Skip to main content Skip to secondary navigation
Article Newspaper/Magazine

Exploring Linguistic Features for Deception Detection in Unstructured Text


Detecting deception in unstructured texts has many applications but faces serious obstacles. Recent research has shown that linguistic traces of deception can be extracted from texts, such as political speech, and a number of theoretically derived features have been linked to deception. In this paper, we explore methods to extract and select computationally- derived linguistic features in order to improve the performance of deception detection in political speech. We extract semantic features from different language tools and use feature selection techniques to select the optimal feature set. The selected features include both theoretically expected features (e.g., negative emotion tone) and empirically-derived features (e.g., narrative and cohesion). Using the selected features can significantly improve deception detection performance compared with a theoretical approach that uses a limited set of features.

X. Liu
J.T. Hancock
G. Zhang
R. Xu
N. Markowitz & Bazarova
Proceedings, Forty-Fifth Hawaii International Conference on System Sciences (HICSS45)
Publication Date
January 7, 2012