A Machine Learning Collaborative Study of Language-Action Cues for Spontaneous Deceptive Communication and Cyber-Ontology Development

In this study, we will examine the use of deceptive language, and explore language usage patterns during online deceptive acts. As interpersonal communication is defined as a dynamic exchange of messages between or among two or more people, our research will focus on social interactions where participants try to mutually influence each other in a dynamic fashion. We thus seek an answer to the following research question: What linguistic cues can be attributed to deception in a computer mediated communication across a pluralistic background of users?

Methodology: Online Game

For this research, we plan to focus on developing specific metrics for language usage patterns and information behaviors. We will set up online game scenarios to analyze communication dynamics that distinguish between different types of communication typologies for further investigation into the underlying patterns and their relationships to words used. To mimic various deceptive scenarios, an interactive online game has been designed for collecting data on the dynamics between two actors as they are presented with choices to deceive – or to detect deception.

Data on participants’ truthful (as baseline) and deceptive statements, as well as their interpretations will be collected and stored in a database. New machine learning techniques will be applied to proceed from previous experiments that proved feasibility, to state of the art tools. The promising results were obtained using Linguistic Inquiry and Word Count analysis; and regression analysis to estimate the relationship between predictor variables. The next stages include text mining, and new cyber-ontology development.

Results of Prior Study

The best result for the dataset of our prior study was R²=0.967, p=0.03. This study suggests promising results for modeling deceptive language cues. Games like this can be further employed to collect a large, synchronous dataset. Future research includes improving the design of the game to minimize the effects of dynamic factors, validating deceptive and trustworthy cues, and broadening the framework to a solid cyber-ontology of deceptive language.