An enhanced approach for arabic sentiment analysis using deep learning / Rania Abdelmonam Kora ; Supervised Ammar Mohammed Ammar

Thesis (M.Sc.) - Cairo University - Faculty of Graduate Studies for Statistical Research - Department of Computer and Information Sciences

Valuable Information has a great impact on decision making and problem-solving processes. Information can come from anywhere, like books, articles, reference books, web sites, expert opinions, personal experiences, and so on. Also, of course social media are considered excellent sources of information which can provide opinions, thoughts, and insights toward many important topics. Due to its importance in decisions making based on opinions derived from analyzing the user's contents on social media, Sentiment Analysis becomes a vital topic in research.The Arabic language is one of the widely spoken languages used for sharing content on the social media. However, the sentiment analysis for Arabic contents is limited. There are several challenges facing the sentiment analysis for Arabic contents including the morphological structures of the language, the varieties of dialects and the lack of the appropriate corpora. From the above discussion, it can be noted that the increase of researches in Arabic Sentiment analysis is grown slowly unlike other languages such as English.This thesis has twofold contributions. First, it introduces a new corpus of forty thousand labeled Arabic tweets covering several topics.Then, it presents three Deep Learning models namely CNN, LSTM and RCNN for Arabic sentiment analysis. Based on the aid of a word embedding technique, the performance of the three models on the proposed Corpus is being validated.The experimental results showed that LSTM which has an average accuracy of 81.31% outperforms CNN and RCNN. Also, applying data augmentation on the corpus increases LSTM accuracy by 8.3%

Issued also as CD