Team Wa’ed Al-Shrida at the Mowjaz Multi-Topic Labelling Task

摘要

This paper describes an attempt in the "Mowjaz Multi-topic Labelling Task", the competition is about classifying the Arabic articles to its topic by using artificial intelligence and neural networks, the programming language that was used to classify the datasets is "Python". The attempt was started by uploading the datasets from the "Github website", the datasets that were used in the system include three groups, train, validation, and test datasets. The "Pyarabic" and simple-transformers" libraries were used to allow the system to manipulate Arabic letters and simplify the usage of Transformer models without having to compromise on utility, respectively. The model’s type that I used is "Bert" and its name is "Asafaya/Bert-base-Arabic". The accuracy of the result that was gotten is as follows: F1 macro: 0.864, F1 micro: 0.869, competition website on Codalab: 0.8430.