Publikationer från Stockholms universitet
Not a member yet
31237 research outputs found
Sort by
Commodity Time Series Momentum and the impact of Market Stress
No full textThis thesis investigates the momentum effect in the commodity market. The research approach used in this thesis is deductive and quantitative, utilizing daily data for 23 commodity future contracts with a time frame ranging from 2007 to 2024. Two different portfolio formation techniques are employed with various formation and holding periods, following the method of Moskowitz et al. (2012). Findings show significant momentum effects in short term horizons for both holding and formation periods, with the optimal portfolio with one-month formation and holding period yielding 0,45% monthly return (see table 4.1). The strategy has exceptional performance during recession periods, with the volatility-adjusted long-short portfolio (one-month formation, three-month holding period) achieving monthly returns of 2.98%. The persistence of significant alpha in the five-factor model, particularly in short horizons, indicates that momentum represents both market inefficiency and risk premium components
Comparison of Machine-Learning Algorithms for Intrusion Detection Systems
No full textIntroduction: As cyberthreats continue to grow in complexity and frequency, efficient systems for identifying malicious traffic have become essential in network security. Machine learning enables detection of both known and unseen threats through pattern recognition. Regardless of the large amount of research on how different algorithms perform, there is no consensus about their performance when they are used for intrusion detection. Therefore, there is a need for future research. Research Question: The primary research question is: “How do various machine-learning algorithms, specifically Decision Trees, Random Forest, Support Vector Machine, Naïve Bayes, K-Nearest Neighbours, and AdaBoost, perform when they are used for intrusion detection on the CICIDS2017 dataset?" Method: The research question is addressed through a simulation-based experiment where training and testing is done on a set of machine-learning algorithms using the CICIDS2017 dataset. The dataset is preprocessed through cleaning, feature selection, normalisation, and data splitting. Models are assessed based on accuracy, precision, recall, F1-score, and computational time, using both binary and multiclass classifications in order to examine overall performance and to create a basis for comparative analysis of the algorithm's effectiveness. Results: The results indicate that tree-based models, such as Random Forest and Decision Tree, consistently outperformed other machine learning algorithms, both in binary and multiclass intrusion detection tasks. In binary classification, they achieved F1-scores above 0.998. In the multiclass setting, they showed good performance on attacks with large sample sizes, though this performance dropped when classifying rare attack types. k-Nearest Neighbours performed competitively but was limited by high testing times, while Support Vector Machine suffered scalability issues. AdaBoost and Naïve Bayes underperformed, especially when faced with class imbalances. Discussion: The findings suggest that model selection for intrusion detection tasks should prioritize algorithms that are both robust to class imbalance, as well as efficient in time complexity, such as Random Forest. The limited performance displayed by the models on rare attack types highlights a key challenge in network intrusion detection systems. Future research could explore ensemble models or deep learning approaches to improve detection rate for low sample size attacks. In addition to this, practical development should consider the trade-offs between computational cost and accuracy
Barnperspektiv och berättarperspektiv i Läsebok för folkskolan : En analys av Läsebok för Folkskolan (1868)
No full textJämförelse av dokumentsegmenteringsmetoder för RAG-system
No full textMed den växande integrationen av AI-drivna teknologier inom olika sektorer har stora språkmodeller (LLM:er) visat betydande potential, särskilt inom kodrelaterade tillämpningar såsom LLM-baserade kodassistenter. Retrieval-Augmented Generation (RAG) har ytterligare förbättrat modellernas faktiska tillförlitlighet genom att minska hallucinationsfel. Trots dessa framsteg har begränsad forskning undersökt huruvida dokumentsegmenteringsmetoder som utvecklats för naturligt språk fungerar effektivt när de tillämpas på källkod. Denna studie undersöker prestandan hos dokumentsegmenteringsmetoder—ursprungligen utformade för både naturligt språk och programkod—i Python-kodbassystem, med fokus på mätvärden som precision, recall och Intersection-over-Union. Med hjälp av ett modifierat verktyg från Chroma Technical Reports genomförde vi ett kvantitativt