Creating a Chinese gender lexicon for detecting gendered wording in job advertisements

Abstract

It is widely assumed that gendered wording in job advertisements can be a source of unconscious gender bias that contributes to occupational gender segregation, and gender lexicons have been developed and employed to detect gendered wording in job advertisements. The goal of this study is to create a Chinese job advertisement gender lexicon, the lack of which has impeded related research in China. Based on 53,786 job advertisements collected from a large employment website in China, the lexicon creation process enabled by supervised learning mainly involved identifying candidate gender words and determining their gender scores. The combination of Word2Vec and SVR yielded the highest performance and generated a new Chinese gender lexicon consisting of 1,429 masculine words and 1,064 feminine words with varying gender scores. The lexicon was successfully applied in gendered wording detection, revealing that masculinely and femininely worded job advertisements dominated different occupations in China. The superiority of the proposed lexicon creation method and the resultant lexicon are verified through comparisons.