CSC 496
King Saud University
College of Computer & Information Sciences
Department of Computer Science
Graduation Project Proposal
For Academic Year/Semester: 1435/1436 First Semester
A. GENERAL INFORMATION
A1. Supervisor(s): Mohamed Maher Ben Ismail, PhD
A2. Project Title: Verbal Offense Detection in Social Networks Commentsusing Global Fusion Approach
A3. Number of Students: 3
B. PROJECT ABSTRACT
The widespread of smart devices yields an exponential growth of social networks. These networks are hosted and managed by very big companies which are employing thousands of people, and investing millions of dollars in order to improve their social network services, features, and performance. Also, millions of users gathered within this virtual society, and formed several communities sharing the same cultural background or interest or skills. For hundreds of thousands of users, these virtual communities replaced their real social activity, and consider them as a major part of their life. Thus, they try to be connected most of the time in order to follow every event happening in their communities/groups. After almost one decade of extensive use of social networks, several challenges emerged as a result of their fast growth, and popular usage. Privacy, spam filtering, and security deficiency are some of the main concerns that social network industry is facing. However, these challenges are giving headache to professional rather than “simple” users. In other words, “simple” users are not sensitive to that kind of problems. On the other hand, the issue that everyone is facing nowadays when surfing within virtual communities is verbal offenses. Moreover, users have to deal manually with these troubles. For instance, administrators of Facebook pages have to read all comments on every single post and wipe out all insults and moral offenses.
This manual deletion is subjective and labor intensive since comments number can be very large. Moreover, comments can be very frequent, which makes the process more tedious. To address this issue, few algorithms that can automatically detect insults in social network comments have been proposed recently. The earliest efforts in this area were directed towards matching comments with a vocabulary of “prohibited” words. In other words, if the comment contains one or more keywords from the banned word list, then the comment is denied. These efforts posed the problem of insult detection as a string matching problem. Learning comments semantics and insult detection can be posed as text mining and supervised learning problem.
In this project we propose to solve the challenge of automatic verbal offense detection in social network comments. The proposed system relies on four main components: (i) text pre-processing and representation, (ii) a supervised learning model (classifier) which classifies comments from social networks as insult or not, (iii) a global fusion approach of the classifier outputs, and (iv) a prototype graphical user interface which illustrates the learning process.
C. REQUIREMENTS (both hardware and software)
* PC or laptops.
* Java IDE
D. PROJECT PHASES
Phase 1: Literature review and system design
Phase 2: Implementation of the text preprocessing and representation technique.
Phase 3: Experimental Investigation of two or three classifiers.
Phase 4: Experimental Investigation of global fusion techniques.
Phase 5: Implement the proposed algorithm/system
Phase 6: Assess the performance of the proposed system.
Phase 7: Document the results.
E. SCHEDULING OF PHASES
Semester 1: phases 1.
Semester 2: phases 2, 3, 4, 5, 6 and 7.