linear logic. Lithium is designed to be an intermediate compilation target for linear functional languages. We give the typing and kinding rules for Lithium before describing a series of transformations to turn Lithium into a language that is easily translated into assembly code. Additionally, we present a compilation scheme, a mapping from types to memory, and the application binary interface (ABI).
Mapping comorbidity patterns in high needs paediatric patients: A machine learning approach
(2025-10-08) Ubogu, Chukwudumebi; Xu, Yunyi; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Background: Of children requiring hospital visits, High-cost, high-need (HCHN) children make up a small proportion of this paediatric population but account for a disproportionately large share of the total healthcare costs for the said population. Understanding comorbidity patterns within this subpopulation is crucial for improving care coordination. Unsupervised learning increasingly identifies patterns in complex healthcare data, offering new possibilities for patient stratification. Objective: This thesis used unsupervised machine learning techniques on administrative data from Västra Götalandsregionen (VGR) to identify distinct patient subgroups among high-cost paediatric patients in order to characterise their comorbidity patterns. Methods: The analysis involved hospital visit data for children aged 0-17 years in VGR, defining HCHN patients as the costliest 5%, accounting for 50% of total costs. Three clustering methods were used: KMedoids, Hierarchical Density-Based Spatial Clustering, and Agglomerative clustering to extract distinct patient groups based on diagnosis patterns. Results: The analysis revealed several distinct clusters amongst which were: (1) complex neonates likely with prolonged healthcare dependency remaining costly throughout childhood; (2) neonates initially high-cost due to early-term birth conditions but requiring minimal subsequent intervention; (3) teenage girls with mental health conditions that could lead to increased high self-harm rates; and (4) a complex ailment subgroup that require care from a multidisciplinary care team for optimum care. Conclusions: This study demonstrates unsupervised machine learning’s utility in identifying clinically meaningful subgroups within HCHN paediatric populations. Findings support the need to dismantle siloed treatment strategies through multidisciplinary care teams tailored to specific clusters. Lastly, identifying distinct clusters provides a foundation for targeted interventions and resource allocation strategies.
Interpretable Methods for Information Removal in Text-Based Learning: Exploring the Use of Large Language Models and SHAP as Information Removal Tools in Text Based Learning
(2025-10-08) Michel, Johan; Lindberg, Sara; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
In text analysis, it is sometimes necessary to remove or obscure certain information that should not be included in downstream processing or interpretation. This thesis addresses the challenge of removing such information from text while preserving as much of the remaining content as possible. Existing successful methods typically operate in the embedding space, which makes it difficult to see which specific parts of the text are being changed. In contrast, this thesis proposes a more interpretable approach that operates directly on the text, taking raw text as input and producing rewritten text as output. While one recent method also attempts direct text-based removal using LLMs, this work extends it by exploring a wider range of prompt strategies and, more importantly, by introducing an intermediate step using SHAP. SHAP is used to extract token level importance scores from a classifier predicting the forbidden variable, providing the language model with more targeted guidance on which parts of the text are most relevant to remove. The proposed method was evaluated on two different datasets: one consisting of professional biographies and another of Amazon product reviews. The results indicate that the method successfully removes the forbidden variable in the first dataset while preserving the remaining content. However, it does not succeed in removing the forbidden variable in the second dataset. Across both datasets, the most effective setups included the SHAP-based guidance step, suggesting that SHAP improves the performance of the information removal method. These findings highlight that LLM-based text disentanglement is not a one-size-fits-all solution, but instead requires adaptable strategies depending on the context and the nature of the sensitive information.
AR as a Tool for Exploring Usage and Potential for the Industrial Division
(2025-10-08) Madhavan, Achyut; Yadav, Tanmay; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This thesis examines the use of Augmented Reality (AR) as a tool to increase support in industrial settings. With increasing interest in industry 4.0 technologies, AR has emerged as a promising solution to assist operators in environment such as agriculture, warehousing and forestry. The purpose of this study was to identify a meaningful use case, develop a functional AR prototype, and to evaluate its viability, purpose and effectiveness in the industrial context of the real world. The prototype integrates a live camera feed in its UI with IP address configuration through the XR-compatible virtual keyboard. It was applied using Unity, PTC Vuforia Engine package and tested on a Magic leap 2. Feedback was collected from 12 play-testers with various professional backgrounds including a forklift operator, UI/UX designers and marketing professionals. While some users initially required orientation, most found the system comfortable and suggested many enrichments, such as voice input, better camera hardware and hands-free interaction. The results demonstrate that AR, when designed keeping in mind the user’s needs and practical obstacles, can improve awareness, reduce cognitive weight, and increase work efficiency in industrial workflows. This study supports the idea that AR is not only a viable tool, but a valuable tool for modern industrial applications.
