Publikationer från Stockholms universitet
Not a member yet
31237 research outputs found
Sort by
AI-drivet beslutsstöd för MDR-klassificering: En chattbot för riskklassificering av medicintekniska produkter
No full textIntroduktion MDR:s förordning för medicintekniska produkter inom EU syftar till att säkerställa en god patientsäkerhet genom transparens och strikta krav. Förordningen har skapat problem för tillverkare, då komplexiteten i regelverket kräver ökade resurser. Konsekvenserna inkluderar fördröjd marknadsintroduktion och dyra kostnader. Denna uppsats undersöker hur AI kan användas som beslutsstöd utifrån MDR:s regelverk. Författarna har utvecklat en prototyp av en AI-baserad chattbot som ska kunna utföra en riskklassificering utifrån MDR. Forskningsmål Studien har som mål att utveckla och utvärdera hur korrekt en AI-baserad chattbot utför riskklassificering av en medicinteknisk produkt enligt MDR. Metod En designvetenskaplig studie genomfördes där en artefakt, i form av en prototyp av en AI-baserad chattbot, utvecklades med en iterativ ansats. Först genomfördes en litteraturgenomgång för att förstå problemområdet. Från litteraturgenomgången formulerades krav för artefakten. Chattboten utvecklades med Python och LangGraph ramverket. Utvärdering skedde genom en ex-post utvärdering med ett stratifierat randomiserat urval av produkter från EUDAMED:s databas. Testobjekten jämfördes med chattbotens genererade riskklass för att mäta procentuell korrekthet. Resultat Resultatet är en prototyp av en utvecklad chattbot som utvecklats utifrån framtagna krav. Resultatet visar att artefakten kan bedöma riskklasser med 70% korrekthet. Artefaktens korrekthet varierar mellan de olika riskklassgrupperna. Högst korrekthet uppkom i den lägsta och högsta riskklassen. Diskussion Studien indikerar potential med användning av AI-baserade beslutsstöd i regelstyrda miljöer, med komplement av mänsklig expertis. I reglerade miljöer är spårbarhet och transparens viktigt, därför ses potential för fortsatta studier inom hybrid arkitektur som möjliggör mänsklig granskning under klassificeringen istället för att endast använda AI.Introduction The new MDR regulation for medical devices in the EU aims to ensure patient safety through transparency and strict requirements. This has created challenges for manufacturers of medical devices, as the complexity of the regulatory framework requires increased resources. The consequences include delayed market introduction and high costs. This thesis examines how AI can be used as a decision-support tool within the MDR framework. The authors have developed a prototype of an AI-based chatbot designed to perform risk classification according to the MDR. Research Goal The study aims to develop and evaluate the accuracy of an AI-based chatbot in performing risk classification of a medical device according to the MDR. Method A Design Science Research study was conducted in which an artefact, a prototype of an AI-based chatbot, was developed using an iterative approach. A literature review was conducted to understand the problem area and to identify requirements. The chatbot was developed using Python and the LangGraph framework. Evaluation was carried out on a stratified randomised sample of products from the EUDAMED database. The test objects were compared with the risk classes generated by the chatbot to measure the percentage of correctness. Results The study resulted in an artefact, a developed chatbot made from formula- ted requirements. The study results show that the chatbot can assess risk classes with 70% accuracy. The artefact’s classification accuracy varies across different risk class groups. The highest accuracy was observed in the lowest and highest risk classes. Discussion The study indicates the potential of AI-based decision support systems in regulated environments, when complemented by human expertise. In regulated environments, traceability and transparency are crucial. We see potential for continued research into hybrid architectures that allow for a ‘human in the loop’ during the clas- sification phase, rather than relying solely on AI
Sveriges neolitiska växter : En studie om växternas roll i det neolitiska samhället i Sverige 4000–1800 f.v.t.
