Towards Dialogue Models for Online Grooming Detection

Abstract: Social networks are a popular way for people to engage in activities and find new friends over the Internet. One central feature that these services share is the possibility for users to interact and share personal information. However, social networks also enable sexual predators with an interest in children to have access to rich hunting grounds with a reduced risk of detection. This thesis applies grounded theory to provide domain knowledge from eight experts on online child sexual abuse. This domain knowledge is used to create informational structures that can be applied in intelligent software systems to potentially detect these types of crimes. The results provide structures for automated dialogue detection including ontological structures, models of influential circumstances (social and psychological) and a normative dialogue structure of a nested deliberation dialogue. Implementation of intelligent software systems in this domain has the potential to detect probable outcomes of dialogues much faster than human analysts. Future work is needed to provide further theoretical sampling to extend the proposed models and to evaluate the models on real-world data sets.