67471 research outputs found
Sort by
An Empirical Framework for Evaluating Semantic Preservation Using Hugging Face
Full text linkAs machine learning (ML) becomes an integral part of high-autonomy systems, it is critical to ensure the trustworthiness of learning-enabled software systems (LESS). Yet, the nondeterministic and run-time-defined semantics of ML complicate traditional software refactoring. We define semantic preservation in LESS as the property that optimizations of intelligent components do not alter the system’s overall functional behavior. This paper introduces an empirical framework to evaluate semantic preservation in LESS by mining model evolution data from HuggingFace. We extract commit histories, Model Cards, and performance metrics from a large number of models. To establish baselines, we conducted case studies in three domains, tracing performance changes across versions. Our analysis demonstrates how semantic drift can be detected via evaluation metrics across commits and reveals common refactoring patterns based on commit message analysis. Although API constraints limited the possibility of estimating a full-scale threshold, our pipeline offers a foundation for defining community-accepted boundaries for semantic preservation. Our contributions include: (1) a large-scale dataset of ML model evolution, curated from 1.7 million Hugging Face entries via a reproducible pipeline using the native HF hub API, (2) a practical pipeline for the evaluation of semantic preservation for a subset of 536 models and 4000+ metrics and (3) empirical case studies illustrating semantic drift in practice. Together, these contributions advance the foundations for more maintainable and trustworthy ML systems
Agentic AI Platform Networks
Full text linkAgentic AI is a recent development in artificial intelligence (AI) technologies that is capable of automating task execution and decision-making. In the context of platforms, they provide novel approaches in interconnecting platforms. This enhances the traditional hub-and-spoke architecture of multi-sided platforms and leads to the formation of networks of platforms. This research analyzes four major AI platforms and identifies three distinct integration patterns: isolated architecture (Alexa), mediated integration (ChatGPT), and protocol-based networking (Claude/Gemini). The findings reveal that integration-centric architectures demonstrate higher cross-platform connectivity than traditional, interconnection approaches . They extend platform theory by introducing network orchestration as a new competitive strategy and provide empirical evidence of an emerging paradigm shift from platform competition to interconnection
Rethinking Retail Location Decisions: Industry insights into Decision-making Practice
Full text linkRetail location decision-making is facing growing challenges as consumer behavior becomes more complex and dynamic. This paper draws on ten in-depth interviews with location decision-makers at major Canadian retail and service firms. While traditional decision-making practices continue to dominate, there is an increasing interest in leveraging spatial big data and applying data science and geospatial artificial intelligence (GeoAI). Yet, many organizations remain cautious, often relying on institutional knowledge or rebranding existing tools rather than wholly embracing innovation. Experimentation with data science and GeoAI is taking place. However, its effective integration will require strong leaders and better collaboration between data science teams and decision-makers to align analytical models with experiential judgment. Nevertheless, the shift from legacy decision-making toward more adaptive data science and GeoAI-informed strategies is underway. This transition marks a strategic inflection point, with the success of new approaches depending on how well firms overcome inertia and foster innovative decision cultures
Simplicity Gone Wrong: Revisiting Implicit Designs for Older Adults
Full text linkThis study investigated how removing implicit design elements, such as gesture-based or symbol-based interactions, affects the usability and learnability of mobile applications for older adults. We used Line, a popular social media platform in Taiwan, for a case study and created a prototype version with implicit designs removed for comparison. We collected data from 16 older adult participants and primarily focused on their verbal responses and task completion performance. Our findings suggest that replacing symbol-based designs with text can improve user experience, while replacing gesture-based designs did not result in better usability. We also found that how the user explores the application greatly impacts whether implicit design affects their experience. This study provides insights into how users interact with different kinds of implicitness and provides a case study of a less frequently reviewed application
Federated Learning for Brain Tumor Classification from MRI: A Comparison of MLP and ConvNeXt Approaches under IID and Non-IID Data Scenarios
Full text linkFederated Learning (FL) effectively addresses privacy concerns in medical imaging by enabling collaborative model training without sharing sensitive patient data. This paper compares two neural network approaches for brain tumor classification (BTC) from magnetic resonance imaging (MRI) in a federated learning setting. Both models operate on regions of interest (ROIs) extracted from brain MRI scans. The first is a lightweight multilayer perceptron (MLP) that classifies ROIs based on radiomic features extracted from them. The second is a deep learning (DL) approach based on the ConvNeXt architecture, which performs classification directly on the ROI images. Two experimental scenarios are considered: a balanced (IID) and an unbalanced (non-IID) distribution of data among federated clients. Results show that the radiomics-based MLP achieves performance comparable to the more complex ConvNeXt model, while requiring significantly lower computational resources. Moreover, federated learning consistently outperforms isolated local training, particularly under non-IID conditions, emphasizing its potential for clinical deployment
Modeling Digital Repression: A Machine Learning Analysis of Shutdowns as Governance Signals
Full text linkThis study advances Digital Government research by applying machine learning to analyze internet shutdowns as structured signals of digital repression. Using a dataset of 566 shutdown events (1995–2011) and regime attributes from the Polity 5 project, the study builds interpretable models to estimate shutdown severity and classify regime type. A decision tree regressor and bootstrapped logistic classifier reveal strong associations between shutdown characteristics and political context, achieving over 93% accuracy. While not designed for real-time prediction, these models demonstrate how event-level data can inform early warning, policy evaluation, and digital rights monitoring. By modeling shutdowns as governance decisions embedded in digital infrastructure, this research shows how computational methods can support accountability in opaque information environments
What Makes or Breaks an Immersive Online Customer Journey? Digital Native Generation Z Customers’ Experiences
Full text linkCustomers may lose awareness of time and their surroundings during online shopping, signaling an immersive experience. Despite the widespread use of everyday digital technologies, it remains unclear at which stage of the customer journey immersion is born or broken—and why. This study addresses that gap by exploring how and why individuals become immersed in online shopping. Data were collected through 33 semi-structured interviews with Finnish digital natives belonging Generation Z who had experienced online shopping immersion. The findings identify seven fostering factors for online shopping immersion: motivation towards products, intentional product search, external impulses, captivating product loop, intense concentration on products, high-quality product visualization and need to collect all the information. On the other hand, four interrupting factors are presented: external distractions, technological issues, fatigue and transaction
Faith in the Feed: A Netnography of Delusional Consumption Ideology
Full text linkThis paper adopts a consumer culture lens to investigate the Delulu phenomenon across social media platforms. Addressing the underexplored entanglement between delusional imaginaries and digital platforms, we conduct a netnography of Delulu content across TikTok, YouTube, Instagram, and Reddit. Drawing on Kozinets’ theorization of digital utopianism we analyze visual, discursive, and ritualized manifestations of delusional consumption. We unveil Delulu as a utopian apparatus of self-actualization, marketplace formation, and ritualized consumption fostering a delusional consumption ideology. This ideology frames fantasy and desires as actionable reality through aesthetic routines, witchcraft practices, marketized symbolism, and algorithmically-reinforced hope, blurring boundaries between aspiration and illusion. Conceptually, we advance the debate on consumption ideology for the digital age, by unpacking the notions of delusional consumption ideology and technosocial realities. Methodologically, we contribute by discussing how netnography foregrounds ideological entanglements that result from researchers’ immersive, reflexive engagements, thus highlighting netnography as a method for studying consumption ideologies