No full textThis essays aim is to understand how the neolithic people of Sweden used specific plants in their day-to-day life. The plants this essay chose to study were apples, nuts, mushrooms and grains. Due to their rich amount of plant material, the neolithic Alvastra pile dwelling as well as the human bog find, Hallonflickan (“the Raspberry girl”) were chosen as locations to study. The results showed that the plant material from both Alvastra and Hallonflickan were outside of being used for its nourishment, primarily used in conjunction with ritualistic feasts and offerings. Certain mushrooms were also used as fire starting material
Designing Immersive Tools for Exposure Therapy: Expert-Guided Development of VR/MR Applications for Acrophobia
No full textIntroduction: Exposure therapy is an established treatment for specific phobias such as acrophobia, the fear of heights. Advances in Virtual Reality (VR) and Mixed Reality (MR) technologies provide new opportunities to enhance exposure therapy by offering accessible, controllable, and safe environments. However, there is limited research on how such applications should be designed to ensure that they can support patient treatment and are practical for psychologists in clinical settings, including considerations of technical stability, usability, and therapeutic workflow integration. Research Question: The primary research question guiding this thesis is: “What are the design considerations necessary to ensure that a VR/MR application is usable by psychologists and can be integrated into exposure therapy treatment for acrophobia?” Method: The study employed a qualitative Design-Based Research (DBR) approach, combining iterative prototype development with input from an expert experienced in Virtual Reality Exposure Therapy (VRET). Two interviews were conducted, one exploratory and one following hands-on evaluation of the prototype. The interviews were analyzed using thematic analysis. In addition, a heuristic usability inspection and an engineering validation were conducted to assess safety, usability, and technical performance of the prototype. Results: The findings highlight three key areas of design considerations: (1) the need for gradual and patient-controlled exposure that balances realism with safety, (2) usability and workflow alignment to support clinical practice, and (3) the potential of MR in providing grounding and embodied interaction as a potential extension of VR-based exposure. These considerations were translated into concrete design requirements and implemented in a VR/MR prototype. Evaluation showed that key requirements, such as safety controls, patient agency, MR grounding, and performance stability, were largely met, while elements such as therapist-facing controls require further development. Engineering validation showed stable framerates with low reprojection across both devices, suggesting that the prototype meets the technical requirements for early-stage clinical use. Discussion: This study contributes both conceptual and practical insights into designing immersive applications for acrophobia treatment. It highlights the importance of aligning technological design with therapeutic principles, clinician usability, and technical reliability. Limitations include reliance on a single expert and the absence of patient testing. Future research should validate these design considerations across multiple practitioners and evaluate therapeutic outcomes in clinical settings
"Du måste man up nu, lillen" : En tematisk analys av hur maskulinitet framställs i serien Snabba Cash.
No full textDet demokratiska priset : En tidsserieanalys av hets mot folkgrupp i samband med valrörelser i Sverige år 2000–2024
No full textPolitik präglas ofta av hård retorik och ämnen som väcker känslor. Under valår ställs politiken på sin spets där partier arbetar hårt för att säkra varje enskild röst. Det är denna särskilt politiskt intensifierade period som är huvudintresset för denna studie. Med hjälp av en avbruten tidsserieanalys (Interrupted time series) ämnar studien undersöka sambandet mellan riksdags- och EU-valår och hatbrott i en svensk kontext. Studien undersöker perioden 2000 till och med 2024 med hänsyn till data tillgänglig i Brottsförebyggande rådets databas över anmäld brottslighet. För att göra hatbrott mätbart operationaliserades det till en variabel som bestod av kvartalsvis förändring i antalet polisanmälda fall av hets mot folkgrupp. Resultatet visar en statistiskt signifikant effekt vid riksdagsval, där antalet anmälningar av hets mot folkgrupp ökar med cirka 19 procent. Detta motsvarar i genomsnitt 100 fler anmälda fall under ett valår givet en befolkning på 10 miljoner invånare. Variabeln för EU-val visar däremot ingen signifikans i samtliga modeller vilket tyder på att effekten ter sig skiljaktigt mellan riksdags- och EU-val. Vidare åskådliggörs flertalet ytterligare faktorer som möjligtvis kan ligga bakom effekten, bland annat samhällsomvälvande händelser, användandet av sociala medier och reaktioner på invandring. Slutsatsen är att riksdagsvalrörelser skapar tillfällesstrukturer som riskerar öka utsattheten för minoritetsgrupper, vilket belyser ett mätbart demokratiskt pris av ett intensifierat debattklimat
School Leadership and the Association to Teachers’ Digital Competence in Supporting Students with Special Educational Needs
No full textThe digitalisation of education has introduced new possibilities for inclusive teaching practices, particularly in supporting students with special educational needs (SEN). While digital tools have demonstrated potential to enhance learning outcomes and engagement for these students, the role of school leadership in fostering teachers’ digital competence remains underexplored. The aim of the study is to investigate the association between school leadership, as rated by teachers, and teachers’ self-reported digital competence in supporting students with SEN. To this end, cross-sectional data from 285 Swedish teachers enrolled in special education training programmes have been used. The data were collected through the SELFIE survey, a European Commission tool designed to assess schools’ digital capacity. A stepwise linear regression analysis was conducted to examine the association between perceived school leadership and teachers’ self-reported digital competence in supporting students with SEN, controlling for teacher collaboration, infrastructure and equipment, and demographic variables. The results show a consistent and significant positive relationship between