Screen-Space Global Illumination Using Radiance Cascades in 3D Video Games
(2025-10-08) Abou Dan, Ghaith; Jensen, Mati; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Achieving accurate Global Illumination (GI) is essential for creating visually compelling and photorealistic imagery in computer graphics. Recent advancements in GI typically require temporal accumulation or reuse of samples in order to achieve real-time performance at a good quality. However, interactive video games may require GI solutions that respond quickly to user interaction and rapidly changing lighting conditions. We develop a lighting method for video games, specifically designed for the constraints present in the game As We Descend by Box Dragon, where the camera has limited movement and all relevant parts of the scene are on screen at all times, contains many short lasting emissive visual effects, with a budget of a few milliseconds per frame on modern hardware. To create a method for these conditions, we make use of ideas first presented by Alexander Sannikov for use in Path of Exile 2, Radiance Cascades (RC), which is an efficient data structure for representing a lightfield [Ale23]. We make use of a recent screen-space lighting technique, Visibility Bitmask Global Illumination (VBGI) [TLG23] combined with RC, by placing probes in screen-space. We detail many of the optimizations and adjustments needed to combine these techniques, and allowing them to run in real-time. During evaluation, we go over various parameters that allow scaling the method to various performance levels, allowing it to run at down to 2ms at the lowest settings, 6 milliseconds at medium settings, and 26 milliseconds on high settings, with a memory usage ranging from 0.03GiB to 3.9GiB depending on the settings. At this performance, it can achieve one bounce screen-space global illumination along with emissive lighting, with a radius that covers the entire screen. The method has no high-frequency noise without the use of a separate denoising pass, and does not use temporal accumulation or reuse which would otherwise lead to temporal artifacts. The method is not without issue, as there are many limitations and artifacts, and is an approximation of ground truth GI. Firstly, anything that is off-screen cannot contribute or occlude lighting, the chosen screen-space tracing method causes over-brightening of the background, and upscaling radiance data causes artifacts at especially lower settings. Worst is that the method is prone to flickering during even small movements, especially at lower settings, significantly worsening any artifacts present as artifacts become flickery.
On Obfuscating JavaScript Code Using Large Language Models
(2025-10-07) Cîrstoiu, Andreea-Ioana; Heng, Siyu; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Large Language Models (LLMs) have become increasingly popular for their proven capabilities in code analysis and synthesis, and code obfuscation is a suitable task. Generally, the purpose of obfuscation is to make a program difficult to understand. More specifically, it is widely applied to JavaScript code, as it is a popular language used to build client-side web applications. One reason to obfuscate JavaScript code is to prevent anyone from copying proprietary work. Code obfuscation is widely applied and studied in the context of cybersecurity, particularly in relation to malware or intellectual property protection. Existing obfuscators have been built to implement obfuscation patterns of ranging complexities. Research in the area of code obfuscation using LLMs has emerged in recent years and is continually evolving. A potential gap in research is whether LLMs can obfuscate code using patterns that the current standard deobfuscators cannot reverse-engineer, and whether it is possible for the LLMs to replace existing obfuscators. To achieve this goal, it is essential to investigate whether and how LLMs can apply obfuscation transformations to code, as well as the impact that prompt engineering techniques have on the results. In this laboratory experiment, our goal is to determine the extent to which an LLM can obfuscate JavaScript code. We choose an open-weight LLM and we craft prompts to obfuscate standalone, relatively simple code snippets. A key component of our work is a dedicated, free-to-use obfuscation tool that serves as our baseline for evaluating the LLM results. We prompt the model iteratively, then, using data visualization and descriptive statistics, we analyze and interpret the results. Our results show that the chosen LLM can obfuscate simple JavaScript code; however, the choice of prompt engineering technique is crucial. Some LLM-obfuscated code snippets differ significantly from the original code, while maintaining the original behavior. However, the LLM also yields obfuscated code that changes the original behavior, has errors, or is significantly similar to the original code.
What are some of the most significant ways that environmental and level design shape player experience and emotions?
