Keng Hoon Gan

The rapid spread of misinformation in Arabic news headlines poses a growing challenge to digital media integrity, given the scarcity of Arabic-specific automated detection tools relative to English-centric systems. Headlines are brief and context-limited, yet their lexical and stylistic patterns encode strong cues of veracity or deception, making headline-only detection practically urgent and linguistically tractable. This study investigates automatic fake-news detection using Arabic headlines, leveraging five heterogeneous corpora and their unified combination. English sets were incorporated via neural machine translation with light post-normalization that preserves stylistic cues, yielding a heterogeneous cross-domain corpus. A systematic analysis of linguistic and stylistic indicators reveals stable asymmetries between fake and real headlines that recur across domains. We evaluate approaches from classical TF-IDF baselines to Arabic-specialized transformers, and propose a late-fusion strategy coupling transformer representations with discriminative engineered features. Transformers consistently outperform classical baselines, confirming that subword representations effectively capture semantic and stylistic regularities in short Arabic texts. Late fusion yields statistically significant improvements only on datasets with prominent numeric or temporal cues; on the unified corpus, McNemar’s exact test confirms that fusion gains are non-significant, indicating that subword encoders already internalize the surface-level cues captured by the engineered features. Even where accuracy differences are marginal, interpretable features enhance explainability.

The headline plays a pivotal role in news content, designed to capture attention and provoke curiosity, thereby encouraging readers to engage with the full article or share it on social media. However, social media platforms have increasingly facilitated the rapid spread of misinformation, with misleading headlines often serving as vectors for fake news. Traditional methods, such as manual fact-checking, are inadequate in addressing the scale and speed of misinformation dissemination. This study investigates computational approaches for fake news detection in Arabic headlines using machine learning, deep learning, and transformer-based models specifically pre-trained for the Arabic language. We utilize two benchmark datasets: the AFND dataset—a large-scale collection of Arabic news headlines—and the Arabic News Stance (ANS) dataset, a smaller, unbalanced corpus of annotated claims. Transformer-based models demonstrated superior performance across both datasets. AraBERTv02 achieved the highest accuracy on the large-scale AFND dataset at 70.41%, while AraELECTRA reached 77% accuracy on the smaller ANS dataset. These results highlight the robustness and effectiveness of transformer architectures in capturing nuanced linguistic features relevant to fake news detection in Arabic. Additionally, our experiments reveal that preprocessing choices—such as the application of stemming—impact performance differently across model types, emphasizing the need for tailored preprocessing strategies. This study contributes to the growing body of research in Arabic natural language processing and offers practical insights to improve automated detection of fake news in Arabic media.

Feature selection is a crucial technique in text classification for improving the efficiency and effectiveness of classifiers or machine learning techniques by reducing the dataset’s dimensionality. This involves eliminating irrelevant, redundant, and noisy features to streamline the classification process. Various methods, from single feature selection techniques to ensemble filter-wrapper methods, have been used in the literature. Metaheuristic algorithms have become popular due to their ability to handle optimization complexity and the continuous influx of text documents. Feature selection is inherently multi-objective, balancing the enhancement of feature relevance, accuracy, and the reduction of redundant features. This research presents a two-fold objective for feature selection. The first objective is to identify the top-ranked features using an ensemble of three multi-univariate filter methods: Information Gain (Infogain), Chi-Square (Chi2), and Analysis of Variance (ANOVA). This aims to maximize feature relevance while minimizing redundancy. The second objective involves reducing the number of selected features and increasing accuracy through a hybrid approach combining Artificial Bee Colony (ABC) and Genetic Algorithms (GA). This hybrid method operates in a wrapper framework to identify the most informative subset of text features. Support Vector Machine (SVM) was employed as the performance evaluator for the proposed model, tested on two high-dimensional multiclass datasets. The experimental results demonstrated that the ensemble filter combined with the ABC+GA hybrid approach is a promising solution for text feature selection, offering superior performance compared to other existing feature selection algorithms.

Cross-modal retrieval, which aims to search for semantically relevant data across different modalities, has received increasing attention in recent years. Deep learning, with its ability to extract high-level representations from multimodal data, has become a popular approach for cross-modal retrieval. In this paper, we present a comprehensive survey of deep learning techniques for cross-modal retrieval including 35 papers published in recent years. The review is organized into four main sections, covering traditional subspace learning methods, deep learning, and machine learning-based approaches, techniques based on large multi-modal models, and an analysis of datasets used in the field of cross-modal retrieval. We compare and analyze the performance of different deep learning methods on benchmark datasets, the result shows that although a large number of innovative methods have been proposed, there are still some problems that need to be solved, such as multi-modal feature alignment, multi-modal feature fusion, and subspace learning, as well as specialized datasets.