الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 112 |  |  |
| | | from | 112 | | |
Abstract
With the emergence of the internet and its platforms, users tend to express their opinions on various blogs and social media platforms. Analyzing these opinions and understanding them are crucial for businesses and their owners. Sentiment Analysis (SA) is concerned with this type of information. It understands human opinions and analyzes them to obtain the required knowledge. SA can be performed at the document, sentence, or aspect levels. The majority of research on the Aspect-Based Sentiment Analysis (ABSA) tends to split this task into two subtasks: one for extracting aspects, Aspect Term Extraction (ATE), and another for identifying sentiments toward particular aspects, Aspect Sentiment Classification (ASC). Although these subtasks are closely related, they are performed independently; while performing the Aspect Sentiment Classification task, it is assumed that the aspect terms are pre-identified, which ignores the practical interaction required to properly perform the ABSA. This study addresses these limitations and presents a perspective on ABSA that can enhance the overall representation and performance of the task. Instead of treating ATE and ASC as separate and independent tasks, our proposed approach aims to integrate them into a an End-to-End (E2E) model. By considering the interplay between aspect extraction and sentiment classification, we can leverage the dependencies between these subtasks to improve the accuracy and coherence of the overall ABSA process. Several experiments were conducted to evaluate the proposed E2E model. We employed the Arabic version of the Bidirectional Encoder Representations from Transformers (AraBERT) model in two ways: as a feature-based model and as a fine-tuned model. Additionally, we evaluated the influence of using both a Fully Connected Neural Network (FCNN) with a softmax activation function and Conditional Random Fields (CRF) as decoding layers, aiming to understand their impact on the classification outcomes. Moreover, we presented the outcomes of our proposed E2E model, utilizing three model variants: Unified, Joint, and Pipeline. The evaluation of the proposed E2E model was performed on the SemEval-2016 Arabic hotel reviews dataset using k-fold cross-validation. The experimental results demonstrated the efficiency of the proposed unified fine-tuned AraBERT-CRF model with 15-fold cross validation, achieving an overall F1 score of 95.11%.The model’s predictions are then subjected to additional processing for comparisons with the previous studies, and the results indicated the superiority of the proposed unified model, achieving an F1 score of 97.78% for the ATE task and an accuracy of 98.34% for the ASC task, outperforming previous studies.