Modelling and Predicting Individual Salaries in United Kingdom with Graph Convolutional Network

Chen, L., Sun, Y. and Thakuriah, P. (2019) Modelling and Predicting Individual Salaries in United Kingdom with Graph Convolutional Network. In: Madureira, A. M., Abraham, A., Gandhi, N. and Varela, M. L. (eds.) Hybrid Intelligent Systems. Series: Advances in intelligent systems and computing (923). Springer: Cham, pp. 61-74. ISBN 9783030143466 (doi:10.1007/978-3-030-14347-3_7)

Full text not currently available from Enlighten.


Job Posting Sites, such as Indeed and Monster, are specifically designed to help users obtain information from the market. However, at the moment, only approximately half of the UK job postings have a salary publicly displayed. Therefore, the aim of this research is to model and predict the salary of a new job, so as to improve the performance of job search and help a vast amount of job seekers better understand the market worth of their desirable positions. In order to effectively estimate the salary of a given job, we construct a graph database based on job profiles of each posting and build a predictive model through machine learning based on both metadata features and relational features. Our results reveal that these two types of features are conditionally independent and each of them is sufficient for prediction. Therefore they can be exploited as two views in graph convolutional network (GCN), a semi-supervised learning framework, to make use of a large amount of unlabelled data, in addition to the set of labelled ones, for enhanced salary classification. The preliminary experimental results show that GCN outperforms the existing ones that simply pool these two types of features together.

Item Type:Book Sections
Glasgow Author(s) Enlighten ID:Chen, Dr Long and Sun, Mr Yeran
Authors: Chen, L., Sun, Y., and Thakuriah, P.
College/School:College of Social Sciences > School of Social and Political Sciences
Published Online:21 March 2019

University Staff: Request a correction | Enlighten Editors: Update this record