MERAA : Matryoshka Embedding Retrieval Assessment for Arabic

Evaluate your Embedding model or any Arabic Sentence Transformer model's performance on context and question retrieval for Arabic datasets for Enhancing RAG (Retrieval-Augmented Generation).

• ARCD evaluates short context retrieval performance.
• MLQA Arabic evaluates long context retrieval performance.
• Arabic Financial Dataset focuses on financial context retrieval.

Evaluation Metrics: The evaluation uses NDCG@10 and MRR@10, which measure how well the retrieved documents (contexts) match the query relevance. Higher scores indicate better performance. Embedding dimensions are reduced from 768 to 64, evaluating how well the model performs with fewer dimensions.