Detecting deception in conversations
Ioana-Raluca Zaman, Ștefan Trăușan-Matu · 2021
Interaction with other people is part of anyone's life, and implicitly comes with the risk of deception.On average, a person is lied up to several dozen times a day, the main purpose being manipulation.Deception is defined as a psychological act of misleading or trying to persuade someone to believe certain facts.One manner to understand deception is via emotions and sentiments of the participants in conversation.This paper's aim is to combine the information extracted from the text with the help of techniques specific to Natural Language Processing with notions of psychology.Even if lies are not a mandatory element, in most cases they play a crucial role in manipulation.Therefore, this project classifies lies based on the emotions extracted from a message, but also from the replies to it and using two alternatives of the Naïve Bayes algorithm.Moreover, the project will focus both on actual lies and suspected lies.The corpus utilized for analyze is Deception in Diplomacy Dataset from Cornell University ConvoKit.