Detecting Liars in Chats using Keystroke Dynamics

Abstract

In this paper we will investigate the possibilities for detecting liars in chat rooms who have taken on a different identity. While using a different identity people might require more time to reply to questions of the chat partner, or might use corrections to change their text to avoid inconsistencies in their answers. These issues will cause differences in the typing behavior, which can be measured in the typing rhythm. We have shown in this paper that, with a high accuracy, we can distinguish between a chat of a person who uses his/her own identity and is honest in his/her answers, and a chat of a person who is lying because his/her answers need to be consistent to an assumed identity. We obtained a correct classification of a single message in a chat with an accuracy of more than 70% and a correct classification of a full chat with well over 90% accuracy.