Lixin Zhang Email & Phone Number

Beijing City, China

微信-京东算法合作项目，微信侧算法负责人合作项目包括“京东订阅号个性化推送”，微信京东“卖快场景”商品排序项目等。负责整体算法框架及方案设计，提升点击率、GMV等相关核心指标。曾获腾讯XP项目奖。"看一看" 早期“看一看”（基于微信场景的Feeds流个性化推荐项目）主要参与者。负责推荐系统数据流pipeline，参与推荐架构重构，以及CTR预估工作。People RankWe build up a pagerank-like algorithm called People Rank in Wechat to find users who have more spread impact according to their forward relationship. And we proposed a solution to measure the quality of document according to its reading and forwarding audience's PR which helps improve the quality of wechat search. SNS Feed Interaction Rate Prediction in WechatGiven a user, a sns feed and this feed's publisher, we want to predict the probability the user take click, comment or like action to the feed. L1-norm Logistic Regression, GBDT and EdgeRank algorithm are involved in this project, and technologies like hadoop, hive, pig are used for data processing and feature engineering.

Beijing City, China

Real-time Data Analytic System powered by Druid.Setup a distributed cluster with Druid(opensource project) for large scale and Real-time data analytic which involves Storm, Kafka, Hadoop, Zookeeper technique framework. I tried to implement the Druid-based D&A tool(large-scale data statistics) and drafted a Druid performance(vs vespa) report, also took part in the refactoring of Druid-based MoneyWatcher(near real-time metrics statistics).Advertisers’ Budget System, data analysis and testing.Data analysis for advertisers’s budget allocation and pacing strategy, and some testing tasks on Curveball Budget.

Beijing City, China

Click Through Rate Prediction and Bidding Optimization.Participate in feature engineering for CTR Prediction with MLR model in a DSP view. And we proposed an improved bidding model considering more factors which outperforms old one.User Interest Points Recommendation for Audience Targeting.We treat audience targeting method based on interest points as a recommendation problem. And we adopt distributed collaborative filtering method to call back more ads and optimize ad click probability.Item Categories Auto Mapping across Different Domains.Due to item categories between DSP and Ad-Exchange are in different domains, we proposed a creative solution to build the relationship between them which results in less manual effort.

Beijing City, China

Work on Cloud Computing (PaaS) Development.Implement a PaaS server cluster management tool based on script tasks.

Master'S Degree, Computer Science

Bachelor'S Degree, Computer Software Engineering