الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 85 |  |  |
| | | from | 85 | | |
Abstract
Social networking sites (SNSs) have gained high popularity among individuals in the past few years. It enables them to share their thoughts, feelings and daily activities, as they contain a huge amount of information about its users. This source of data cannot be underestimated. from the psychologist’s point of view, words of individuals reflect who they are. This information can be used in predicting gender, age [1], personality traits [2–5], first impressions [6] and interests [7]. In this thesis, I am interested in predicting personality traits.
I can define personality as the set of habitual behaviors, emotional patterns and cognitions that evolve from environmental factors and biological, which can be used to characterize a unique individual. Many studies have been conducted by psychologists to reach a method to define human’s behavior. As a result, they reached a collection of human features that identifies it, which are then called personality traits. Some researchers in psychology divided these features into four groups: sensation (S), intuition (N), feeling (F) and thinking (T) called Myers-Briggs Type Indicator (MBTI) [8]. Others divided them to five groups: openness, conscientiousness, extraversion, agreeableness and neuroticism called The Big Five (BF) or the Five Factor Model (FFM) [9], which is the most widely used nowadays.
The prediction of personality traits, which is carried out using social media websites as a source of data can be very accurate, effective and
Chapter 1 Introduction
3
useful for many business models like virtual assistants, healthcare such as mood detection, detection of personality disorder, social network analysis, and customer profiling. Therefore, the computer science researchers aim to reach an artificial intelligent system that can make a use of the data extracted from the social media users to predict their personality traits in different languages as English [1, 10], Spanish [1, 10], Italian [1, 10], Dutch [1, 10], Chinese [11, 12], Filipino [13] and Portuguese [14].
However, there is no Arabic personality traits prediction intelligent model has been published yet, although Arabic is one of the six official languages of the United Nations Organization, which is the native language for 467 million of the world’s population. However, less attention is paid to it in the field of scientific research, as it is classified within the most difficult languages in dealing with in scientific researches. In this thesis, a new-labeled dataset for Egyptian dialect twitter users is provided. I discuss the method of gathering, annotating, properties, and statistics of the dataset. I present a set of benchmark experiments.
1.1 Motivation
Nowadays, personalization of intelligent systems gains the interest of investors in many fields. As they add value in many business models, such as virtual assistants, healthcare like mood detection, detection of personality disorder, inter-personal relations, job satisfaction,
Chapter 1 Introduction
4
professional and romantic relationship success. Predict personality for social media users can be very useful for many applications like recommender systems, personalized advertising and on-line marketing. Marketers believe in the importance of profiling the interests of those they consider as their targeted customers to achieve the best prospects for purchasing the product/service. It can be applicable by predicting the customer’s personality.
1.2 Problem Definition
Despite of the importance and the massive contribution of the Arabic users in different social media platforms as they are estimated to reach 11.1 million in March 2017, almost doubling up from 5.8 million three years ago [15]. I cannot find any research on extracting the personal characteristics of the Arab social media users or a well-trained dataset that can help researchers to carry on such research.
1.3 Research Objectives
As Arabic language suffers from lack of interest by researchers. The research objectives are:
Create the first labeled personality traits prediction dataset for Egyptian dialect twitter users.
Provide Arabic users with means to predict their personality traits through their twitter feed.
Help building Artificial Intelligent (AI) system for personality traits detection.
Chapter 1 Introduction
5
1.4 Thesis Contributions
The contributions as follows:
The first labeled dataset for Egyptian dialect twitter users was provided.
A baseline model for binary personality traits classification problem with average accuracy 66.4% according to f1-score.
A baseline model for multiclass personality traits classification problem with average accuracy 35.8% according to f1-score.
1.5 Thesis Organization
This thesis is organized as follows: Chapter 2 discusses some topics that are relevant to this thesis subject such as personality traits, social media and machine learning. In addition, Chapter 2 is reviewing similar studies in the same topic but using different languages. Chapter 3 presents proposed architecture, which is used to achieve the model. Chapter 4 presents implementations and the results for different experiments to reach the best result in each trait. Chapter 5 includes conclusions of the thesis and the suggestions for future works.