Mining Labor Market Requirements Using Distributional Semantic Models and Deep Learning

Abstract

This article describes a new method for analyzing labor market requirements by matching job listings from online recruitment platforms with professional standards to weigh the importance of particular professional functions and requirements and enrich the general concepts of professional standards using real labor market requirements. Our approach aims to combat the gap between professional standards and reality of fast changing requirements in developing branches of economy. First, we determine professions for each job description, using the multi-label classifier based on convolutional neural networks. Secondly, we solve the task of concept matching between job descriptions and standards for the respective professions by applying distributional semantic models. In this task, the average word2vec model achieved the best performance among other vector space models. Finally, we experiment with expanding general vocabulary of professional standards with the most frequent unigrams and bigrams occurring in matching job descriptions. Performance evaluation is carried out on a representative corpus of job listings and professional standards in the field of IT.