On Utility of Temporal Embeddings in Skill Matching

Abstract

Candidates routinely use a set of key phrases or keywords to succinctly describe their expertise or skillset. This is useful for both matching candidate profiles to jobs and for comparing different candidates. Constant development of businesses and labour market has dynamic impact on importance of such skills, where importance of each skill may evolve with time. At any given time, some skills may be more important than others due to seasonality in job markets. While, existing approaches consider lexical or semantic match between candidate profile and each skill, they do not consider the time biased importance of the skill for ranking. Word embeddings have emerged as an effective tool to represent vocabulary in lower dimensional space. In this work, we exploit word embedding models that also encodes time information or seasonality of key phrases. In this work, we explore utility of these time biased skill embeddings in ranking both skills and candidates. Our experiments indicate that incorporation of skill trends improves candidate-skill matching performance.