(2025-10-07) Likus, Aleksandra; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
The relationship between players and game environments is a vital aspect of game design, shaping how players experience and connect with the virtual world. This thesis explores how environmental and level design influences player emotions, narrative engagement, and the overall gaming experience. The research focuses on how spatial layout, the interaction between design elements, and the flow of levels evoke specific emotions such as happiness, fear, and sadness within game environments. Through examining how design elements and interaction between components evoke specific emotions, the study uncovers actionable insights for creating immersive and emotionally resonant games. The findings offer game designers practical methods to craft experiences that captivate players emotionally and deepen their connection to the narrative.The research was based on the question: What are some of the most significant ways that environmental and level design shape player experience and emotions?
An exploratory field study on the use of data management and data quality requirements in ML-enabled software applied in environmental research
(2025-10-07) Mahagamarachchi, Devasinghage Sara Nirmani; Pamali Chathurika, Hikkaduwa Liyanage; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Integrating machine learning into environmental science has shown great promise in improving research outcomes. However, the effective application of machine learning and the reliability of the results depend heavily on data quality and management practices, which are often overlooked or addressed inconsistently. It is important to have a proper data pipeline that includes good practices for quality data and data management. This thesis introduces SPADES-ML (Scientific Pipeline Assessment and Data-Centric Evaluation Scorecard for Machine Learning), a structured assessment framework developed to evaluate the quality and transparency of data-related practices in machine learning based research. SPADES-ML is demonstrated through a case study of machine learning based environmental research. A total of 28 research papers were analysed using SPADES-ML. The framework was applied to assess five critical areas: data selection and suitability, data quality, adherence to the FAIR principles, data preprocessing, and challenges in preprocessing. A survey was conducted to validate the findings targeting practitioners in machine learning based environmental research. Results from the literature and survey analyses revealed recurring challenges in ensuring data quality, reproducibility, and methodological excellence. The analysis of SPADES-ML and the survey revealed recurring challenges in ensuring data quality, reproducibility, and methodological excellence. Furthermore, this study provides initial recommendations to improve data practices in machine learning-based research by adhering software engineering principles in the results. This thesis contributes to the emerging field of research software engineering by offering a structured evaluation and guidelines for robust methodology pipelines in interdisciplinary, machine learning based research.
From Challenge to Change: A Human- Centric Framework of Actionable Guidelines for AI Transformations
(2025-10-07) Tavantzis, Theocharis; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
The rapid growth of Artificial Intelligence (AI) is reshaping Software Engineering (SE), offering new opportunities but also introducing human-centered challenges. While existing research highlights behavioral barriers in AI integration, most studies focus on technical roll-out, neglecting how teams adapt and trust AI. This study aims to address this gap by proposing a Behavioral Software Engineering (BSE)-informed framework to support early-stage AI incorporation in SE organizations. To achieve this, a mixed-methods approach was applied. Initially, we conducted a literature review to explore the degree at which former change management models address the behavioral challenges of the AI integration, forming the core of the proposed framework. Then, we performed thematic analysis to enrich this core, using an existing qualitative dataset. Finally, we employed a systematic way to supplement actionable steps per guideline and form the final solution. The proposed framework comprises nine dimensions, AI Strategy Design, AI Strategy Evaluation, Collaboration, Communication, Governance and Ethics, Leadership, Organizational Culture, Organizational Dynamics, and Up-skilling, each supported by guidelines and actionable steps to ensure their applicability. Quantitative evaluation involved a survey and two expert review workshops. The findings of the survey data analysis suggest that AI Strategy Design, Leadership, and Up-skilling are the most critical dimensions for seamless AI integration-results, aligning, to a significant extent, with prior change management initiatives. Specifically, survey respondents allocated on average 15.4% of their budget to AI Strategy Design, 14.9% to Leadership, and 12.9% to Up-skilling, highlighting their priority in early AI adoption. Age- and domain-related patterns also emerged. Despite the small sample size of the expert review sessions, the workshop data analysis indicates the need to focus more on real-life industrial insights. These insights provide a structured, humancentered foundation for AI adoption and offer guidelines for future research studies to contribute to the research body of human-AI integration.
