NIMS: MagIC - Documentos de conferências nacionais
URI permanente para esta coleção:
Navegar
Entradas recentes
- Is Education Spatial?Publication . Santos, Samuel; NOVA Information Management School (NOVA IMS)High stakes in access to Higher Education decisions: National Exams + Internal grades in Secondary Ed., decided at the decimal points. Systematic differences, even if small, suggest inequity. Geospatial analysis can detect regional differences. Public policies have different bandwidths; not everything has to be done at a national level. Source and data are available, to foster further research and discussion.
- Developing Cycling Networks to Improve Active Accessibility to Public Schools Using the 15-minute City ConceptPublication . Diogo Pinto, José; NOVA Information Management School (NOVA IMS)• 83% of children do not meet 60 min of daily physical exercise • In Lisbon, 44% of school kids are driven to school by car • 15% of urban traffic is generated by school journeys • 70% of EU population lives in cities, 78-80% by 2050 • < 4% of cycling share in most southern European cities What should be done in Southern European cities? • Promote walking and cycling to schools ✓ promotes children’s independence ✓ shapes future travel behavior • Plan proximity-accessible schools: ✓ family-friendly, improves quality of life ✓ sustainable and competitive ✓ improves social resilience and participation What is being done? • Bicycle trains to schools (e.g. Lisbon, Barcelona, Milan) • School surroundings projects (e.g. kiss and ride in Lisbon) • Cycling literacy workshops at schools (e.g. traffic schools) • Mobility to school surveys (e.g. Hands up! in Lisbon) RQ1 – Where is the current cycling infrastructure not sufficient to provide 15-minute active accessibility to schools? RQ2 – What is the contribution of cycling to improving active accessibility to schools? RQ3 – How to operationalize 15-minute accessibility indicators to guide cycling infrastructure investment decisions?
- Uncovering Stroke Mortality Profiles Through Clinical, Socioeconomic, and Environmental Risk FactorsPublication . Ramalhete, Sara Ventura; Information Management Research Center (MagIC) - NOVA Information Management School; NOVA Information Management School (NOVA IMS)Stroke remains one of the leading causes of mortality worldwide1. While traditional risk factors for stroke are well recognized, they do not completely account for variations in stroke-related deaths. Increasing evidence suggests that non-traditional factors, including environmental conditions and social determinants of health (SDOH), may also influence mortality outcomes. However, previous studies have produced inconsistent results, and only a limited number have examined the combined impact of both classical and non-classical risk factors on survival. This study aimed to identify profiles of stroke patients with differing probabilities of survival by considering a broad range of traditional and non- traditional risk factors. Initial statistical analyses were conducted to identify potential predictors of mortality within three months following hospital discharge, including both classical and non-classical variables. These predictors were then used in a clustering approach, followed by survival analysis to evaluate differences between patient groups. Furthermore, machine learning classification models combined with explainable artificial intelligence techniques were employed to determine the relative contribution of each variable to cluster formation. We investigated both classical and non-classical risk factors to identify subgroups with different survival probabilities. This analysis revealed a set of variables that distinguished two groups with significantly different survival probabilities at 3 months after discharge. By identifying combinations of classical and non-classical factors associated with distinct survival outcomes, this study contributes to a more comprehensive understanding of stroke mortality. These findings may help to inform rehabilitation planning, improve patient management, and support strategies aimed at reducing adverse outcomes following stroke.
- Moral Accounting of Credit in LLM-Assisted Content CreationPublication . Nunes, Joana Rita; Information Management Research Center (MagIC) - NOVA Information Management School; NOVA Information Management School (NOVA IMS)Every day we see news from around the world about the use of AI, GenAI and LLMs in content creation: LLMs are faster, more efficient, and, sometimes, more creative than humans. At first, this seemed incredible, but the integration of LLMs into professional and creative content production raises a business ethical question: When work assisted by these models is recognized and rewarded, how much of that recognition does the human creator legitimately earn? In this research we tested the moral accounting of credit, examining whether the benefit received by a creator is morally proportional to the contribution perceived by individuals, and if there is a moral penalty for creators using LLMs. Research on AI and business ethics has asked whether AI systems can be blamed, trusted, or judged as moral agents, however, it places AI as the object of moral judgment and leaves the human who receives the awards for the LLM-assisted work less examined.
