Ryan Huynh¹, Lee Gillam² and Alison Callwood¹, ¹University of Surrey, United Kingdom, ²Sammi-Select Ltd, United Kingdom

Multiple mini-interviews (MMIs) are a widely used and validated interview method for eliciting soft skills. By using multiple, separate, and timed interviews in which each has a distinct scenario, MMIs purportedly reduce possibilities such as a biased individual dictating results, although potentially inconsistent scoring by interviewers may still impact on fairness. However, MMIs overall can be seen as challenging to run due to the number of interviewers and assessments required. In this paper, we discuss the progress in automatically, and consistently, extracting soft skills from transcriptions of MMI responses to support such assessment. While previous research has focused on extracting soft skills from job postings and written responses, to the best of our knowledge there is no other published research on soft skill extraction from MMI responses. We begin by annotating collected MMI responses to assure presence of soft skills, then evaluate the effectiveness of combining word embeddings with classifiers to identify soft skill indicators. The most promising result, F1-Score = 0.79, compares favourably to previous literature on extracting soft skills from other datasets and is encouraging of further exploration.

Soft Skills Extraction, Multiple Mini Interviews, MMI, Word2Vec, BERT