Melbourne, Australia
This presentation has the aim of presenting a Content-Based (combination of Retrieval Phase; + Learning2Rank algorithm) solution made in Julia to improve the quality of Job Recommendations on the SEEK Asia homepage. As a topic to make accessible to any audience, I will present all components in high level and focus in how Julia helped to make it possible to process all features and natural language models necessary to generate high-quality recommendations using Genie API. The solution was deployed in AWS for production using Docker and Ansible. Key Points of Julia in the Presentation: • Speed-up the process in more than x10 • Make easy the natural language preprocessing • Give total control how to manipulate arrays and matrix for efficiency • Turns flexible the feature engineering calculation without adding much complexity by the language • It is capable to be used as API in a real system environment that need to scale to thousands of users • Last version v0.6.x is a remarkable achievement and stable to make the code go to production • Present performance improvements showing Sparse Matrix, @views, @. and special use of @inbounds, @simd, and @fastmath. • Quick highlights on the packages Genie, DataFrames, and XGBoost
I`m a Lead Data Scientist specialized in Recommender Systems, Machine Learning, and Natural Language Processing.
