Enhancing semantic question similarity in Arabic using hybrid models and knowledge graph alignment
Semantic question similarity (SQS) plays a crucial role in enhancing questionanswering (QA) systems, particularly for low-resource languages, such as Arabic,which present challenges due to complex morphology, semantic ambiguity, andlimited linguistic resources. Despite prior efforts, knowledge graphs (KGs), which arestructured networks representing entities and relationships, remain underutilized inimproving semantic understanding and reducing ambiguity. Existing semanticsimilarity approaches have inherent trade-offs: KGs capture deep meaning but lackadaptability; corpus-based methods, such as word embeddings, rely on statisticalpatterns without deep semantic understanding; and deep learning models efficientlyperform but lack interpretability. In addressing these limitations, this study proposesa hybrid SQS model that integrates knowledge graph embedding (KGE) withAraBERT embeddings to enrich semantic representation. The model utilizes abidirectional long short-term memory network with an attention mechanism topredict question similarity, leveraging the strengths of each approach whilemitigating their individual shortcomings. Furthermore, we align LughaNet (a variantof Arabic WordNet) with Wikidata using the similarity flooding (SF) algorithm toenhance KG-based semantic representations, creating a comprehensive andlarge-scale Arabic knowledge graph. Experimental evaluation on an augmentedMawdoo3 dataset demonstrates that our approach achieves an F1-score of 93.88%,outperforming state-of-the-art benchmarks, including GPT-based models. Thesefindings confirm the effectiveness of hybrid models in capturing structured andcontextual semantic relationships, advancing Arabic natural language processing(NLP) research. Additionally, this study introduces a large enriched Arabicknowledge graph, providing a valuable linguistic resource that lays the foundation forfuture advancements in low-resource language processing, with direct applications inautomated QA systems, intelligent search, and recommendation engines.