jämförande experiment på öppna källkodsprojekt, där resultaten analyserades statistiskt med ANOVA och icke-parametriska tester. Resultaten visar inga signifikanta prestandaskillnader överlag, men utmanar tidigare antaganden genom att påvisa att en tidigare föredragen metod presterar sämre i kodkontexter. Dessa insikter antyder att antaganden om segmenteringsmetoders effektivitet bör omvärderas vid utformning av RAG-baserade kodassistenter, även om ytterligare forskning krävs på grund av begränsningar såsom storlek och representativitet hos det data-set som används.With the growing integration of AI-driven technologies across various sectors, large language models (LLMs) have shown significant potential, particularly in code-related applications such as LLM-powered code assistants. Retrieval-Augmented Generation (RAG) has further enhanced the factual reliability of these models by reducing hallucination errors. Despite this progress, little research has explored whether chunking methods developed for natural language function effectively when applied to source code. This study investigates the performance of chunking methods—originally designed for both natural language and program code—in Python code bases, focusing on metrics such as precision, recall, and Intersection-over-Union. Using a modified tool from Chroma Technical Reports, we conducted a quantitative comparative experiment on open source projects, statistically analyzing the results using ANOVA and non-parametric tests. The findings indicate no significant performance differences overall, yet challenge prior assumptions by revealing that a previously favored method underperforms in code contexts. These insights suggest that assumptions about segmentation method effectiveness should be re-evaluated when designing RAG-powered code assistants, although further research is needed due to limitations such as dataset size and representativeness
What's the problem represented to be? : A poststructural policy analysis of South Africa's Just Energy Transition Partnership
No full textThis study critically interrogates South Africa’s Just Energy Transition Partnership (JETP). Rooted in poststructuralism, it adopts the ‘What’s the Problem Represented to be?’ (WPR) approach to policy analysis. The research aims to interrogate how the JETP constructs the ‘problem’ it addresses by asking, what its implicit problem representations are, and what underlying presuppositions, assumptions, omittances, and effects they have. It applies a theoretical framework grounded in WPR together with the concepts of financial subordination, the wall street consensus and renewable energy financialisation. By using a qualitative design, the analysis focuses on South Africa’s JETP texts: the Just Transition Framework, the Just Energy Transition Investment Plan, and the Just Energy Transition Implementation Plan. The study identifies two key problem representations: mobilising finance for an ‘uninvestable South Africa’ and ensuring a just transition by diversifying the coal-dependent economy. These are underpinned by a wall street consensus discourse and the political rationalities of Public-Private-Partnerships and financialisation. The study concludes that the JETP risks perpetuating financial subordination and that it frames justice primarily as market participation. The JETP also produces effects, both discursive effects but also how it produces particular types of ‘subjects’, ‘objects’ and ‘places’
Urban Wealth and Climate Impact: Investigating Climate Attitudes in Östermalm, Stockholm
No full textTo truly address climate change, action is required across all sectors of society worldwide. This thesis explores how socio-economic wealth influences climate attitudes and perceptions of responsibility among residents of Östermalm, Stockholm. By using eight semi-structured interviews and thematic analysis, the study examines daily practices, awareness of climate change and its related initiatives, and barriers to sustainable behavior. The analysis is guided by four theoretical frameworks: Urban Political Ecology, Green Urbanism, Environmental Justice, and Green Gentrification. The findings reveal that even though the residents express some levels of climate awareness, and support some climate initiatives, their engagement remains mainly superficial or conditional based on comfort and convenience. Socio-economic wealth enables eco-friendly choices, but also high-emission lifestyles. The study further highlights the need for stronger alignment between economic privilege and environmental responsibility and contributes to the growing body of research on urban sustainability and environmental justice in wealthy, high-emission settings
Growing Up Between Two Homes : A quantitative study of socioeconomic differences in Swedish adolescents’ post-separation living arrangements, 2002-2022.
