Selected work

Representative Papers

Multi-Agentic System Leveraging Open-Source LLMs to Mitigate Disinformation Threats

Jun 29, 2026

This study addresses the growing challenge of misinformation amplified by social media and AI technologies, where traditional manual fact-checking proves inadequate. The authors propose a novel multi-agent system that uniquely integrates consensus mechanisms from human annotations, cognitive and knowledge diversity, and hierarchical collaborative structures. Built upon open-source large language models—including LLaMA, Qwen, Kimi, Deepseek, and LLaMA-Nemotron—the framework enables automated detection and verification of false claims. Evaluated on English, Polish, Slovak, and Bulgarian datasets, the approach significantly outperforms monolithic models such as GPT-4 and GPT-3.5 across three key tasks: direct misinformation identification, filtering of claims requiring verification, and detection of verifiable factual statements. The system also demonstrates high transparency and reproducibility.

0 citationsRead paper

mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection

May 04, 2026

This study addresses the challenge of detecting conspiracy theory beliefs in Reddit comments under few-shot learning scenarios. To overcome the scarcity of labeled data, the authors propose a fine-tuning approach that combines data augmentation with self-training, effectively adapting machine-generated text detection techniques to the task of conspiracy theory identification. The method employs binary classification modeling using the Qwen3-32B large language model, demonstrating significant performance gains despite limited supervision. Evaluated on SemEval-2026 Task 10, the approach achieved 8th place out of 52 participating teams (ranking within the top 15%), thereby validating its effectiveness and methodological innovation in low-resource settings for belief detection in social media discourse.

0 citationsRead paper

Capturing Multivariate Dependencies of EV Charging Events: From Parametric Copulas to Neural Density Estimation

Mar 31, 2026

Existing approaches struggle to accurately capture the complex nonlinear dependencies among arrival time, charging duration, and energy demand in electric vehicle charging events. This work proposes the first application of Vine copulas within the Copula Density Neural Estimation (CODINE) framework to this domain, integrating a conditional Gaussian mixture model network to efficiently model high-dimensional joint dependence structures. The proposed method significantly enhances the ability to capture tail dependencies and intricate correlation patterns. Experimental results on three real-world datasets demonstrate that the approach outperforms conventional parametric copula models, achieves performance comparable to state-of-the-art benchmarks, and generates high-quality synthetic charging events.

0 citationsRead paper

Algorithmic Audit of Personalisation Drift in Polarising Topics on TikTok

Mar 21, 2026

This study investigates whether TikTok’s personalized recommendation system exacerbates user polarization on contentious topics such as politics, climate change, and vaccines. By deploying controlled simulated accounts and integrating content annotation, longitudinal analysis of recommendation trajectories, and cross-topic comparisons, the research systematically distinguishes and empirically evaluates three distinct forms of personalization-driven drift: preference-aligned drift, issue-level polarization drift, and stance-level polarization drift. Findings reveal that TikTok’s algorithm exhibits marked topic dependency: it reinforces users’ preexisting political stances in U.S. political content while simultaneously promoting opposing viewpoints, yet demonstrates a neutralizing effect in the context of conspiracy-related material. These results underscore the highly contingent nature of algorithmic influence on polarization, varying significantly across thematic domains.

0 citationsRead paper

The DSA's Blind Spot: Algorithmic Audit of Advertising and Minor Profiling on TikTok

Mar 05, 2026

This study addresses a regulatory gap in the EU Digital Services Act (DSA) Article 28(2), which prohibits profiling-based advertising targeting minors but adopts a narrow definition of “advertising” that excludes undisclosed influencer marketing. Through an algorithmic audit on TikTok, the authors deployed simulated minor and adult accounts, combined with automated content annotation and statistical analysis, to empirically demonstrate that—despite nominal compliance—minors are exposed to substantial volumes of undisclosed commercial content driven by high-intensity interest-based profiling. The intensity of such profiling reaches five to eight times that of formal advertisements shown to adults. The findings underscore the need to broaden the legal definition of “advertising” to encompass emerging forms of commercial content, thereby closing critical enforcement loopholes in digital platform regulation.

