Models for Arabic Document Quality Assessment

Digital content has been increasing rapidly. This content can be generated, accessed and used by anyone and thus the need for quality assessment of web content before usage becomes an important issue. Devising methods to assess the quality of Arabic digital content is the focus of this paper. Our work was partially based on Wikipedia articles annotated into featured and good according to quality guidelines of Wikipedia. Our analysis was directed at finding features that can serve as best quality indicators. Using the defined features, we trained a high accuracy quality assessment model using machine-learning algorithms. Our work went beyond the Wikipedia documents to build a general model that can assess the quality of Arabic documents that lack Wikipedia metadata with acceptable accuracy. The model was trained and built using features from documents we collected from Arabic online news sites and blogs, and annotated in collaboration with university students.