- Quo vadis?Publication . Sturm, Niclas Frederic; Candia, Cristian; Damásio, Bruno; Pinheiro, Flávio L.; Information Management Research Center (MagIC) - NOVA Information Management School; NOVA Information Management School (NOVA IMS)Diversification is at the heart of many economic processes. The principle of relatedness captures this through first-order co-occurrence: local overlap between activities. However, diversification may also be dependent on an entity's structural position in the capability network, which quantifies structural similarities in capability profiles and global connectivity patterns. To overcome this limitation, we propose using node-level embeddings that encode multi-step connectivity as well as neighborhood structure. Next to the structural factors affecting diversification, which is a key contribution of Economic Complexity, forecasting methods from the Machine Learning domain are increasingly applied to the task of diversification paths.
- Validation of Early Vineyard Yield EstimationPublication . Santos Costa, Diogo; NOVA Information Management School (NOVA IMS); Information Management Research Center (MagIC) - NOVA Information Management SchoolReliable early season vineyard yield estimation is increasingly important for planning, quota interpretation and regulatory decision support in wine regions exposed to strong interannual variability. However, the practical value of a predictive model depends not only on its error metrics, but also on the realism of the validation design used to assess its performance. This study evaluates an open data Long Short-Term Memory (LSTM) neural network pipeline for parish level wine grape yield estimation in the Douro Demarcated Region (DDR), comparing Leave-One-Year-Out (LOYO) validation with Walk-Forward (WF) back-testing. The modelling workflow combines Sentinel-derived Normalised Difference Vegetation Index (NDVI) time series with open-access gridded climate variables from AgERA5 and CHIRPS. Estimates are assessed across two early seasonal windows, flowering and veraison, and across multiple territorial levels, from parishes to sub-regions and the whole DDR. LOYO is used as a benchmark-oriented validation strategy that enables comparison across years, while WF imposes a stricter chronological structure in which each target year is estimated using only information available from previous campaigns. The comparison shows how validation design can change the interpretation of model readiness. LOYO supports methodological benchmarking and model comparison, whereas WF provides a more operationally realistic assessment of temporal generalisation. By making this distinction explicit, the study strengthens the methodological credibility of AI-based vineyard yield estimation and contributes to more transparent decision support for precision viticulture and regional wine governance.
- SEATSPublication . Almeida Diogo, Diogo; NOVA Information Management School (NOVA IMS)Cultural venues increasingly use data to guide resource planning and revenue decisions, and a central question is what drives visitor attendance. Yet, prior work is limited on three fronts: it typically considers only weather and holidays as predictors, it relies on proxies such as Google Trends in place of actual attendance, and it studies a single venue, ignoring how drivers differ across venues of different sizes. Using weekly ticketing together with a set of contextual signals from seven venues in Lisbon, we estimate which factors credibly predict attendance and by how much across all venues. We present a Hierarchical Bayesian Structural Time-Series model spanning venues of very different sizes. Beyond the usual weather and holidays, it weighs a set of predictors against weekly ticketing: tourism, website traffic, exhibitions, online reviews, search interest, and macroeconomic conditions.
- Beyond teleoperationPublication . Andrade, José António Nunes; Castelli, Mauro; NOVA Information Management School (NOVA IMS); Information Management Research Center (MagIC) - NOVA Information Management SchoolVision-Language-Action (VLA) models have demonstrated remarkable capabilities in generalized robotic control, yet their scalability is fundamentally bottlenecked by the high cost and low diversity of teleoperated data. While abundant, human demonstration videos cannot be directly utilized for policy training due to the severe morphological differences between human anatomy and robotic manipulators. To bridge this embodiment gap, this work proposes a lightweight retargeting pipeline that kinematically retargets human interaction data (DexYCB) onto a 6-DoF (degree of freedom) manipulator trajectories to fine-tune policies based on pi0.5 architecture. By prioritizing Cartesian positional alignment via constrained Inverse Kinematics (IK) and introducing an object-based grasping heuristic, smooth geometric priors are generated without relying on computationally heavy visual synthesis. Physical evaluations demonstrate that retargeted models significantly outperform standard teleoperation (40.6% success rate), achieving 65.6% success via co-training and a peak 78.1% success rate via two-stage cross-embodiment co-training. Furthermore, evaluations under extreme visual clutter reveal that explicitly retargeted policies exhibit immunity to semantic visual distractors. Finally, we it is examined and analysed the "Terminal State Ambiguity (TSA)", a temporal failure mode where generative models fail to terminate the task, caused when exposed to scenariosby the similar to the nature of human video priors when exposed to extreme visual noise.
