loading page

A Quantum Beta Distributed Multi-Objective Particle Swarm Optimization Algorithm for Twitter Fake Accounts Detection
  • +6
  • Ahlem Aboud ,
  • Nizar Rokbani ,
  • Seyedali Mirjalili ,
  • Abdulrahman M. Qahtani ,
  • Fahd S. Alharithi ,
  • Omar Almutiry ,
  • Amir Hussain ,
  • Adel M. Alimi ,
  • Habib Chabchoub
Ahlem Aboud
University of Sousse, University of Sousse

Corresponding Author:[email protected]

Author Profile
Nizar Rokbani
Author Profile
Seyedali Mirjalili
Author Profile
Abdulrahman M. Qahtani
Author Profile
Fahd S. Alharithi
Author Profile
Omar Almutiry
Author Profile
Amir Hussain
Author Profile
Adel M. Alimi
Author Profile
Habib Chabchoub
Author Profile

Abstract

Fake account detection is a topical issue when many Online Social Networks (OSNs) encounter problems caused by a growing number of unethical online social activities. This study presents a new Quantum Beta-Distributed Multi-Objective Particle Swarm Optimization (QBD-MOPSO) system to detect fake accounts on Twitter. The proposed system aims to minimize two objective functions simultaneously: specifically features dimensionality and classification error rate. The QBD-MOPSO has two optimization profiles: the first uses a quantum behaved equation for improving the exploratory behaviour of PSO, while the second uses a beta function to enhance PSO’s exploitation. Six variants of the QBD-MOPSO approach are proposed to account for various data distribution types. The QBD-MOPSO system provides a feature selection technique based on the sigmoid function for position binary encoding. Each particle has a binary vector as a potential solution for feature subset selection, and a bit with the value of “1” indicating selection of a feature and “0” otherwise. Machine learning based classification models are trained and tested using a subset of selected features. An extensive experimental study is carried using two benchmark Twitter datasets with 1982 and 928 accounts. From 46 original features, QBD-MOPSO has selected 32 and 25 pertinent features and accurately classified 99.19% and 97.52% account on the datasets.