KUNUZ: A Multi-Purpose Reusable Test Collection for Classical Arabic Document Engineering
Ibrahim Bounhas, Souheila Ben Guirat · 2019
Corpora are important resources for several applications in Information Retrieval (IR) and Knowledge Extraction (KE). Arabic is a low resourced language characterized by its complex morphology. Furthermore, most existent Arabic language resources focus on Modern Standard Arabic (MSA). This paper describes KUNUZ a multi-purpose test collection composed of voweled and structured classical Arabic documents. Its goal is to provide a unique benchmark for assessing applications in several areas of document engineering including IR, document classification and information extraction. The documents are also translated in English to allow Arabic-English cross-lingual IR and machine translation. As far as IR is concerned, we follow the standard topic development and results sampling used in international campaigns. The paper, describes the process of topic development, results pooling and relevance judgment. It also analyses the results of some processing tools and IR models used in the runs. In order to enhance the results of our experiments, we also proposed to combine the results based on a meta-search approach using Support Vector Machines (SVM) classification.