Scaling Personality Traits of Interviewees in an Online Job Interview by Vocal Spectrum and Facial Cue Analysis

Abstract

In present world unlike in the past era, interview patterns have changed with time and innovation. An appropriating employee should be recruited for the job position which is a vital act, considering balanced personality he or she acquire. An interview is defined as a method of obtaining information of a person through oral responses to the oral inquiries raised. Different types of job interviews prevail. The interview is used as a mechanism to understand the aspects of the interviewee that cannot be measured by the application or other written tests. It’s used by the interviewers in judging the personality, appearance, communication skills, motivation and attitude of the candidate. As the main motive of our study is to improve the accuracy of hiring decisions and there were positive outcomes achieved from similar studies, we were motivated to extract vocal and facial information from an interview. In the aspect of organizational psychology, we have investigated on the relationship between the suitability of an applicant for a certain job position with his or her vocal spectrum and the variations in facial expressions during the interview. So we have based our research on Big Five personality model, where chosen four facets – Passion, Confidence, Cooperation and Emotional Stability were tested among job positions that require high communication skills, such as IT Business Analysts. By following qualitative and quantitative data gathering approaches, we were able to extract certain low level non-verbal cues and determined correlation among them. Later on through using popular classifiers such as LibSVM, BayesNet and other few popular techniques, we were able to classify the face-to-face interview recordings through feature correlation. Thereafter these clips were trained and used to test the online interview dataset. Due to several constraints found in VoIP communication, outcomes for the traits didn’t have much relationship. Still through the correlation identified among gender and traits, we were able to provide recommendations of classifiers to identify certain traits. Furthermore we figured out that, in supervised learning studies, less classification classes means higher the accuracy. Finally, our research outcome would be attractive and interesting for the researchers from psychology and sociology fields, as it provides insights on what are the non-verbal cues that might have influenced an interviewer to give a high score to a candidate.