Deliverable D4.2 “Software libraries built on Graphos.ml using data mining for the detection of aggressive or distressed behaviors in OSN”

One of the main objectives of the ENCASE project is to develop a browser add-on and its corresponding machine learning algorithms for the detection of malicious and problematic behavior such as cyberbullying, hate speech, aggressive behavior, distressed behavior, and sexual grooming.

This document describes the efforts and algorithms developed to identify and quantify the various types of online abuse. It also explains the research conducted and the methodologies studied to detect hateful content, raid, abuse, sexual grooming and bullying.

This deliverable also provides a brief description of the completed projects carried out towards automatically detecting malicious behavior in the context of tasks T4.1 - “User profiling to detect and prevent malicious and criminal activities” and T4.2 - “Sentiment and affective analysis on individual and collective basis”, which were listed as “ongoing” and were included in deliverable D4.1 – “Development of automated techniques to detect early indications of malicious behavior of social network users”. Importantly, the work and efforts that took place during Task 4.3 - “OSN malicious users time-dependent detection” is thoroughly presented herein.

Task 4.3 aims at modeling of time-dependent interactions and activities of social network users, and the application of graph mining and text processing methodologies to detect latent patterns of activity by users/entities and their interactions. Advanced data mining and analytics techniques weredeveloped in order to leverage the OSN users’ concurrent activities that indicate behavioralvariations and spikes with emphasis on advancing the state of the art on anomaly detection in OSN.