[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86549-en":3,"doc-seo-86549-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86549,13056703019662,"Evangeline","https://ap-avatar.wpscdn.com/avatar/be000253a8e92610077?_k=1778726343310543188",8,"Research & Report","FAD-SA-GRU Enhancing Hate Speech Detection in Algerian Dialect Through Feature-Augmented Self-Attention GRU Networks","The widespread adoption of social media enables instant exchange of information and opinions, but also accelerates the distribution of abusive and hateful content, creating social, psychological, and ethical risks. This paper targets automatic hate speech detection in Algerian Arabic (Darija), where linguistic diversity across Arabic, French, and Arabizi complicates modeling. A comparative study covers TF–IDF baselines, recurrent networks, and Transformer models, and proposes FAD-SA-GRU using multi-embedding fusion and a self-attention-enhanced GRU encoder, achieving 93.2% accuracy and 97.0% ROC-AUC on binary-labeled comments.","arXiv :2607 . 1 1279v 1 [ cs .CL] 13 Jul 2026  \nFAD-SA-GRU: ENHANCING HATE SPEECH DETECTION IN ALGERIAN DIALECT THROUGH FEATURE-AUGMENTED SELF-ATTENTION GRU NETWORKS  \nA PREPRINT  \n(1)Sara YAKOUBI,(1)Ikram KHALFALLAH,(2)Kenza KHELKHAL,(2)Dihia LANASRI  \n(1)USTHB,(2)ATM Mobilis  \nAlgiers, Algeria  \n[yakoubi.sara21@gmail.com](yakoubi.sara21@gmail.com) , [ikramkhalf736@gmail.com](ikramkhalf736@gmail.com)  \n[khelkhalkenza88@gmail.com](khelkhalkenza88@gmail.com) , [dihia.lanasri@gmail.com](dihia.lanasri@gmail.com)  \nJuly 14, 2026  \nABSTRACT  \nThe widespread adoption of social media platforms has transformed online communication by enabling users to exchange information and opinions instantly. However, these platforms have also facilitated the rapid dissemination of abusive and hateful content, posing significant social, psychological, and ethical challenges. Hate speech can incite discrimination, harassment, and violence against individuals or communities based on attributes such as ethnicity, religion, gender, nationality, or political affiliation. Consequently, the automatic detection of hate speech has become a major research topic in natural language processing (NLP), attracting considerable attention from both academia and industry as an essential component of content moderation systems.  \nThis paper investigates the automatic detection of hate speech in Algerian Arabic dialect (Darija) on social media. Hate speech detection in Algerian Darija remains a challenging task due to the dialect’s linguistic diversity, characterized by the coexistence of Arabic, French, and Arabizi (Arabic written using the Latin alphabet) . To address these challenges, we conduct a comprehensive comparative study of four categories of text classification approaches: (1) traditional machine learning models using TF–IDF feature representations,(2) deep learning models based on recurrent neural networks,(3) Transformer-based language models, including DziriBERT and multilingual BERT, and (4) a novel hybrid architecture, FAD-SA-GRU, which combines semantic representations from DZ FastText, DZ AraVec, and DziriBERT through a multi-embedding fusion strategy, followed by a self-attention-enhanced GRU encoder.  \nExperiments are conducted on a manually annotated dataset of Algerian Darija social media comments labeled for binary hate speech classification. The proposed FAD-SA-GRU model consistently outperforms all baseline approaches, achieving an accuracy of 93.2%, precision of 93.4%, recall of 91.0%, F1-score of 92.1%, and ROC-AUC of 97.0% . The experimental results demonstrate the effectiveness of combining complementary embedding representations with attention-based sequence modeling for robust hate speech detection in low-resource dialectal Arabic.  \nKeywords Hate Speech detection · Algerian dialect · Arabizi · FAD-SA-GRU · Natural Language Processing · GRU  \n1 Introduction  \nThe rapid expansion of social media has profoundly transformed the way people communicate, share information, and express opinions. Platforms such as Facebook, X (formerly Twitter), Instagram, and YouTube have become indispensable channels for public discourse, allowing millions of users to interact in real time regardless of geographical boundaries. While these platforms promote connectivity, collaboration, and freedom of expression, they have also facilitated the proliferation of harmful online content, including cyberbullying, misinformation, abusive language, and  \nhate speech. The sheer volume of user-generated content makes manual moderation increasingly impractical, motivating the development of automated solutions capable of detecting and filtering harmful content efficiently. Consequently, hate speech detection has emerged as a major research topic in Natural Language Processing (NLP), with applications in online content moderation, public safety, and the protection of vulnerable communities.  \nHate speech is generally understood as any form of communication that attacks, ","cbCaibNoQItnuDm3","https://ap.wps.com/l/cbCaibNoQItnuDm3","pdf",5455094,5,1,17,"English","en",105,"# Introduction\n## Hate Speech Detection on Social Media\n## Definition and Impact of Hate Speech\n## Challenges for Low-Resource Dialects\n## Specific Difficulties in Algerian Darija","[{\"question\":\"Why is hate speech detection challenging for Algerian Darija on social media?\",\"answer\":\"Darija lacks standardized orthography, users use multiple spellings for the same words, and online text often includes code-switching between Arabic and French plus Arabizi written in Latin letters and numerals.\"},{\"question\":\"Which model categories are compared in the paper?\",\"answer\":\"The study compares traditional TF–IDF machine learning, recurrent neural networks, Transformer-based language models (including DziriBERT and multilingual BERT), and the proposed hybrid FAD-SA-GRU architecture.\"},{\"question\":\"What key components make FAD-SA-GRU effective?\",\"answer\":\"FAD-SA-GRU fuses semantic representations from DZ FastText, DZ AraVec, and DziriBERT via a multi-embedding fusion strategy, then applies a self-attention-enhanced GRU encoder for sequence modeling.\"}]",1784212563,43,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"fad-sa-gru-enhancing-hate-speech-detection-in-algerian-dialect-through-feature-augmented-self-attention-gru-networks","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/fad-sa-gru-enhancing-hate-speech-detection-in-algerian-dialect-through-feature-augmented-self-attention-gru-networks/86549/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-27","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"Why is hate speech detection challenging for Algerian Darija on social media?","Question",{"text":76,"@type":77},"Darija lacks standardized orthography, users use multiple spellings for the same words, and online text often includes code-switching between Arabic and French plus Arabizi written in Latin letters and numerals.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"Which model categories are compared in the paper?",{"text":81,"@type":77},"The study compares traditional TF–IDF machine learning, recurrent neural networks, Transformer-based language models (including DziriBERT and multilingual BERT), and the proposed hybrid FAD-SA-GRU architecture.",{"name":83,"@type":74,"acceptedAnswer":84},"What key components make FAD-SA-GRU effective?",{"text":85,"@type":77},"FAD-SA-GRU fuses semantic representations from DZ FastText, DZ AraVec, and DziriBERT via a multi-embedding fusion strategy, then applies a self-attention-enhanced GRU encoder for sequence modeling.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":20,"slug":138},19,"General","general"]