school leadership and teachers’ digital competence, even when other factors are accounted for. Teacher collaboration also contributed positively, though to a lesser extent, while infrastructure and equipment and demographic variables showed no significant effect. The study contributes knowledge by showing that teachers’ digital competence development depends not only on individual efforts but also on organisational factors, such as supportive school leadership, highlighting the importance of recognising school leadership as vital alongside digital resources in schools. Given the cross-sectional design, the findings should be interpreted cautiously and not as evidence of causal relationships. These findings suggest that school leadership is important in enabling teachers to use digital technologies to support students with SEN, highlighting practical and policy implications for strengthening school leadership in developing teachers’ digital competence in supporting students with SEN
Arbetsmässig anpassningsförmåga och dess påverkan på AI transformation i offentlig sektor
No full textIntroduktion: AI införs i allt högre grad i offentlig verksamhet, men nyttan realiseras inte enbart genom tekniken i sig. För att AI ska skapa värde i det dagliga arbetet krävs ändamålsenliga arbetssätt och organisatoriskt stöd. Studien fokuserar därför på arbetsmässig anpassningsförmåga, som avser medarbetares och teams förmåga att anpassa arbetssätt vid förändrade förutsättningar. Forskningsfråga: Vilket samband finns mellan arbetsmässig anpassningsförmåga och upplevd nytta av AI-driven transformation i offentliga verksamheter? Metod: Datainsamlingen genomfördes genom en anonym webbenkät riktad till anställda inom svensk offentlig sektor (N = 308). Enkäten mätte upplevd AI-nytta och arbetsmässig anpassningsförmåga på individ- och teamnivå samt bakgrundsvariabler om roll, yrkeserfarenhet, AI-användning och organisatorisk AI-mognad. Materialet analyserades med deskriptiv statistik, korrelationsanalyser och regressionsmodeller. Resultat: Resultaten visar positiva samband mellan arbetsmässig anpassningsförmåga och upplevd AI-nytta. Arbetsmässig anpassningsförmåga på individnivå framträder som den starkaste prediktorn för upplevd AI-nytta, medan arbetsmässig anpassningsförmåga på teamnivå uppvisar ett svagare samband. Diskussion: Studien bidrar med empirisk kunskap om hur arbetsmässig anpassningsförmåga på individ- och teamnivå relaterar till upplevd AI-nytta i svensk offentlig sektor. Sammantaget indikerar resultaten att upplevd AI-nytta i offentlig sektor hänger samman med organisatoriska förutsättningar och medarbetares anpassningsförmåga. Studiens tvärsnittsdesign och självskattningar begränsar möjligheten att dra kausala slutsatser.Introduction: AI is increasingly adopted in the public sector, but its benefits are not realised through technology alone. For AI to create value in everyday work, appropriate work practices and organisational support are required. This study therefore focuses on workforce agility, defined as employees’ and teams’ ability to adapt work practices under changing conditions. Research question: What is the relationship between workforce agility and perceived benefit of AI-driven transformation in public sector organisations? Methodology: Data were collected through an anonymous web-based survey targeting employees in the Swedish public sector (N = 308). The survey measured perceived AI benefit and workforce agility at the individual and team levels, along with background variables related to role, work experience, AI use, and organisational AI maturity. The data were analysed using descriptive statistics, correlation analyses, and regression models. Results: The results show positive relationships between workforce agility and perceived AI benefit. Individual agility emerges as the strongest predictor in relation to perceived AI benefit, while team agility shows a weaker association. Discussion The study contributes empirical knowledge on how workforce agility at the individual and team levels relates to perceived AI benefit in the Swedish public sector. The results indicate that perceived AI benefit in the public sector is associated with organisational conditions and employees’ adaptive capacity, while the cross-sectional design and reliance on self-reported data limit causal inference
Swedish Upper-Secondary Students’ Attitudes Toward African American Vernacular English
No full textThis study examines how Swedish upper-secondary students evaluate African American Vernacular English (AAVE) in comparison with a Southern White American English (SWAE) accent and how these attitudes relate to motivation and perceived classroom appropriateness. While previous research in Sweden has largely focused on learners’ preferences for “standard” varieties such as General American and Standard Southern British English, little attention has been paid to racialised vernaculars, despite the growing global visibility of AAVE through music, film and social media. Drawing on work on standard language ideologies and language attitudes, the study explores how Swedish learners position two Southern U.S. accents that are historically intertwined yet unequally valued. The data come from a verbal guise experiment with 48 students enrolled in English 6–7 at a Swedish upper-secondary school. Participants listened to two one-minute recordings of spontaneous speech by middle-aged women from Mississippi: one using a SWAE accent and one using AAVE accent. After each recording, students rated the speaker on seven semantic-differential scales targeting status-related and solidarity-related traits, and then answered comparative questions about which accent they found more appropriate, more motivating, and more suitable for inclusion in teaching materials. Results show that the SWAE accent received significantly more positive status evaluations (e.g. educated, clear), while solidarity ratings (e.g. friendly, likeable) were broadly similar across accents. A clear majority of students preferred the SWAE accent for classroom use and representation in teaching materials. The findings indicate that Swedish learners reproduce institutional hierarchies of English varieties by granting greater educational legitimacy to a SWAE variety than to AAVE, even though both speakers in the study are from the U.S. South. The study contributes to debates on linguistic justice in English education by showing how standard language ideology and perceived classroom appropriateness intersect in the Swedish school context
