Career Path Modeling and Recommendations with Linkedin Career Data and Predicted Salary Estimations

Micaela Tayoto Cerilla, Aaron Santillan, Carl John Vinas, Michael B. Dela Fuente

Keywords: Career Path, Attention Layers, LSTM Layers, Recommendation Systems, Data Mining

TL;DR: Our research is about improving existing career path recommendation models in terms of accuracy by adding more variables/features namely, Social Networks and Predicted salaries per job, and using a new technique utilizing Attention Layers.

Abstract: Career planning involves devising a sequence of steps that build up an ideal career path for a person. However, career planning has become more complex in recent years, demanding the need for better models and systems for recommending Career Paths. With that in mind, we explored new variables and techniques that could help in predicting better career paths. We built a Long-Short Term Memory Network with Self Attention Layers called LSTM-ATT, that predicts a person’s possible career path using Linkedin Career History and new variables such as salary estimations and social networks. We measured the model’s performance in terms of Mean Percentile Rank and Precision at 50 and 100. We found that LSTM and self-attention layers were able to show good predictive performance for multi-class classification even with over 6000 classes for companies and skills, effectively beating a multi-channel CNN for all metrics. However, by checking the versions of either model with added features, they did not yield any major increase in predictive accuracy against the models without it. This leads us to conclude that the added variables did not help in predicting better career paths.