وسام المعموري

The study presents a model that can be employed to enhance the named entity recognition (NER) of Arabic text with specific reference to hadiths of Islam. Using sequence-to-sequence models can better an‎d more precise text analysis to improve entity recognition. With the help of Noor Hadith dataset that consists of 59,430 hadiths a BIO tagging scheme was created, cleaning of text, removal of stop words an‎d breaking up of text occurred. The distribution was shown to be right skewed by the analysis of the text length where short texts were more frequent. The classification of words in the dataset was made to eight categories of entities namely, person, imam, location, narrator, book, tribe, date, an‎d event with a choice of four models, namely, AraBERT, BiLSTM, CNN-BiLSTM Hybrid an‎d AraBERT-LSTM Hybrid. Precision, recall an‎d F1 score of four models AraBERT, LSTM, hybrid CNN-BiLSTM, an‎d hybrid AraBERT-LSTM model were eva‎luated using performance which gave the latter hybrid model an accuracy of 0.981. The hybrid NER model that has been created during the study is also very promising to the natural language processing in Arabic. The application of different modeling strategies to our hybrid model gives us more accuracy an‎d consistency, which lead to the solution of challenging tasks more successfully. The experiment with the hybrid model on the Arabic Islamic texts of Hadith has been fruitful an‎d therefore paves way to some exciting applications in the future in the natural language processing an‎d related studies. The research constitutes a useful resource of other efforts to enhance such models an‎d render it to other collections of Arabic texts.

تشخيص موجوديت‌هاي اسمي عربي (NER) , تحليل متن حديث , مدل‌هاي توالي به توالي , مدل‌هاي عصبي تركيبي , طرح برچسب‌گذاري زيستي

Arabic Named Entity Recognition (NER) , Hadith Text Analysis , Sequence to Sequence Models , Hybrid Neural Models , BIO Tagging Scheme

Wessam Almamoori

Dr.Minaei- Dr Etemadi