A Latent Dirichlet Allocation Framework to Analyse and Forecast Employability Skills

Abstract

Recently globalization, technological advancement, reorganization of many job positions, have substantially influenced and profoundly transformed the nature and attributes of the labour market in both industrialized and emerging nations. Data pertaining to the online labour market has become a valuable resource for comprehending the dynamics and trends of the job market. Policymakers in Albania must benefit from machine learning algorithms to analyse and forecast skill needs. This paper presents a Latent Dirichlet Allocation (LDA) framework, as a probabilistic topic-modelling technique employed to find latent topics within a dataset of job vacancies. The analysis employs a two-step LDA model. The initial step is the Latent Dirichlet Allocation Model, which detects latent semantic components. The second step of the model maps semantic components calculated by the LDA algorithm to map job vacancies from the topics identified from LDA model into each category of skills. A Shiny app has been developed to forecast the top 10 skills and interests needed, categorized by skill type. The methodology and findings of this study, of this paper are expected to support Albania’s public employment institutions to perform a screening of the current soft skills and interests and predict future skills required by companies operating in Albania.