الفهرس | Only 14 pages are availabe for public view |
Abstract In the era of big data and massively generated textual data, computationally intensive techniques are needed to facilitate people’s lives. Natural Language Processing (NLP) is the area of Artificial Intelligence (AI) that keeps developing efficient models to process, understand, and generate textual data. Negation and uncertainty are universal linguistic phenomena that may change the polarity and factuality of the text meaning; therefore, both affect the performance of many NLP applications, such as sentiment analysis, machine translation and biomedical information retrieval. The negation and uncertainty detection problems have been addressed in many studies, mostly focusing on the English and Spanish languages. This is due to the lack of annotated corpora in other languages, including Arabic language. This thesis focuses on detecting the negation and uncertainty in Arabic texts using modern deep-learning techniques. Firstly, an exhaustive study is carried out to discuss the ongoing research techniques addressing these crucial linguistic features. Then, a set of rules is developed to highlight the commonly used negation and uncertainty patterns, leading us to build two negation and speculation-aware corpora: ArNegSpec and NSAR. A Multi-Task Learning (MTL) negation- and uncertainty-aware learning system is proposed to detect the negation and uncertainty scopes in Arabic Language texts using the NSAR corpus, achieving Percentage Correct Scope (PCS) of values 97% and 95% for the negation and uncertainty scope detection, respectively. This proposed system is applied to the Arabic Sentiment Analysis (ASA) as a main task with the support of negation and uncertainty as auxiliary tasks. The results show an enhancement in the F1-score of 5% and 3% to the main task using negation and uncertainty auxiliary tasks, respectively. |