Sentiment analysis in Iraqi Arabic dialects based on distributed representations of sentences and machine learning approach
Thesis Type: Doctorate
Institution Of The Thesis: Gazi University, Fen Bilimleri Enstitüsü, -, Turkey
Approval Date: 2019
Thesis Language: English
Student: ANWAR ADNAN MZHER ALNAWAS
Principal Supervisor (For Co-Supervisor Theses): Nursal Arıcı
Co-Supervisor: Mehmet Hakkı Suçin
Open Archive Collection: AVESIS Open Access Collection
Abstract:Sentiment analysis is a sub-discipline of computer science involved in computational linguistics and data mining. The purpose of Sentiment analysis is the inference of individuals 'and communities' feelings and thoughts about a topic from textual documents. In the field of Sentiment analysis, which has become an interesting research topic for researchers in recent years, there are many studies on English in the scientific literature. However, not enough studies have yet been published on Arabic. Arabic; it is an important language in terms of number of speakers, history, and religious heritage. The official language in Arabic consists of classical and modern standard Arabic. Classical Arabic represents the language of the Qur'an. Modern Standard Arabic is used in newsletters and education. Although the use of Arabic on the Internet is increasing, these two types are not used in social networking environments. Local dialects used in daily practice are more preferred. Therefore Sentiment Analysis of the Arabic texts based on dialects, is an important research topic. In this doctoral dissertation, Sentiment Analysis is conducted in the Arabic-Iraqi dialect. In the first stage of the study, three types of data were collected. These are: data sets classified from previous studies, unclassified Iraqi Arabic dialect and classified Iraqi Arabic dialect. The second stage is the pre-processing stage. At this stage, unnecessary terms from the datasets have been eliminated to minimize complexity and standardize text format. In the third stage, features were extracted to represent a word as a vector using Doc2Vec model. In the fourth step, the vectors created were trained through four machine learning algorithms to create a sentiment estimation model. Lastly, the sentiment predictive model was evaluated. Moreover, at the experimental phase, the effects of variable parameters and the background corpora on classification performance was evaluated.