Seyed Sahand Mohammadi Ziabari

Standard models for predicting speaker attributes from text often fail to manage multiple, interdependent attributes simultaneously, and existing Reinforced Multiple-Instance Learning (RL–MIL) frameworks are typically limited to single-task prediction. To address this, we propose and evaluate a Reinforced, Multi-Task, Multiple-Instance Learning framework, a novel framework that enhances Reinforced Multiple-Instance Learning with Multi-Task Learning (MTL) to predict a speaker’s age, gender, and political party from congressional speeches. A central goal of our work was to investigate the optimal parameter-sharing strategy. We compared a fully shared architecture against a dynamic task clustering mechanism designed to mitigate negative transfer by adaptively grouping related tasks. Our results demonstrate that the multi-task approach significantly outperforms single-task baselines. Interestingly, the model with a fully shared representation achieved the highest macro-average F1 score of 0.668, suggesting the tasks in this dataset were sufficiently correlated to benefit from shared features without needing adaptive separation. This work contributes a more effective method for weakly supervised, multi-attribute prediction and provides crucial insights into the trade-offs between different parameter-sharing strategies in MTL.

The discovery of various CRISPR–Cas systems has revolutionized genome engineering by enabling precise and programmable nucleic acid targeting. Continued exploration of CRISPR diversity, together with advances in computational modeling and deep learning (DL)-based design, has expanded the potential to manipulate nearly any genomic locus, thereby accelerating both basic research and therapeutic applications. This review systematically provides a structured and up-to-date overview of CRISPR–Cas technologies, including their classification, computational modeling strategies, and the integration of machine learning (ML) and DL approaches to predict guide RNA (gRNA) efficiency and specificity. The emphasis is placed on studies published between 2019 and 2025, which highlight significant progress in modeling Cas–gRNA–DNA interactions, optimizing on/off-target prediction, and developing comprehensive CRISPR-related datasets. By synthesizing recent developments in CRISPR biology, computational simulations, and artificial intelligence, this review underscores the importance of interdisciplinary integration to improve the accuracy, safety, and scalability of next-generation genome-editing systems.

Sports spectatorship is fundamentally affective, and the rise of fan-generated data has enabled large-scale computational analysis of fan sentiment and emotion. This paper synthesises 64 studies (2000–2025) that infer fan affect from (i) digital text and online discourse, (ii) broadcast-linked second-screen interaction, and (iii) in-venue or non-textual signals such as crowd audio, video, and wearable sensing. We organise the literature into three modality-based categories: text-centric discourse analytics, second-screen/social-TV behaviour, and in-venue or multimodal sensing. Across studies, a consistent empirical pattern is event-driven synchrony: aggregate affective signals shift rapidly around salient match events and controversy. However, three structural limitations constrain behavioural inference and generalisability: strong platform dependence (especially Twitter/X), overreliance on coarse polarity sentiment, and conceptual slippage between affective expression and behaviour. Research on harmful expressions (e.g., toxicity, hate speech) is expanding and behaviourally relevant, but introduces methodological and ethical challenges. Overall, the literature is effective at detecting affective expression but weaker in linking affect to explicit behavioural outcomes or integrating evidence across modalities. We highlight directions for behaviour-centred, multimodal fan analytics, including improved construct validity, clearer outcome definitions, cross-modality integration, and stronger cross-platform and ethical considerations.

Accurate determination of human epidermal growth factor receptor 2 (HER2) status is a critical component of breast cancer prognosis and treatment planning. Conventional diagnostic techniques, such as immunohistochemistry (IHC) and fluorescence in situ hybridization (FISH), are clinically established but remain invasive, time-consuming, costly, and sensitive to pre-analytical and interpretative variability. Motivated by the need for scalable and data-driven decision-support tools, this study proposes a hybrid ensemble machine learning framework for non-invasive HER2 status prediction using routinely available clinical and immunohistochemical features. A retrospective dataset comprising 624 breast cancer patients from Mahdieh Clinic (Kermanshah, Iran) was analyzed using a structured preprocessing pipeline including normalization and class balancing. The proposed framework integrates multiple tree-based classifiers, Random Forest, XGBoost, and LightGBM, through ensemble strategies and enhances predictive robustness using membership-function feature engineering to capture gradual transitions in clinically relevant biomarkers. Decision threshold optimization was further applied to improve classification balance in borderline cases. The proposed ensemble framework achieved an accuracy of 0.816, an F1-score of 0.814, and an area under the receiver operating characteristic curve (AUC) of 0.862 on a held-out test set, demonstrating performance comparable to the best-performing individual classifier. These results indicate that ensemble learning combined with smooth membership-based feature representations can provide a reliable decision-support framework for HER2 status prediction, although further external validation is required before clinical use.

HER2 status determination is a crucial task in breast cancer prognosis and treatment,1 yet traditional diagnostic methods such as immunohistochemistry (IHC) and fluorescence in situ2 hybridization (FISH) are invasive, time-consuming, and costly. Motivated by the need for scalable3 and data-driven predictive approaches, we propose a hybrid machine learning framework that4 integrates ensemble learning with fuzzy modeling for HER2 prediction using routinely available5 clinical and immunohistochemical data. A dataset comprising 624 breast cancer patients from6 Mahdieh Clinic (Kermanshah, Iran) was analyzed, with extensive feature engineering, scaling, and7 class balancing applied. We developed an ensemble framework based on tree-based learners (Random8 Forest, XGBoost, and LightGBM), combined through ensemble strategies and enhanced using fuzzy9 feature representations and decision threshold optimization. The proposed hybrid model achieved10 an accuracy of 0.816, an F1-score of 0.814, and an area under the ROC curve (AUC) of 0.862 on11 the held-out test set, demonstrating strong discriminative capability and balanced classification12 performance. This work highlights the potential of hybrid fuzzy–ensemble learning for uncertainty-13 aware predictive analytics in biomedical decision support, aligning with the journal’s focus on14 information processes, intelligent systems, and data mining.

Subseasonal forecasting of the North Atlantic Oscillation (NAO), which spans two to eight weeks, is essential for climate-sensitive sectors such as energy and agriculture. However, it remains challenging due to the chaotic nature of atmospheric processes and the influence of tropical conditions, resulting in limited forecast skill and interpretability beyond approximately two weeks. This study aims to address the limited predictive accuracy in forecasting the NAO beyond 14 days by focusing on decadal representations of the NAO and utilizing "forecasts of opportunity." We propose two machine learning models: an Artificial Neural Network (ANN)-based model and a Gradient Boosting Decision Tree (GBDT)-based model. These models use daily mean sea level pressure (SLP) and 50 m wind (V50) data from the CESM2-LE dataset and are enhanced by explainable artificial intelligence (XAI) techniques, namely SHapley Additive exPlanations (SHAP) and Integrated Gradients (IG) to ensure transparency and interpretability. The models achieve accuracies of 0.769 (ANN) and 0.773 (GBDT) at a 14-day lead time for the top 20% most confident predictions, surpassing the baseline models across lead times. Validation against ERA5 reanalysis data shows accuracies of 0.732 (ANN) and 0.739 (GBDT), confirming robust predictive relationships. By analyzing the NAO over decadal timescales, we uncover patterns in variability and, through the use of XAI, identify key spatial predictors that drive these predictions. These findings not only enhance transparency and predictive skill in subseasonal NAO forecasting but also establish a solid foundation for future research. This future research can build upon our work by integrating additional climate variables known to influence the NAO.