Experience

JDT

• China
• IT Services and IT Consulting
• 300 - 400 Employee

• senior manger

  • Jun 2016 - Present

  JD is a fortune 500 company. it is famous in China and already ipo in America. I lead a team with data scientists to solve the user growth and risk about the JDT core product Baitiao/Jintiao. My work focused on the four parts. The first part is digital marketing and growth hacking. Machine learning/deep learning methods were used to promote core products' GMV/balance, new user registration number, or other important indicators. The data such as browsing records, order records, and loan… Show more JD is a fortune 500 company. it is famous in China and already ipo in America. I lead a team with data scientists to solve the user growth and risk about the JDT core product Baitiao/Jintiao. My work focused on the four parts. The first part is digital marketing and growth hacking. Machine learning/deep learning methods were used to promote core products' GMV/balance, new user registration number, or other important indicators. The data such as browsing records, order records, and loan records were used to build consumers' portrait, ltv model, conversion rate model, etc. The second part is about loan risk control. application score card, behavior score card and collection score card were built to control risk. Different from the traditional risk model, we built an “action-model” system to offer the users customized loan limits to optimize product profit in the pre-loan phase. And in the post-loan phase, an intelligent collection system was built to analyze user's repayment probability and allocate the optimized sources such as phone call numbers and matched employees. The third part of my job is anti-fraud. Fraud has touched nearly every area of business. My job is to use machine learning to fight against fraud. The main fields are payment fraud, account fraud, and loan fraud. We built risk awareness system to monitor the whole user life cycle. Because of the lack of fraud label, the system used many unsurprised technologies like an autoencoder, iforrest, lpa, etc. In the detail, the system monitors the risk of user important actions like registration, login, browsing, payment, delivering, and so on. And then it will protect user privity immediately and recall more unknown threats. Finally, we build a huge network consisting of 13 million nodes and 10 million edges. Through the network, we can judge users’ risk not only by they own feature but also consider they friends’ risk situation and detect fraud communities. Show less JD is a fortune 500 company. it is famous in China and already ipo in America. I lead a team with data scientists to solve the user growth and risk about the JDT core product Baitiao/Jintiao. My work focused on the four parts. The first part is digital marketing and growth hacking. Machine learning/deep learning methods were used to promote core products' GMV/balance, new user registration number, or other important indicators. The data such as browsing records, order records, and loan… Show more JD is a fortune 500 company. it is famous in China and already ipo in America. I lead a team with data scientists to solve the user growth and risk about the JDT core product Baitiao/Jintiao. My work focused on the four parts. The first part is digital marketing and growth hacking. Machine learning/deep learning methods were used to promote core products' GMV/balance, new user registration number, or other important indicators. The data such as browsing records, order records, and loan records were used to build consumers' portrait, ltv model, conversion rate model, etc. The second part is about loan risk control. application score card, behavior score card and collection score card were built to control risk. Different from the traditional risk model, we built an “action-model” system to offer the users customized loan limits to optimize product profit in the pre-loan phase. And in the post-loan phase, an intelligent collection system was built to analyze user's repayment probability and allocate the optimized sources such as phone call numbers and matched employees. The third part of my job is anti-fraud. Fraud has touched nearly every area of business. My job is to use machine learning to fight against fraud. The main fields are payment fraud, account fraud, and loan fraud. We built risk awareness system to monitor the whole user life cycle. Because of the lack of fraud label, the system used many unsurprised technologies like an autoencoder, iforrest, lpa, etc. In the detail, the system monitors the risk of user important actions like registration, login, browsing, payment, delivering, and so on. And then it will protect user privity immediately and recall more unknown threats. Finally, we build a huge network consisting of 13 million nodes and 10 million edges. Through the network, we can judge users’ risk not only by they own feature but also consider they friends’ risk situation and detect fraud communities. Show less



百分点-Baifendian

• China
• Software Development
• 1 - 100 Employee

• 业务数据分析师

  • Oct 2014 - May 2016

  主要算法分析工作涉及分析、清洗和推荐三方面： 数据分析，主要包括 电商品类数据分析、行业分析调研工作 数据清洗，主要包括 用户浏览行为数据、设备采集数据的清洗加工和特征工程工作 推荐系统，主要包括 多路召回策略和一些精排序算法实现工作 主要算法分析工作涉及分析、清洗和推荐三方面： 数据分析，主要包括 电商品类数据分析、行业分析调研工作 数据清洗，主要包括 用户浏览行为数据、设备采集数据的清洗加工和特征工程工作 推荐系统，主要包括 多路召回策略和一些精排序算法实现工作