Career Pathway Recommendation (CPR) generalizes the idea of job recommendation. Instead of a single job only, a sequence of jobs can be recommended that identifies a career pathway either consistent with the user’s job history or targeting a specific career goal. While job recommendation strives to suggest the most appropriate job for a given candidate, the purpose of CPR is to discover a pathway that leads a candidate towards her career goal, dealing with uncertainty, e.g. about alternative career choices, the likelihood that the application to a job succeeds, and so on.
Career pathway recommenders can be applied to use cases where job recommenders would inherently fail. Let us say for example that a job is exactly what a candidate is looking for, but she does not have an adequate profile to get the job. A job recommender can choose whether to recommend the job or not. If it is able to capture that the candidate’s profile is not appropriate, it will not recommend the job, but the candidate will never know about a job she would have been very interested in. If it recommends the job, it will be leading the candidate towards a failure. Consequently, the candidate might lose confidence in herself, or even stop pursuing her career goal. On the other hand, a career pathway recommender will recommend a policy that eventually leads the user towards her goal, minimizing the intermediate career steps while simultaneously improving the user’s profile and increasing the likelihood that the user succeeds when applying for her career goal. In other words, a career pathway recommender will take the career goal into account, and will help the user finding a pathway to it rather than recommending it when the user’s application is very likely to fail.
40 CHAPTER 5. RECOMMENDER SYSTEMS FOR JOB SEARCH
Despite the fact that many approaches have been proposed for job recommendation, little effort has been made to the problem of recommending a pathway that leads users towards a career goal. Elayidom et al. [EIA11] proposed a decision tree based approach which helps students choosing a good branch that may fetch them placement in either rural or urban sectors. In other works [ASE14, RHN+14], the epithet career pathway recommendation is used with a different meaning. The authors employed it alluding to a branch of potentially interesting jobs rather than to a sequence of jobs aiming at achieving a final goal. Mimno and McCallum [MM08] analyzed the text of a corpus of resumes and their job transitions to mine latent skills that make up each job description and modeling career path trajectories. Differently from their work, where the past job history of all users is used to learn general path trajectories, the approach that will be presented in Chapter 11 provides a personalized recommendation for each user based on her profile.
On the other hand, some researchers focused on factors that can affect talent career steps. Li et al. [LGZ+17] analyzed two critical issues in talent management, namely turnover and career progression, showing that they can effectively be predicted through a survival analysis, which considers sequences of time intervals. Xu et al. [XYY+16] created a job transition network where vertices stand for organizations and a directed edge represents the talent flow between two organizations for a time period. Talent circles including the organizations with similar talent exchange patterns may be detected from such a network, and their characteristics can be used for talent recruitment and job recommendation. Xu et al. [XYX+15] tried to discover the job change motivations as well as correlations between professional and daily life. The authors used the career mobility and daily activity patterns at the individual level to assess to what extent such data can be predictors of a job change. Xu et al. [XZZ+17] introduced a data driven approach for modeling the popularity of skills based on the analysis of large-scale recruitment data. After having used a large corpus of job postings to build a job skill network, they created a topic model that integrates different criteria of jobs (e.g. salary, company size) and the latent connections within skills to effectively rank the skills based on their popularity.
The career pathway recommendation problem can be formulated as a Markov Decision Process (MDP) [Put14]. MDP-based systems have already been proved useful for product recommendation [SHB05], where the goal is recommending a product given any sequence of user purchases. In career pathway recommendation the goal is somewhat related, and the MDP allows extracting an optimal sequence of jobs (i.e. policy) considering every possible career evolution. Differently from product recommendation, where the recommendation of an item increases its probability to be bought, the recommendation of a job does not affect the probability for a user to get it.
6
Big Data Analytics
This chapter focuses on the analytic process, putting the emphasis on what is needed to properly perform an analysis. As the reader might expect, the focus is not only on the algorithms, but also on Big Data mining platforms. In particular, a description of the main characteristics of such platforms is provided, such as computational resources, storage, together with advantages and drawbacks.
The rationale is to provide readers with some background knowledge that helps understanding the possibilities that big data offer along with the challenges that programmers have to face when conducting an analysis. Specifically, the large variety of platforms available makes it hard to decide which solution fits best the current requirements. Because of this awkward choice and the continuously changing requirements, porting algorithms to different platforms is often needed. In order to reduce the development time of algorithms, which typically need to be rewritten from scratch, parallel programming primitives for platform-independent algorithm implementa- tion will be introduced in Part V.