Dialogue-based Deception Detection

Previous research has shown that humans are not better than random
guessing at detecting lies. However, recent research has shown, that
automatic methods can increase deception detection accuracies. This is
partly due to machine learning techniques and increasing amounts of
behavioral data availability. Multimodal deception detection is an
approach that combines various data sources, such as language, speech
or visual cues (facial expression, gestures etc.) in an attempt to
merge the individual advances in deception detection in the
corresponding fields. This approach was able to show, that with the
joint use of multiple modalities, automatic deception detection
accuracies can be improved. Everyday deception often takes place
within dialogues, but this setting stills remains largely unexplored.
In this talk, I will describe my current research on expanding
previous multimodal deception detection techniques onto a dialogue
setting. This project attempts to incorporate the dimension of the
dialogue interaction into the current multimodal deception detection
techniques. While the project is ongoing, the presentation will cover
methodological aspects related to data collection, annotation, and
preprocessing as well as a first glimpse into classification
performances on deceit detection in dialog settings.

Felix Soldner is a Master Student at the University of Amsterdam in
the "Brain and Cognitive Science" program. He is currently doing an
internship at the University of Michigan with the Language and
Information Technologies group. His background lies in the
intersection of psychology and biomimetics. His research interests are
in deception detection and crime science. He has interned previously
at the "lie-lab" in the psychology department in Amsterdam. During
that time, he worked on a project that examined automated methods of
detecting fake online reviews. He compared different classification
methods and the impact of data collecting procedures on their
performances. Felix enjoys working on interdisciplinary topics,
especially examining psychological phenomena with methods from
computer science.