Hire Jinhui W. - AI Researcher

Led projects to study and identify advertising features, and established an end-to-end machine learning pipeline to optimize superior ad delivery algorithms. - Trained computer vision models to identify elements of creative ads videos, enabling the analysis of the relationship between ads video quality and user engagement as well as viewability metrics. This insight was leveraged to assist clients in enhancing ad quality, thereby boosting conversion rates. - XFN with global teams on full-funnel research for regional best practices, organized and automated machine learning tools across the organization designed to empower user behavior changes.

1. TV ratings prediction: Built and automated a step-wise regression model to improve existing TV ratings prediction capability. 2. Target audience prediction: Built an mml-LightGBM model to predict the demographics of target audience given program's information.

* Nautilus Research: Developed evaluation metrics on mml-LightGBM model and parallelized grid search on hyperparameters optimization, and produced a visualization module for feature importance and partial dependency. * NCS Audience Targeting: Forecasted sales lift for brands with LSTM model to help target consumers with time-series input. * Nielsen panelist Clustering: Clustered Nielsen panelist using k-means by incorporating geographic information to inform other Nielsen products, and validated the ideal number of clusters by investigating the Within Group Sum of Square (WSS) and silhouette score. * NCS Sales Effect: Applied Targeted Maximum Likelihood Estimator (TMLE) to reduce selection bias due to suboptimal A/B testing results and estimated campaign lift. * Watch Effect: Collaborated with stakeholders to build an automated pipeline to recommend optimal reach and frequency for ad planning with Bayesian inference on Pyro. * Total Media Resonance: Built a pipeline for Nielsen TMR project, including variable selection by using GBM, NumPyro model based on Bayesian Inference, cross-validation and preparation for post-modeling, visualization and client facing outputs. * Automation: Automated Identity Engine Evaluation, modularized the SparkSQL queries in Airflow and stored the output into data cache for future implementation, the automation is 30% more efficient.