MORSE-QE: A Morphology-Aware, Embedding-Driven Framework with Root Extraction for Arabic Dialectal Query Expansion

Information retrieval (IR) in diglossic and morphologically complex Arabic includes major difficulties since dialectal searches usually do not retrieve documents written in Modern Standard Arabic (MSA). The following paper introduces MORSE-QE (Morphology‑aware, Optimized, Root‑driven Semantic Expansion for Query Enhancement), a four‑stage approach that combines rule‑based morph-based processing and embedding-based expansion in parallel structures. The process includes: (i) dialect‑to‑MSA normalization using curated lexicons, (ii) root extraction via AlKhalil Morpho Sys 2, (iii) semantic expansion with AraVec embeddings, and (iv) root‑driven filtering to reduce morphological noise. Experiments were conducted on the QADI dataset (2,000 dialectal queries) and the TREC 2001 Arabic Corpus (383,872 MSA documents), using Mean Average Precision (MAP), Precision@10, and Root Recall as evaluation metrics. MORSE‑QE achieves MAP gains of 15–18% over neural baselines (DANs, DMNs) on QADI and 15% over RM3 on TREC 2001, with a root recall improvement from 65% (DMNs) to 88%. Ablation studies show that dialect normalization and root‑based filtering contribute 19.5% and 28.9% relative MAP improvements, respectively. These results demonstrate that MORSE‑QE provides a scalable and interpretable solution for bridging dialectal and morphological gaps in Arabic IR.