0 citationsRead paper

Recent publications

Latest Papers

Multi-Agentic System Leveraging Open-Source LLMs to Mitigate Disinformation Threats

Jun 29, 2026

This study addresses the growing challenge of misinformation amplified by social media and AI technologies, where traditional manual fact-checking proves inadequate. The authors propose a novel multi-agent system that uniquely integrates consensus mechanisms from human annotations, cognitive and knowledge diversity, and hierarchical collaborative structures. Built upon open-source large language models—including LLaMA, Qwen, Kimi, Deepseek, and LLaMA-Nemotron—the framework enables automated detection and verification of false claims. Evaluated on English, Polish, Slovak, and Bulgarian datasets, the approach significantly outperforms monolithic models such as GPT-4 and GPT-3.5 across three key tasks: direct misinformation identification, filtering of claims requiring verification, and detection of verifiable factual statements. The system also demonstrates high transparency and reproducibility.

0 citationsRead paper

mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection

May 04, 2026

This study addresses the challenge of detecting conspiracy theory beliefs in Reddit comments under few-shot learning scenarios. To overcome the scarcity of labeled data, the authors propose a fine-tuning approach that combines data augmentation with self-training, effectively adapting machine-generated text detection techniques to the task of conspiracy theory identification. The method employs binary classification modeling using the Qwen3-32B large language model, demonstrating significant performance gains despite limited supervision. Evaluated on SemEval-2026 Task 10, the approach achieved 8th place out of 52 participating teams (ranking within the top 15%), thereby validating its effectiveness and methodological innovation in low-resource settings for belief detection in social media discourse.

0 citationsRead paper

Capturing Multivariate Dependencies of EV Charging Events: From Parametric Copulas to Neural Density Estimation

Mar 31, 2026

Existing approaches struggle to accurately capture the complex nonlinear dependencies among arrival time, charging duration, and energy demand in electric vehicle charging events. This work proposes the first application of Vine copulas within the Copula Density Neural Estimation (CODINE) framework to this domain, integrating a conditional Gaussian mixture model network to efficiently model high-dimensional joint dependence structures. The proposed method significantly enhances the ability to capture tail dependencies and intricate correlation patterns. Experimental results on three real-world datasets demonstrate that the approach outperforms conventional parametric copula models, achieves performance comparable to state-of-the-art benchmarks, and generates high-quality synthetic charging events.

0 citationsRead paper

Algorithmic Audit of Personalisation Drift in Polarising Topics on TikTok

Mar 21, 2026

This study investigates whether TikTok’s personalized recommendation system exacerbates user polarization on contentious topics such as politics, climate change, and vaccines. By deploying controlled simulated accounts and integrating content annotation, longitudinal analysis of recommendation trajectories, and cross-topic comparisons, the research systematically distinguishes and empirically evaluates three distinct forms of personalization-driven drift: preference-aligned drift, issue-level polarization drift, and stance-level polarization drift. Findings reveal that TikTok’s algorithm exhibits marked topic dependency: it reinforces users’ preexisting political stances in U.S. political content while simultaneously promoting opposing viewpoints, yet demonstrates a neutralizing effect in the context of conspiracy-related material. These results underscore the highly contingent nature of algorithmic influence on polarization, varying significantly across thematic domains.

0 citationsRead paper

The DSA's Blind Spot: Algorithmic Audit of Advertising and Minor Profiling on TikTok

Mar 05, 2026

This study addresses a regulatory gap in the EU Digital Services Act (DSA) Article 28(2), which prohibits profiling-based advertising targeting minors but adopts a narrow definition of “advertising” that excludes undisclosed influencer marketing. Through an algorithmic audit on TikTok, the authors deployed simulated minor and adult accounts, combined with automated content annotation and statistical analysis, to empirically demonstrate that—despite nominal compliance—minors are exposed to substantial volumes of undisclosed commercial content driven by high-intensity interest-based profiling. The intensity of such profiling reaches five to eight times that of formal advertisements shown to adults. The findings underscore the need to broaden the legal definition of “advertising” to encompass emerging forms of commercial content, thereby closing critical enforcement loopholes in digital platform regulation.

0 citationsRead paper