Can We Detect Non-playable Characters’ Personalities Using Machine And Deep Learning Approaches?

Authors

  • Jérôme Hernandez Sorbonne Université, CNRS, LIP6
  • Mathieu Muratet Sorbonne Université, CNRS, LIP6
  • Matthis Pierotti Origamix-RH
  • Thibault Carron Sorbonne Université, CNRS, LIP6

DOI:

https://doi.org/10.34190/ecgbl.16.1.627

Keywords:

Natural Language Processing, Personality Recognition, Machine Learning, Visual Novel Game

Abstract

Personality recognition and computational psychometrics data have become prevalent in personnel selection processes. Such assessment tools are adequate for human resources seeking tools to assess a large volume of diverse player personalities in the current "war of talents." Recently, studies about using Gamified situational judgment test approaches have shown positive results in assessing players' behavior and personality.

Gamified situational judgment tests combine the advantages of gamification, such as enhancing players' reactions and flow state, with the acknowledged traditional situational judgment test approach. To gamify a situational judgment test, an innovative approach using the visual novel game genre has shown positive results in the gamification by adding game elements such as narrative scripts, non-player characters, dialogs, and audiovisual assets to the test. Indeed, these elements play an essential role in the validity of the players' personality results by using a stealth-assessment method to minimize social bias and player's stress. However, to our knowledge, as gamification in personality detection is still recent, little is known on the possible positive outcomes of designing game elements such as the dialogues and non-player character personalities in the validity of the team cohesion measure. To this end, we propose an empirical study to build personality trait models based on non-players characters' speeches. 

We used the Myers–Briggs Type Indicator based on four dichotomies to classify the personalities as one of companies and organizations' most used personality typology. For each of the four dimensions, we train twenty-four separate binary classifiers and one 16-class classifier, using well-established machine learning and a convolutional neural network in the domain of natural language processing, text analytics, and computational psychometrics. The results of this study show that it is possible to recognize non-playable characters’ personalities and thus can help game designers to understand their characters' personalities using natural language processing.

Downloads

Published

2022-09-29