Stephen Lee Email & Phone Number

Hangzhou, Zhejiang, China

1) Responsible for optimizing the recommendation system for AliExpress Russian site.2) Achieved a 30% increase in CTR in the Recall Stage through embedding-based algorithms such as GraphSAGE.

Shanghai, China

1) Achieved a 10% to 200% CTR increase and a 10% to 70% CVR increase on client projects using deep learning algorithms and knowledge distillation technique.2) Built a machine learning platform from scratch, increasing the success rate of the company's B2B bid wins.3) Developed AutoML for the machine learning platform, including automated feature engineering, feature selection, model training, model deployment and hot updates.4) Developed over 20 recommendation models for the machine learning platform, covering the recall, ranking, and re-ranking stages.

Beijing, China

1) Achieved a 100% CTR increase in the Ranking Stage of e-commerce product recommendations on the JD Finance app with deep learning algorithms.2) Achieved a 10% increase in CVR (conversion rate) in the Ranking Stage using multi-objective algorithms such as ESMM and MMOE.3) Extracted textual features including product titles and descriptions using BERT series models such as RoBERTa.4) Extracted image features using image classification algorithms such as ResNet and object detection algorithms like YOLOv3.5) Responsible for the development of the online workflow for feature engineering and model inference, resulting in a 30% improvement in online performance.

Beijing, China

1) Achieved a 20% CTR (click-through rate) increase in the Ranking Stage on the 'Right-Side Recommendations' of 360 Search through comprehensive feature engineering and deep learning algorithms such as DeepFM, DIN and Dien.2) Improved CTR on "Related Recommendations" and "News Recommendations" in the Ranking Stage of 360 Search through machine learning algorithms such as Logistic Regression.3) Built a knowledge graph using data from the search engine.4) Filtered semantically irrelevant POIs (points of interest) using Word2Vec and a synonym dictionary on 360 Maps to enhance search relevance as the POIs were sourced from web crawlers and user-generated content.

Beijing, China

1) Achieved 90% Accuracy in extracting valuable information from the financial reports of publicly traded companies using deep learning models such as TextCNN.2) Intercepted user chat logs on the WeChat app for Mac using reverse engineering and hooking techniques to better understand customer needs.