Job matching is a process that involves decision to whether a job vacancy is relevant, given the profile of job seeker and vice versa. It requires thorough understanding of job seeker and vacancy in order to match them bidirectionally. Bidirectional matching through measuring the degree of semantic similarity of job descriptions in vacancies and candidate job seekers has been a challenging task in the job recruitment industry. The challenges are associated with i) lack of information due to job seeker inability or resistance to provide sufficient data, ii) difficulty in modeling job seekers and/or vacancies and iii) the complexity of matching process itself.
Fortunately, the Internet and advancement of information technology provide opportunities that help deal with these challenges. Availability of huge online data about job descriptions which has been entered by job seekers and job holders can be utilized to understand job seekers. Large volume of online vacancy data can be exploited to portray the current demand of the job market. Prevalence of technological advancements to handle the size of big data and the complexities of the matching process makes job matching feasible to address.
Understanding the job seeker presupposes obtaining more information about the job seeker directly from himself, i.e., through resumé and web survey, or through others, i.e., from social network. This research investigates tools and techniques, and implements a web-based user interface, i.e., a context-aware Dynamic Text Field (DTF), that allows users to enter data of their choice but with guidance using autocompletion. Moreover, this study identifies methods that measure the skills, expertise and experience of a job seeker and investigates the importance of using social networking data as input to user modeling that determines the strength of skills to be used for recommending matching job vacancies.
In addition to job seekers, job matching requires understanding and modeling of vacancies. Though online vacancies are publicly available, due to overwhelming volume of data, job seekers are not able to easily find relevant vacancy for their skill or are unable to analyze the requirements to estimate its relevance. Analyzing vacancies as well as optimizing the matching process, on one hand, and exploiting the available opportunities of big data and technological advancement, on the other hand, are of paramount importance to pursue a novel approach of job matching.
This research employs solutions that learn from data (as opposed to rules) because they perform better at handling job seekers and vacancy data in the ever changing market. One of the methods to address these challenges is applying Machine Learning – data-intensive techniques to model job seekers and
vacancies – and get a better matching. It explores matching job vacancies with job seekers using data from online vacancies, occupational standards, resumés, job seeker’s self-assessment, and social network data. Deep unsupervised feature learning, which is a kind of machine learning, is applied to develop
a novel bidirectional matching of job seeker and vacancy through modeling the former using data from self-assessment, resumé parsing and social network, and the latter using vacancy parsing and enriching it via occupational standards. The choice of the data is based on its suitability to model different aspects of job seeker and vacancy. Machine learning is chosen because of the dynamics of job market, i.e, the jobs change so frequently that developing rules is practically infeasible, whereas learning from data is feasible with the availability of large online data, and robust computational resources.
The results of this endeavor are i) development of algorithm for context-aware DTF for user profile survey; ii) a new technique to measure skill relevance using social network-enhanced job seeker modeling; iii) improved relevance ranking of job vacancies through feature enriching by job titles and descriptions from standard occupations.