- Delimitando a Interface Urbano-Rural em PortugalPublication . Barbosa, Bruno; Oliveira, Sandra; Caetano, Mário; Rocha, Jorge; Information Management Research Center (MagIC) - NOVA Information Management School; NOVA Information Management School (NOVA IMS)O ano de 2023 apresentou uma das temporadas de incêndios mais severas das últimas duas décadas no continente europeu. Eventos extremos como ocorridos em 2017 em Portugal, superam a capacidade de controle e aumentam a exposição e o risco da população. Áreas de transição entre os territórios urbanos e os espaços rurais/florestais, denominadas como Interface Urbano-Uural (IUR), são caracterizadas pela coexistência entre edificações e vegetação. A disposição das habitações e sua localização influencia na vulnerabilidade da edificação frente ao incêndio, sendo as perdas materiais mais prováveis em áreas com densidade estrutural de baixa a intermediária e em regiões com histórico frequente de incêndios. Nosso objetivo é mapear a IUR em Portugal continental para avaliar a exposição aos incêndios entre 2000-2023 utilizando diferentes bases de dados (BD) de edificações — (i) Base Geográfica de Edifícios de Portugal (IURB), (ii) GlobalMLBuildingFootprints (IUR-M) e (iii) World Settlement Footprint (IUR-W) — para avaliar as diferenças geradas nos mapeamentos. Considerou-se área candidata (AC) à IUR aquela que apresente densidade de edificações superior a 6,17 edifícios/km² em uma janela móvel circular com raio de 100-m. Se a AC apresentar mais de 50% de material combustível (MC) é classificada como intermix; as AC que não atinjam este critério, mas estejam localizadas a até 600 metros de fragmentos de MC com área superior a 5 km2 , são classificadas como interface. A exposição aos incêndios foi quantificada utilizando os perímetros das áreas ardidas (AA) entre os anos de 2000-2023. A área total de IUR variou de 7.620 km² (IUR-B) a 13.175 km² (IUR-M). A nível nacional, somente 29% das áreas de Intermix foram identificadas consistentemente pelas três BD, 35 enquanto 47% foram mapeadas por apenas uma BD – predominantemente pela IUR-M (39,5%). As áreas de Interface apresentaram maior concordância, aproximadamente 60%. A proporção de AA na IUR variou significativamente conforme o tipo de interface e a BD utilizada. Nas zonas de Intermix, a IUR-M apresentou a maior proporção média de AA (7,1%), superando a IUR-W (2,9%) e a IUR-B (2,2%). Nas áreas de Interface a proporção média de AA foi inferior, não superando os 3,5% (IUR-M). Embora existam diferenças nos mapas de IUR gerados foram encontradas áreas de sobreposição. Essas áreas mapeadas de forma recorrente representam pontos críticos de exposição aos incêndios e devem receber atenção prioritária.
- Who Reaches the Hubs?Publication . Filonchik, Khristina; NOVA Information Management School (NOVA IMS)Every day, thousands of people travel from suburban neighbourhoods to Lisbon, each depending on a transit hub to start their journey. Yet, which hubs they turn to and how easily they can reach them remain open questions. RQ1 – How does network structure relate to accessibility? To what extent do Node2Vec embeddings and graph centrality metrics explain suburban passengers' ability to reach transit hubs within 15 minutes? RQ2 – Where do structural importance and accessibility diverge? Which hubs are highly central in the metropolitan bus network yet offer poor short-travel-time access, indicating potential first-mile bottlenecks? RQ3 – Can graph-based representations reveal functional hub types? Do embeddings and clustering uncover distinct categories of hubs with characteristic accessibility and network roles?