Predicting Personality Using Answers to Open-Ended Interview Questions

Abstract

One’s personality is widely accepted as an indicator of job performance, job satisfaction and tenure intention. The ability to measure an applicant’s personality in the selection process helps recruiters, hiring managers and the applicant make better hiring decisions. Our work shows that textual content of answers to standard interview questions related to past behaviour and situational judgement can be used to reliably infer personality traits. We used data from over 46,000 job applicants who completed an online chat interview that also included a personality questionnaire based on the six-factor HEXACO personality model to self-rate their personality. Using natural language processing (NLP) and machine learning methods we built a regression model to infer HEXACO trait values from textual content. We compared the performance of five different text representation methods and found that term frequency-inverse document frequency (TF-IDF) with Latent Dirichlet Allocation (LDA) topics performed the best with an average correlation of r = 0.39. As a comparison, a large study of Facebook messages based inference of Big 5 personality found an average correlation of r = 0.35 and IBM’s Personality Insights service built using twitter text data reports an average correlation of r = 0.31. We further validated our model with a group of 117 volunteers who used an agreement scale of yes/no/maybe to rate the individual trait descriptors generated based on the model outcomes. On average, 87.83% of the participants agreed with the personality description given for each of the six traits. The ability of algorithms to objectively infer a candidate’s personality using only the textual content of interview answers presents significant opportunities to remove the subjective biases involved in human interviewer judgement of candidate personality.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 73 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.