Leonidas Boutsikaris

Graph neural networks remain challenging to apply to traffic forecasting, often failing to consistently outperform simple baselines. We observe that the strongest baseline is regime-dependent: last-value persistence dominates at short horizons on smooth freeway data, while time-of-day climatology prevails at longer horizons and on bursty arterial networks. Rather than treating this as a limitation, we propose a learnable approach that automatically selects the optimal baseline. Our method anchors predictions to a learned, per-horizon convex blend of persistence and climatology, introducing only twelve scalar parameters. This learned anchor recovers whichever baseline is locally most effective, allowing the model to focus on capturing residual variations that neither baseline captures. We evaluate on six public benchmarks (METR-LA, PEMS-BAY, PEMS03/04/07/08) spanning traffic speed and flow data under standard 70/10/20 chronological splits with masked evaluation metrics and holiday-aware climatology. Our anchored temporal models consistently beat both baseline methods on the 12-step average across all datasets, and outperform at every horizon on five of the six benchmarks. When integrated into two strong architectures (STID and Graph WaveNet), the anchor yields substantial gains at long horizons where climatology is most informative. Notably, within our lightweight framework, learned spatial graph components do not improve accuracy and can slightly degrade performance, a finding we analyze and discuss.

Understanding how political narratives evolve and maintain coherence in digital discourse presents fundamental challenges for computational analysis. We introduce a framework that combines subject-verb-object (SVO) triplet extraction with graph-based reinforcement learning to model narrative coherence in communication. Our methodology constructs dynamic narrative graphs from congressional tweets, where nodes represent political entities and edges capture semantic relationships through verb connections. We develop a coherence metric based on graph triangulation, clustering coefficients, and small-world properties, enabling quantitative assessment of narrative structure. The reinforcement learning component continuously optimizes graph configurations to maximize coherence while incorporating sentiment analysis to capture emotional resonance. Experimental validation using 100 congressional tweets demonstrates stable coherence scores, with clear Q-learning convergence patterns. Our analysis reveals distinct narrative strategies: Republicans employ more tightly interconnected security-focused messaging, while Democrats utilize broader, less centralized discourse spanning diverse social issues. The framework successfully identifies three dominant narrative clusters around Trump accountability, Biden policy implementation, and national security themes.

An ethical evaluation of Large Language Models (LLM) requires good background knowledge regarding their technical bases; only then is it possible to see how their architectures and training methodologies lead to ethical concerns and biases. The present review evaluates qualitatively the reports produced for GPT, DeepSeek, Claude, Gemini, LLaMA, and Qwen on different aspects of their technical specifications in terms of how design choices shape ethical considerations like bias, fairness, and transparency. We examine how model architecture, data curation strategies, and ethical bias mitigation and propagation is being handled by different large language models. Through our findings, it appears that proprietary models adopt advanced solutions to manage bias but without being transparent. On the other hand, open models are amenable to an independent audit; however, the datasets that are available for public scrutiny usually harbor some biases that are inherent to the datasets themselves. Thus, in our discussions, we emphasize the need for standardized evaluation frameworks to ensure that ethical considerations are parallel to advances in AI in the future.

This paper introduces a novel hybrid algorithm that enhances the Louvain community detection algorithm by integrating spatial mapping based on opinion distributions on social networks. Nodes are represented by individuals' opinions embedded in a geometric space. Minimum Bounding Rectangles (MBRs) encapsulate the spatial distribution of opinions within communities and by incorporating individual opinions the algorithm identifies communities that are both structurally and opinion-connected. Experimental results show that the modified algorithm preserves the structural integrity of networks while improving opinion cohesion within communities and enhancing opinion separation across them. These findings highlight its effectiveness in balancing structural and opinion-based clustering.

In this article, we explore the impact that social media has on information consumption and community dialogue. We discuss concepts like filter bubbles, echo chambers, and polarization. Although filter bubbles tailor content to individual preferences, echo chambers reinforce ideological conformity by surrounding individuals with like-minded peers. These phenomena intensify the polarization that society is facing by intensifying divisions. However, research challenges the conventional understanding of filter bubbles, echo chambers, and polarization, suggesting that our online communities are less polarized and that conflict and disagreement play a greater role in online discussions. We also explore the Friedkin and Johnsen model, which offers us insights into opinion formation within social networks, highlighting internal and external conflict and disagreement dynamics. The nature of social media amplifies disagreement and conflict, underscoring the need for strategies that promote diversity. Research should explore opinion models and societal dynamics that address platform design in shaping information ecosystems.

This paper provides a review of classification methods for irony and sarcasm in textual data. It explores different approaches to detecting irony and sarcasm, their definitions, distinguishing features, and detection methodologies. The study examines a range of datasets used in irony and sarcasm detection research, including short-text datasets from social media platforms and long-text datasets from product reviews and discussion forums. Additionally, the paper discusses the various features employed in sarcasm detection experiments, such as lexical, pragmatic, hyperbole, semantic, syntactic, sentiment, and contextual features. It also explores the classification methodologies used. The paper concludes by analyzing each classification method and highlighting the latest trends in irony and sarcasm detection.

In this work, we propose a multidimensional extension of the input of Friedkin-Johnsen (FJ) model to more accurately capture the diversity of opinions in social networks. Opinion dynamics models, including the FJ model, represent opinions on a binary scale, limiting their capacity to reflect real-world opinion formation. Our approach expands this binary framework by introducing multiple dimensions for each perspective within an opinion, allowing for the representation of varied intensities and sub-dimensions. We model opinions as polygons, project them into 3D space using the z-axis for metrics such as mean opinion or social influence, and analyze the resulting polyhedra to quantify opinion diversity within communities. This approach enables us to compute a new metric-Opinion Diversity-based on the volume of polyhedra in the network, gaining a clear insight into the relationship between opinion diversity and social dynamics.