أزل العصواد
| آخرین بروزرسانی: 1405/6/17 |
|
|
|
Stemming is an important task in Natural Language Processing (NLP), particularly for information retrieval, text mining, and text classification. Arabic stemming is challenging due to the rich morphology, orthographic variation, and complex structure of Arabic words. This thesis proposes a deep learning-based approach for Arabic word stemming using a character-level Sequence ۲ Sequence model. The proposed model, called Char Stemmer, employs a Bidirectional Long Short-Term Memory (Bi-LSTM) encoder and an LSTM decoder to generate Arabic stems character by character.
The model was evaluated using the Noor-Sharaye benchmark dataset, which contains ۲۰۵,۰۰۰ Arabic words collected from the Holy Quran and classical Arabic texts. The dataset was normalized and split into ۹۰% for training and ۱۰% for testing. Experimental results show that the proposed Char Stemmer achieved an accuracy of ۹۳.۸۸%, demonstrating its ability to handle the morphological complexity of Classical Arabic.
In addition, the thesis investigates the impact of stemming on a downstream task, namely Arabic morphological word classification. Words were classified into three categories: noun, verb, and particle. The stems generated by Char Stemmer were used as input to the AraBERT model, and the results were compared with a Convolution Neural Network (CNN) baseline. The proposed Char Stemmer + AraBERT approach achieved ۹۹% accuracy, with macro-averaged precision, recall, and F۱-score of ۰.۹۱, ۰.۸۹, and ۰.۹۰, respectively.
The model was evaluated using the Noor-Sharaye benchmark dataset, which contains ۲۰۵,۰۰۰ Arabic words collected from the Holy Quran and classical Arabic texts. The dataset was normalized and split into ۹۰% for training and ۱۰% for testing. Experimental results show that the proposed Char Stemmer achieved an accuracy of ۹۳.۸۸%, demonstrating its ability to handle the morphological complexity of Classical Arabic.
In addition, the thesis investigates the impact of stemming on a downstream task, namely Arabic morphological word classification. Words were classified into three categories: noun, verb, and particle. The stems generated by Char Stemmer were used as input to the AraBERT model, and the results were compared with a Convolution Neural Network (CNN) baseline. The proposed Char Stemmer + AraBERT approach achieved ۹۹% accuracy, with macro-averaged precision, recall, and F۱-score of ۰.۹۱, ۰.۸۹, and ۰.۹۰, respectively.
|
|
|
|
دفعات مشاهده: 256 بار |
دفعات چاپ: 26 بار |
دفعات ارسال به دیگران: 0 بار |
0 نظر