A CNN-BiLSTM Hybrid Architecture with Resampling Techniques for Arabic Legal Text Classification

Résumé

Arabic legal text classification has played a major role in improving judicial systems by automating the categorization of legal texts and facilitating access to legal information. Despite these benefits, developing a model to classify Arabic legal text faces significant challenges, including the rich morphology and the inherent complexities of the Arabic language. Additionally, the imbalanced distribution within the legal specialties adds more challenges to the development of such a model. To address these challenges, this paper proposes a hybrid Deep Learning (DL) model that combines Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks, using a pre-trained Arabic Bidirectional Encoder Representations from Transformers version 2 (AraBERTv2) model as a word embedding technique. Additionally, extensive experiments were conducted to explore the impact of resampling techniques on the legal text classification model and to achieve an equal class distribution. Furthermore, a newly collected Arabic legal dataset was used to evaluate the performance of the developed model, and several evaluation metrics were employed, including accuracy, precision, recall, F1-score, and Matthews Correlation Coefficient (MCC). The findings demonstrate that our model yielded superior performance, with a score of more than 95% across all employed metrics. Moreover, the Random Oversampling (RO) technique showed the best results among other resampling techniques.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0