الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 179 |  |  |
| | | from | 179 | | |
Abstract
Sentiment analysis or opinion mining is a Natural Language Processing technique that analyses people’s opinions, sentiments, attitudes, and emotions expressed in the author’s generated text. Personal thoughts and ideas commonly target products, sports, public figures, issues, and politics. Author-composed text is spread on social media, online reviews, newspapers, and opinion surveys. The vast amount of composing text that encloses opinion stimulated the need to analyse and exploit this resource for informed decision-making. Despite the importance of sentiment analysis, it has not been sufficiently studied in the Arabic language compared to the English language. Arabic sentiment analysis has been conducted based on word features using statistical approaches, traditional machine learning algorithms, and shallow models of deep learning architectures. It has not been tackled by applying deep learning models and exploiting character features. The introduced work aims to employ deep learning models based on character-level features for performance improvement of Arabic sentiment analysis.
First, two CNN models applied for English sentiment analysis are modified and customised for Arabic sentiment analysis. Then, two variants of each model are evaluated using two settings of the filters parameter and two versions of the dataset (raw, processed). The hybrid constructed dataset includes opinions from various domains, namely book reviews, tweets, product reviews, hotel reviews, movie reviews, and restaurant reviews composed in Modern Standard Arabic and Dialectal Arabic. The CNN with the highest performance has been evaluated under the imbalance condition of the dataset. Finally, evaluation measures are computed with different settings of the filter parameter, the imbalance handling techniques, and the version of the dataset (raw, processed).