CAPER: Enhancing Career Trajectory Prediction using Temporal Knowledge Graph and Ternary Relationship

TL;DR

CAPER introduces a temporal knowledge graph-based approach to career trajectory prediction that jointly models the mutual ternary dependency among user, position, and company and captures time-evolving characteristics of users and companies. It reframes CTP as extrapolated reasoning on a sequence of KG snapshots, with three core modules: modeling career trajectories, learning mutual dependency via a timestamp-aware GCN, and learning temporal dynamics via a GRNN. Across a real-world dataset, CAPER significantly outperforms baselines, two recent TKGE methods, and five state-of-the-art CTP methods in predicting future companies and positions, validating the importance of ternary relations and temporal shifts. The work provides practical impact by enabling more accurate forecast of labor-market movements and offers code accessibility, facilitating reproducibility and further research.

Abstract

The problem of career trajectory prediction (CTP) aims to predict one's future employer or job position. While several CTP methods have been developed for this problem, we posit that none of these methods (1) jointly considers the mutual ternary dependency between three key units (i.e., user, position, and company) of a career and (2) captures the characteristic shifts of key units in career over time, leading to an inaccurate understanding of the job movement patterns in the labor market. To address the above challenges, we propose a novel solution, named as CAPER, that solves the challenges via sophisticated temporal knowledge graph (TKG) modeling. It enables the utilization of a graph-structured knowledge base with rich expressiveness, effectively preserving the changes in job movement patterns. Furthermore, we devise an extrapolated career reasoning task on TKG for a realistic evaluation. The experiments on a real-world career trajectory dataset demonstrate that CAPER consistently and significantly outperforms four baselines, two recent TKG reasoning methods, and five state-of-the-art CTP methods in predicting one's future companies and positions--i.e., on average, yielding 6.80% and 34.58% more accurate predictions, respectively. The codebase of CAPER is available at https://github.com/Bigdasgit/CAPER.

Figures (5)

• Figure 1: An extrapolated career reasoning task on the input temporal knowledge graph consisting of career trajectories.
• Figure 2: Overview of CAPER, which consists of three key modules: (M1) modeling career trajectories, (M2) learning mutual dependency, and (M3) learning temporal dynamics. Given a career $(u_1,p_5,c_3,t_9)$, CAPER infers the corresponding career's likelihood using the temporal and evolution embeddings of $u_1$ and $c_3$ before $t_9$ and the embedding of $p_5$.
• Figure 3: A toy example of modeling career trajectories.
• Figure 4: An example to show the limitations of the random-sampling-based evaluation employed in existing studies.
• Figure I: Distributions of our dataset (i.e., $x$-axis indicates positions or companies in frequency order, while $y$-axis indicates the number of users in log scale).