TawjihiNavigator: a novel hybrid information retrieval system for educational guidance in Morocco

Abstract

In this paper, we propose a novel hybrid method for improving Arabic educational information retrieval (IR) in Moroccan high schools. Traditional search methods often struggle with Arabic’s rich morphology and educational terminology, hindering students’ access to accurate guidance information. The proposed method TawjihiNavigator that combines vector-based semantic search with lexical matching, enhanced by advanced Arabic natural language processing (NLP) techniques. Using a comprehensive dataset collected from official Ministry of education sources. To validate the IR-Abhato system, we integrate CAMeL Tools and Farasa stemmer for Arabic preprocessing, testing multiple embedding models including Word2Vec, FastText, and AraT5. The obtained results demonstrate that our hybrid method’s superiority over standalone vector and full-text search approaches, achieving a mean reciprocal rank (MRR) of 0.7987 and mean average precision (MAP) of 0.5628. The AraT5 model achieved the highest precision@5 score of 0.4500, specially in educational query processing. These findings indicate that our model enhances Arabic educational IR accuracy, that can be improve student decision-making processes.