Image Analysis of Cyberbullying Using Machine Learning Techniques

Open Access
Author:
Li, Hao
Graduate Program:
Electrical Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
June 18, 2015
Committee Members:
  • Anna Cinzia Squicciarini, Thesis Advisor
Keywords:
  • Cyberbullying
  • Instagram
  • Machine Learning
  • Image
Abstract:
This work concentrates on predicting bullying attacks on social media site images as advice for users. Using achine learning algorithms, we train and test models on data crawled from Instagram labeled by Internet crowd on Amazon Mechanical Turk. Methods and techniques involved in this project are introduced and explained. We used several approaches that are popular for pictorial semantic information retrieval to extract features, which include color histogram, edge direction coherent vector, faces and scale-invariant feature transform. Also we involved other features of caption and user information to improve the prediction accuracy. The classification algorithms involved are k-NN, support vector machine and binary decision tree. To enhance the performance, we tried pruning the data set and feature selection. The best result we achieved is 65.54% (Bully: 75.34%, Non-bully: 55.36%). We also ttempted community structure detection, a method widely used in network analysis, to perform prediction. The results are also exhibited and explained.