FLAT DATA CLASSIFICATION
The method was designed to address the same problem from a slightly different perspective. The aim was to design a method that was assumption free and could scale well. For that we chose a flat representation of the data, something that can ease their parti-tioning and distributed processing. We also chose a method in which we make no assumption for the users, the videos or their comments. The only input of the system is a list of negatively and non-negatively predisposed users that are used to train our system. This enables us to use the system irrespective of language or other barriers as long as we provide the system with a good initial pool of identified users, as they were indicated by the field expert. To evaluate our system’s performance we conducted extensive expe-riments using the same dataset as in the evaluation of the above mentioned method (Section 4.1.1), and compared the results.