A New Model for the Arabic Language Based on the Semantic Representation of Words and Sentences

We addressed, in this work, the problem of developing a new computerized model for the Arabic language. Language modeling is crucial in modern NLP applications as natural language modeling could help a machine to grasp a high understanding of human language. The proposed model is based on the contextual and semantic representation of words in this language. It is a new and robust version of Word2Vec that aims to feed NLP applications with rich word representation, which is done using three main additions to the basic Wrod2Vec model: it is built on a labeled dataset instead of using unlabeled text data, named entities are detected and then represented as independent entities from their constituents, and the context window is defined to include only the other words in the same sentence that are linked with the target word by dependency relations regardless their position in the sentence. The proposed model was tested and evaluated by assessing the similarity of semantics between the words using a dataset collected and annotated by a homemade system developed for this purpose.