Fixation and Machine Learning A new method for measuring fixation in internet users using machine learning and natural language processing

Sammanfattning: Fixation — the state of mind where an individual is intensely fixated on one individual, activity or idea —  can be an early warning sign of an individual that will later go on to commit violent acts. This paper details the development of a method that measures fixation from texts written on internet forums, a method that could (when expanded) be a part of threat assessment and prevention of violence.  Text documents gathered from Flashback Forum were pre-processed, analysed, clustered and scored using a new fixation scoring algorithm.  Three different clustering algorithms were tested — k-means, agglomerative hierarchical and DBSCAN. After evaluation with both ARI and AMI, DBSCAN was found to be the best fit.  The method was tested on a set of 100 users from Flashback Forum and gave satisfactory results.