[Paper Review] JobComposer: Career Path Optimization via Multicriteria Utility Learning

Richard J. Oentaryo, Xavier Jayaraj Siddarth Ashok+2 more|arXiv (Cornell University)|Sep 4, 2018

Reinforcement Learning in Robotics18 references4 citations

TL;DR

JobComposer is a data-driven framework that optimizes career paths using multicriteria utility learning to improve upon suboptimal real-world career trajectories from online professional networks. By decomposing utility optimization across salary, job level, and desirability, it generates superior, non-greedy career paths that reduce time-to-advancement by up to 80 months compared to common or actual paths.

ABSTRACT

With online professional network platforms (OPNs, e.g., LinkedIn, Xing, etc.) becoming popular on the web, people are now turning to these platforms to create and share their professional profiles, to connect with others who share similar professional aspirations and to explore new career opportunities. These platforms however do not offer a long-term roadmap to guide career progression and improve workforce employability. The career trajectories of OPN users can serve as a reference but they are not always optimal. A career plan can also be devised through consultation with career coaches, whose knowledge may however be limited to a few industries. To address the above limitations, we present a novel data-driven approach dubbed JobComposer to automate career path planning and optimization. Its key premise is that the observed career trajectories in OPNs may not necessarily be optimal, and can be improved by learning to maximize the sum of payoffs attainable by following a career path. At its heart, JobComposer features a decomposition-based multicriteria utility learning procedure to achieve the best tradeoff among different payoff criteria in career path planning. Extensive studies using a city state-based OPN dataset demonstrate that JobComposer returns career paths better than other baseline methods and the actual career paths.

Motivation & Objective

• To address the lack of long-term, data-driven career path optimization in online professional networks (OPNs), which often reflect suboptimal, non-personalized, and popularity-biased career trajectories.
• To overcome limitations of existing career recommendation systems that prioritize commonly taken paths without considering trade-offs among multiple payoff criteria such as salary, job level, and desirability.
• To develop a novel, multicriteria utility learning framework that identifies optimal career paths by maximizing a composite utility function across multiple, potentially conflicting, career goals.
• To demonstrate that observed career paths in OPNs are often suboptimal due to incomplete transitions, subjective decisions, or lack of strategic planning.
• To provide a scalable, efficient method for generating personalized, high-quality career path recommendations that outperform both actual user paths and greedy common-path heuristics.

Proposed method

• JobComposer models career progression as a Markov process over job transitions derived from a large-scale OPN dataset of 265,000+ job transitions.
• It formulates career path optimization as a multicriteria utility learning problem, jointly optimizing for salary, job level, and desirability gains along a path.
• A decomposition-based iterative algorithm splits the multicriteria problem into multiple scalar subproblems, enabling efficient simultaneous optimization across criteria.
• The method uses utility learning to estimate the trade-off between competing criteria, allowing the system to identify the best compromise path for a given user preference.
• It computes optimal career paths by composing sequences of job transitions that maximize the cumulative utility, based on observed career trajectories.
• The framework supports non-greedy, non-popular career paths that may yield better long-term outcomes than commonly taken routes.

Experimental results

Research questions

• RQ1 Can observed career paths in online professional networks be systematically improved using data-driven optimization?
• RQ2 To what extent can a multicriteria utility learning framework outperform popular or actual career paths in terms of time-to-advancement, job level, and desirability?
• RQ3 How effective is a decomposition-based multicriteria utility learning approach in balancing competing career goals such as salary, job level, and job satisfaction?
• RQ4 Are non-popular, non-greedy career paths more beneficial than commonly taken ones in terms of long-term career outcomes?
• RQ5 Can utility-based optimization generate career paths that are both sensible and significantly better than baseline methods?

Key findings

• JobComposer reduced the time-to-advancement by up to 80 months compared to the greedy most common path method in one case study, while achieving higher job level and desirability gains.
• In a qualitative analysis, JobComposer generated career paths that saved 70–80 months compared to actual career paths, with consistent improvements in job level and desirability.
• For a user in the telecommunications industry, JobComposer recommended remaining in the same industry but switching to a smaller company, saving 70 months and achieving higher desirability than the most common path.
• In the financial services sector, JobComposer’s recommendation saved 56 months and led to a higher-level position (Assistant Vice President) with greater desirability than the path suggested by the greedy most common method.
• In the semiconductor industry, JobComposer’s recommendation to switch to banking yielded a Managing Director role in 67 months less than the greedy path, with significantly higher level and desirability gains.
• The method consistently outperformed both actual career paths and the greedy most common path in all evaluated metrics, demonstrating that popular paths are not always optimal.