Over the years, social media has had a considerable impact on the way we share information and send messages. With this comes the problem of the rapid distribution of fake news which can have negative impacts on both individuals and society. Given the potential negative influence, detecting unmonitored ‘fake news’ has become a critical issue in mainstream media. While there are recent studies that built machine learning models that detect fake news in several languages, lack of studies in detecting fake news in the Arabic language is scare. Hence, in this paper, we study the issue of fake news detection in the Arabic language based on textual analysis. In an attempt to address the challenges of authenticating news, we introduce a supervised machine learning model that classifies Arabic news articles based on their context’s credibility. We also introduce the first dataset of Arabic fake news articles composed through crowdsourcing. Subsequently, to extract textual features from the articles, we create a unique approach of forming Arabic lexical wordlists and design an Arabic Natural Language Processing tool to perform textual features extraction. The findings of this study promises great results and outperformed human performance in the same task.

The extensive use of social media platforms has promoted the rapid spread of fake news on the internet, such as fake reviews, rumors, and propaganda. Although these terminologies have different objectives, they share the aim of causing harm in the form of fake news. This study presents an Arabic fake news detection framework to overcome the widespread fake news phenomenon. The proposed framework introduces the first Arabic fake news dataset compiled by passing through strict guidelines to produce fake articles composed of humans and the generative pre-trained transformer (GPT). First, we performed human-based experiments to evaluate the ability of humans to distinguish real news articles from fake news articles. Our findings reveal that humans could roughly identify half of the fake articles from humans or GPT, raising concerns about their ability to detect fake news. This highlights the growing concern surrounding fake news, especially because GPT demonstrates the ability to generate fake news that closely resembles human-created content, further amplifying the issue. To address this issue, we performed the same task using Deep Learning (DL) and transformer-based methods with different word embeddings. Across all the employed models, the study revealed that the innovative transformer-based model, ARBERT, outperformed the DL models, reaching an accuracy of 78% in classifying real and fake news generated by humans and GPT. The findings suggest effective techniques for addressing and resolving this issue.

While customer reviews are crucial for businesses to maintain their standing in the marketplace, some may employ humans to create favorable reviews for their benefit. However, advances in artificial intelligence have made it less complex to create these reviews, which now rival real ones written by humans. This poses a significant challenge in distinguishing between genuine and artificially generated reviews, thereby impacting consumer trust and decision-making processes. Research has been conducted to classify whether English reviews were authored by humans or computers. However, there is a notable scarcity of similar studies conducted in Arabic. Moreover, the potential of ensemble learning (EL) techniques, such as soft voting, to enhance model performance remains underexplored. This study conducts a comprehensive empirical analysis using various models, including traditional machine learning, deep learning, and transformers, with an investigation into ensemble techniques, like soft voting, to classify human and computer-generated Arabic reviews. Integrating top logistic regression (LR) and convolutional neural network (CNN) models, it achieves an accuracy of 89.70%, akin to AraBERT’s 90.0%. Additionally, a thorough textual analysis, covering parts of speech (POS), emotions, and linguistics reveals significant linguistic disparities between human and computer-generated reviews. Notably, computer-generated reviews exhibit a substantially higher proportion of adjectives (6.3%) compared to human reviews (0.46%), providing crucial insights for discerning between the two review types. The results not only advance natural language processing (NLP) in Arabic but also have significant implications for businesses combating the influence of fake reviews on consumer trust and decision-making.

This paper demonstrates Tasaheel, an automative Arabic textual analysis tool. It offers two types of textual analysis utilities: traditional natural language processing tasks; such as stemming, segmentation, normalization, name entity recognition, and part of speech tagging, by integrating open-sourced Arabic natural language processing packages. The second type of utility offers novel corpus analysis methods, including a detailed word use summary of part of speech, emotion, polarity, linguistics, and domain-specific word labeling. Tasaheel is the first Arabic tool that analyses affixes as a type of comprehensive textual analysis, along with options to search and handle data. We anticipate that Tasaheel will be a conducive tool for Arabic textual analysis.

The rapid dissemination of fake news in the digital era has become a pressing concern. The ease of generating and manipulating fake content, including images, text, audio, and videos, has significantly fueled the spread of misinformation on social media platforms. These platforms often lack rigorous editorial scrutiny, exacerbating this problem. Although recent studies have explored multimodal fake news detection to learn shared representations of textual and visual information, they often learn discrete latent representations, merely concatenations of multimodal features. Simple concatenation or summation operations hinder the dynamic interaction of multimodal features. Furthermore, most models rely on additional subtasks, such as reconstruction and event discrimination. The performance of these models depends heavily on subtasks, which can be mathematically complex and time-consuming. This reliance limits the ability of researchers to explore different modeling assumptions freely. This study introduces a novel approach that integrates a probabilistic algorithm with a deep neural network to effectively capture the uncertainties and diversities in the shared latent representation of multimodal data. Specifically, our model utilizes continuous latent representations by leveraging a smoothed Dirichlet distribution, facilitating the identification of shared hidden patterns across textual and visual modalities. In addition, our model demonstrates the powerful properties of generative models when integrated with neural network models. Our results underscore the potential of integrating a probabilistic algorithm with a deep neural network to address the challenges of fake news detection in a multimodal setting. To support further research and reproducibility, we made the code related to this work publicly accessible.

A well-known issue for social media sites consists of the hazy boundaries between malicious false news and protected speech satire. In addition to the protective measures that lessen the exposure of false material on social media, providers of fake news have started to pose as satire sites in order to escape being delisted. Potentially, this may cause confusion to the readers as satire can sometimes be mistaken for real news, especially when their context or intent is not clearly understood and written in a journalistic format imitating real articles. In this research, we tackle the issue of classifying Arabic satiric articles written in a journalistic format to detect satirical cues that aid in satire classification. To accomplish this, we compiled the first Arabic satirical articles dataset extracted from real-world satirical news platforms. Then, a number of classification models that integrate a variety of feature extraction techniques with machine learning, deep learning, and transformers to detect the provenance of linguistic and semantic cues were investigated, including the first use of the ArabGPt model. Our results indicate that BERT is the best-performing model with F1-score reaching 95%. We also provide an in-depth lexical analysis of the formation of Arabic satirical articles. The lexical analysis provides insights into the satirical nature of the articles in terms of their linguistic word uses. Finally, we developed a free open-source platform that automatically organizes satirical and non-satirical articles in their correct classes from the best-performing model in our study, BERT. In summary, the obtained results found that pretrained models gave promising results in classifying Arabic satirical articles.