No full textExpertberoende i SSBI: Användarnas insikter om verktygsutmaningar
No full textSelf-Service Business Intelligence (SSBI)-verktyg är avsedda att göra dataåtkomst och dataanalys mer tillgängligt för icke-tekniska affärsanvändare. Det finns dock många utmaningar som användare av dessa verktyg kan möta. Dessa utmaningar bidrar ibland till ett beroende av expertanvändare, såsom dataanalytiker och BI-specialister. Denna studie undersöker vilka verktygsspecifika utmaningar användare kan uppleva, och vilka aspekter av verktygen som kan orsaka ett beroende av expertanvändare. För att återspegla båda användargruppernas erfarenheter genomfördes semistrukturerade intervjuer med fem deltagare, tre icke-tekniska affärsanvändare och två expertanvändare. Data analyserades sedan med hjälp av tematisk analys och detta påvisade sex teman. Studien identifierar flera verktygsspecifika utmaningar, såsom dold funktionalitet och ett behov av korrekt konfiguration från expertanvändare. Dessutom finns det utmaningar med att förstå visualiseringar, och begränsade alternativ eller ovilja att använda verktygen när det gäller samarbete. Dessutom visar resultaten att användare ofta söker hjälp från andra mer erfarna, icke-tekniska användare, istället för expertanvändare.Self-Service Business Intelligence (SSBI) tools are intended to make data access and data analysis more accessible to non-technical business users. However, there are numerous issues facing users of these tools. These issues sometimes cause a dependence on expert users, such as data analysts and BI specialists. This study explores what tool-specific challenges users might experience, and what aspects of the tools might cause a dependence on expert users. To reflect both user groups experiences, semi-structured interviews with five participants were conducted, with three non-technical business users and two expert users. The data was then analyzed with thematic analysis and uncovered six themes. The study identifies several tool-specific challenges, such as hidden functionality and a need for correct configuration from expert users. In addition, there are issues with understanding visualizations, and limited options or unwillingness to use the tools when it comes to collaborating. Furthermore, the results show that often users look for help from other more experienced, non-technical users, instead of expert users
Teacher Intentions and Student Interpretations: A Meeting of Diverse Frames of Reference
No full textI syfte att utvärdera den egna undervisningen i historia på gymnasieskolan vars huvudmål var attutveckla historiemedvetande genom metoden diskursanalys, lutar sig denna undersökning mot empirifrån vederbörande lektioner i form av observationer samt en svarsenkät kopplat till förädlandet avnämnda förmåga. Frågeställningen - På vilket sätt kan diskursanalys användas som arbetssätt för attfrämja historiemedvetande hos gymnasieelever? styrde arbetet som bearbetats dels genom återgivelserutifrån ‘thick description’ metoden, dels genom kvalitativ analys av enkätsvar och dels ensammanställning av dessa elevuppfattningar. Resultaten låter tyda att diskursanalys som arbetssättmed historieämnet som underlag bäst lämpar sig när instruktioner och uppgifter är konkreta,företrädesvis kopplat till ett specifik material som exempelvis bilder. Språkliga barriärer i form avkomplicerad terminologi nödgar mer tid än arbetet hade utrymme för. Arbetssättet hade för enutförligare utvärdering behövt implementerats över ett helt läsår, vilket är ett huvudargument förundervisningsidéns tillkomst. Slutsatserna från undersökningen föranleder även en diskussion kringelevernas upplevda behov av ett ‘historiemedvetande’, där förslaget för framtida forskning är attgenomföra samma studie i ett område med andra socioekonomiska förutsättningar
The Collapse of The Protestant Ascendancy and its Subsequent Psychological Impact on Bram Stoker's Dracula
No full textThe aim of this thesis is to explore the Anglo-Irish subtext of Bram Stoker’s Dracula (1897) with particular focus on how recurring themes highlight the tensions brought on by the Protestant Ascendancy. I perform a close reading of Stoker’s novel through the lens of Michel Foucault’s theory of New Historicism as connected to power as well as Terry Eagleton’s ideas on how hegemony affected Irish politics in the nineteenth century. In this thesis I ask whether the downfall of the Protestant Ascendancy had any psychological impact on the Anglo-Irish community and if so, how was this depicted in Dracula? I argue that Bram Stoker employs the character of Count Dracula, Catholic symbols, and Gothic imagery to unravel the psychological impact of the fall of the Protestant Ascendancy which presents itself in the form of loneliness and the lack of a sense of belonging. I present my results in three different sections: First, I explore how this shift in power between the Protestant Ascendancy and the tenants manifests in Dracula in the form of the Count himself including his violent tendencies and anxieties toward the future. I analyse how the discussions surrounding religion throughout the novel and the use of religious symbols mirrored deeper religious tensions in Ireland at the time. I also consider these religious tensions in terms of how they gave rise to the Irish Gothic, a genre that became synonymous with Anglo-Irish writers, and how Stoker utilises and builds upon this genre in Dracula. By examining how Stoker employs the figure of the aristocrat, religious symbols, and gothic conventions I show how these aspects of Dracula reveal loneliness and a lack of sense of belonging within the novel’s main antagonist.