A Chalmers University of Technology Master’s thesis template for LATEX: Investigation of Curriculum Learning in Deep Generative Modelling Using Western Classical Music
(2025-10-07) Chen, Qi; Joshy, Gritta; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
Current deep learning approaches for symbolic music generation typically train on randomly ordered musical sequences, which can hinder the development of coherent musical structure and effective learning of long-term dependencies. While curriculum learning has demonstrated significant benefits in natural language processing and computer vision by progressively introducing task complexity, its application to symbolic music generation remains largely unexplored. The hierarchical nature of Western classical music, with structures ranging from simple motifs to complex compositions, makes it an ideal candidate for progressive learning strategies. This thesis investigates the effectiveness of curriculum learning strategies for symbolic music generation using the Music Transformer architecture. A loss-based complexity ranking system is implemented to order musical sequences from simple to complex, combined with progressive data exposure schedules. Our experimental framework compares baseline training against three curriculum learning variants (CL-60%, CL-80%, and CL-60% with learning rate adaptation) using MAESTRO dataset of Western classical piano compositions. While curriculum models demonstrated faster early convergence, they produced broader loss distributions with higher variance compared to the baseline’s concentrated performance. The CL-60% LR variant emerged as a good performer, achieving superior results in multiple musical features including polyphony, qualified note ratio, and rhythmic structure. Importantly, curriculum learning preserved generalization capability while offering enhanced musical expressiveness. Although curriculum models did not significantly outperform the baseline in loss metrics, they generated outputs that were more harmonically rich, structurally coherent, and aligned with real music characteristics. These findings demonstrate that curriculum learning offers valuable trade-offs in symbolic music generation, producing more musically compelling outputs when properly designed. This work establishes curriculum learning as a promising training paradigm for music generation and highlights the importance of moving beyond loss-based evaluation toward music-informed assessment metrics. All code and experiments are available at: Curriculum_learning 1
Cross-Platform Combat: Analyzing Player Experience Differences Between Mobile and PC
(2025-10-07) Zeng, Shouhui; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This study focuses on the experience difference in cross-platform combat and explores the differences in player operation experience, feedback perception and satisfaction on PC and mobile terminals under the same combat system structure. By building a horizontal 2D action game demo, the game mechanism is kept consistent on different platforms, and only the input method and UI interaction are adjusted to eliminate the influence of other interfering variables. The study adopts a mixed method, combining preliminary questionnaire surveys, custom development of demos for the project, three rounds of field trial tests and subsequent data analysis, to systematically evaluate the player experience from five dimensions: accessibility, operability, satisfaction, immersion and inclusiveness. The results show that even if the core mechanism is the same, different platforms still have a significant impact on the player’s operation fluency and emotional feedback in terms of input accuracy, interaction rhythm and UI layout, especially in terms of skill release, strike perception, rhythm control and other aspects. Different preferences are shown. This study not only provides a theoretical framework and empirical basis for the design of cross-platform game combat systems, but also puts forward a series of optimization suggestions to help developers more effectively improve the consistency of operation experience and player satisfaction on multiple platforms.
Design of Engaging Community-Centered Cooperative Multiplayer Sandbox Games with Wide Demographic Appeal
(2025-10-07) Erdelský, Andrej; Monteiro, Alexandre; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This thesis explores the design of cooperative multiplayer sandbox games that foster community-centered engagement and appeal to a broad demographic of players. In collaboration with FunRock & Prey Studios, the research focuses on identifying key mechanics that encourage the formation of organic role hierarchies within player groups, without explicit direction from the game. By combining theoretical analysis with an iterative design and playtesting process, a functional prototype was developed to observe and evaluate player behavior in a large-scale cooperative environment. The study draws from game design theory, psychology of player motivation, community dynamics, and empirical playtest feedback to develop design guidelines that balance complexity, accessibility, and player agency. Findings reveal that ambiguous mechanics, interdependent gameplay systems, synchronizable player tasks, diverse mechanical complexity, and scalable role structures are critical to promoting emergent leadership, group cooperation, and engagement across casual and hardcore players alike. The resulting insights contribute to both academic research and industry practice in the design of socially driven multiplayer games.
Space Entropy: Emergent narrative through systemic game design
(2025-10-07) Bakalakos, Spyros; Fragkos, Georgios; Göteborgs universitet/Institutionen för data- och informationsteknik; University of Gothenburg/Department of Computer Science and Engineering
This thesis explored emergent narrative techniques in game design and attempted to explain what is needed to create the potential for emergence in a game. Emergent narrative refers to stories that arise organically from player interaction with the game’s systems rather than being pre-scripted. Literature review and research showed many different ways in which unique narrative can emerge from systems in all kinds of games and research prototypes. Based on collected information from previous work on the topic, we developed our own prototype game, named Space Entropy, to test narrative emergence through systemic design with users in a first-person, sandbox game in a Sci-Fi setting. In this prototype, we utilized some of the techniques that we found most appropriate for our case. From the quantitative and qualitative data gathered from the playtests, we were able to establish the validity of the prototype as well as identify its weaknesses. Finally, our findings led us to formulate a set of guidelines aimed at other developers or teams that want to achieve unexpected, emergent narrative without explicitly designing for it.

Browse

Recent Submissions