Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yeganeh Abdollahinejad has a total of 9 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • E-CaTCH: Event-Centric Cross-Modal Attention With Temporal Consistency and Class-Imbalance Handling for Misinformation Detection

    by Ahmad Mousavi, Yeganeh Abdollahinejad, Roberto Corizzo, Nathalie Japkowicz, Zois Boukouvalas

    Detecting multimodal misinformation on social media remains challenging due to inconsistencies between modalities, changes in temporal patterns, and substantial class imbalance. Many existing methods treat posts independently and fail to capture the event-level structure of evolving narratives that connect them across time and modality. We propose Event-Centric Cross-Modal Attention with Temporal Consistency and Class-Imbalance Handling (E-CaTCH), a modular and scalable framework that provides interpretable intermediate representations through event-centric modeling and attention-based fusion. E-CaTCH clusters posts into pseudoevents based on textual similarity and temporal proximity, then processes each event independently. Within each event, textual and visual features are extracted using pretrained BERT and ResNet encoders, refined via intramodal self-attention, and aligned through bidirectional cross-modal attention. A soft gating mechanism fuses these representations to form contextualized, content-aware embeddings of each post. To model temporal evolution, E-CaTCH segments events into overlapping time windows and uses a trend-aware LSTM to encode narrative progression over time. Classification is performed at the event level, enabling better alignment with real-world misinformation dynamics. To address class imbalance and promote stable learning, the model integrates adaptive class weighting and temporal consistency regularization. Extensive experiments on three benchmark datasets (Fakeddit, IND, and COVID-19 MISINFOGRAPH) demonstrate the effectiveness of this approach, with E-CaTCH achieving state-of-the-art accuracy of 95.5% on Fakeddit and consistently outperforming strong baselines. Cross-dataset evaluations further demonstrate its robustness, generalizability, and practical applicability across diverse misinformation scenarios.

  • Open Access

    Systematic Literature Review of Machine Learning Methods for Emotion Recognition Using EEG and Physiological Signals in Healthcare

    by Yeganeh Abdollahinejad, Sayed Mohsin Reza, B. Ashwini, Ahmad Mousavi

    Emotion recognition (ER) from physiological signals, particularly electroencephalography (EEG), is increasingly studied for healthcare uses such as mental health monitoring, stress assessment, and patient-adaptive support. Compared with behavioral observation alone, biosignals including EEG, electrocardiography (ECG), electrodermal activity (EDA), photoplethysmography (PPG; reported as blood volume pulse, BVP), and respiration provide objective correlates of affect when overt cues are limited or unreliable. This systematic review synthesizes 70 peer-reviewed studies published from 2014 to 2024. We summarize (i) physiological modalities and multimodal fusion strategies, (ii) machine learning and deep learning methods most frequently used, (iii) datasets, evaluation protocols, and metrics with explicit separation of subject-dependent (SD) and subject-independent (SI) or leave-one-subject-out (LOSO) settings, and (iv) explainable AI (XAI) practices that aim to improve interpretability and clinical trust. Across the literature, SD results are consistently higher than SI/LOSO results; domain adaptation and multimodal fusion partially close this gap. Paired within-study contrasts show small but reliable gains on DEAP for fusion over EEG alone by about 3 to 5 percentage points, and larger gains of about 25 to 35 percentage points on SEED-IV or SEED-V under participant-generic or LOSO protocols; WESAD also shows meaningful improvements. Common barriers to clinical translation include artifacts and noise, inter-subject variability, limited cohort diversity and external validation, privacy and governance constraints, and nonstandardized interpretability. We outline directions for real-world adoption, including protocol-aware benchmarking, longitudinal and cross-site validation, streamlined sensor configurations, personalization through brief calibration or domain adaptation, and clinically useful XAI that supports auditability and aligns with healthcare regulatory expectations.


  • Open Access

    EMO-CARE: EEG Multi-Scale Temporal Modeling With Channel-Aware Feature Attention for Robust Subject-Independent Emotion Recognition

    by Yeganeh Abdollahinejad, Ahmad Mousavi, Petros Siaplaouras, Zois Boukouvalas, Roberto Corizzo

    Electroencephalography (EEG)-based emotion recognition holds promise for real-time mental health monitoring, adaptive interfaces, and affective computing. However, accurate prediction across individuals remains challenging due to inter-subject variability and the non-stationary nature of EEG signals. To address this, we propose EMO-CARE, a lightweight deep learning framework that integrates multi-scale temporal convolutional networks with feature-level self-attention operating on multi-scale temporal representations. This architecture captures emotional patterns across diverse neural timescales while adaptively weighting multi-scale temporal features based on their relevance. Evaluated under the rigorous Leave-One-Subject-Out (LOSO) protocol on three benchmark datasets: SEED, SEED-V, and DREAMER, EMO-CARE achieves state-of-the-art accuracy with low inference latency. Extensive ablation experiments demonstrate the contribution of each architectural component, and the learned attention patterns align with known emotion-related neural activity. These findings collectively highlight EMO CARE's effectiveness in achieving subject-independent generalization and real-time applicability for EEG based emotion recognition.