Online chat maintains a persistent record of chat conversations, mining these conversations can lead to beneficial information. With the increase of cybercrimes targeting minors, online chat can be used to detect and report harmful behaviour to the responsible authorities to take necessary action/s. This can make a valuable contribution to the protection of young people on social networking sites from being exploited by online predators. This paper presents the use of topic modelling techniques as means to cluster participants in conversations (depending on their involvement in the conversation) with the intention of distinguishing predators from victims or other participants. We also use t-SNE for visualising clusters