Improve job experiences prediction with attention mechanism

Abstract

Employee in the information technology domain is extremely active in their career. In that case, the company usually deals with human resource exhaustion. The understanding of their job mobility is useful for the company in a variety of ways, for example, employee recruitment, employee retention. While most studies focus on predicting the next job title or job recommendation based on previous experiences, the problem of forecast duration of employees working on the company at the individual level receives little attention. Moreover, previous methods treat the information from different experiences as similarly important so they cannot utilize the potential connection between experiences. To solve the above problems, we contribute a new model with attention mechanism. In particular, the attention mechanisms give more understanding to learned representations and give better results. We also predict next job title duration. Different from previous works, our model can effectively utilize the previous employee experiences and flexibly adapts to the information of different importance. Our methods are applied for 10.000 real-world employee profiles and show significant results that outperform the strong baseline model and other state-of-the-art methods.