Bert (Mu-Chia) Lee Email & Phone Number

New York City, Us

Burbank, Ca, Us

☑️Established "Unified User Segmentation" for all AVOD and SVOD users (~ 300 million) across all teams in the company, including Marketing, Partnerships, and Growth teams. This initiative aimed to facilitate business communication and improve user targeting in campaigns. Key achievements:①Created "Content Taste Clusters" using Matrix Factorization (ALS), K-Means, and Decision Trees to gain insights into users' content preferences.②Built a pipeline to automate user segmentation updates on Hive table using AWS Pipeline and EMR clusters.☑️Developed "Content ROI Measurement" with the Content Insights team to enhance content performance evaluation and support content marketing and finance budgeting decisions. Key contributions:①Analyzed the relationship between users' retention and engagement with content.②Built a user retention attribution model to associate users' retention revenue with content, measuring the return on content investment. The model's RMSE increased by 45.0%.☑️Created the Short-Form Feed feature by leveraging user-watching behavior in long videos to automatically generate fascinating clips. Developed a pipeline to generate Content Heatmap from user-watching signals, helping to identify the attractive parts of long content and appear on the seekbar. The initial demo in the Disney+ Hackathon achieved 5th place, and the project is scheduled to launch in July 2023.☑️ Built a Taste Prediction Model to forecast potential new content tastes, aiming to predict their performance and improve content supply. Implemented a Deep Neural Network (DNN) Model with PyTorch, achieving a top-3 precision of 0.89 and a cosine loss of 0.12. Conducted feature importance analysis with SHAP.☑️ Explored multiple methods to handle the challenge of taste cluster changes over time by leveraging concepts from K-Means, Hierarchical Clustering, and ISODATA. Developed a new clustering algorithm that provides the optimal balance between cluster stability and sensitivity.

Singapore, Sg

☑️ Built "RM-Customer Reshuffle Model" for over 40K treasury customers of all 34 branches in Taiwan, leading to a 25.9% growth of daily investment revenue in the first month of deployment. Implemented a recommendation system integrating content-based filtering, propensity scores, and decision trees to optimize RM-Customer pairings, aligning RM rewards with sales performance.☑️ Built Bond Tracking Dashboard for Investment PMs to monitor selling performances of branches, RMs and Products using Power BI.☑️ Automated reporting process (including reports of Credit Card, Investment Products & Loans to Taiwan GM) with Python(openpyxl, xlwings, win32com) to reduce 1 hour of manual work daily.☑️ Fulfilled 20+ Credit Card Reward Campaigns (including Pchome, Shopee, Momo...).☑️ Aligned with PMs to analyze and backtest the new customer risk scoring model.

Taipei, Northern Taiwan, Tw

Multimedia Information Retrieval Lab - FinTech Group (Advisor: Jyh-Shing Jang)☑️ Implemented Google's neural network language model - BERT - in a 2-stage stock prediction model using Chinese financial news articles, resulting in a 65.0% directional accuracy in 4,421 sentences.☑️ Analyzing the correlation between optimistic rates, key information of daily news and stock movements.

New Taipei City, Taipei, Tw

Improved the NLU system of smart speaker "OLAMI".☑️ Improved the intention detection performance of smart speaker "OLAMI" by implementing an efficient text classification model using TF-IDF to calculate sentence similarities.☑️ Enhanced the completeness of language corpus by web data crawling using Java's Seimicrawler.☑️ Built a skip-gram word embedding model with text data extracted from novels and movie subtitles.☑️ Implemented a negation detection model setting "dependency relationship" and "part of speech" as feature functions in CRF model. Achieved a 19.6% improvement in detection accuracy.

Taipei, Northern Taiwan, Tw

Research Assistant of Prof. Chung-Ying Lee at Dept. of Economics☑️ Created tables summarizing the data of 25,112 medical organizations in Taiwan using Python's libraries such as numpy, pandas.☑️ Drew maps to visualize the regional competitiveness of medical organizations in Taiwan with QGIS.

☑️ Wrote an article about overrating/underrating players' defensive impact in the game.☑️ Wrote an article analyzing the causal relationship between joining National Teams during the off-season and the performance of players in the next season.☑️ Translated English articles into Chinese.

Master Of Science - Ms, Statistics And Data Science

Bachelor Of Science - Bs, Engineering Science