Tian Wei Email and Phone Number

Azure Synapse ML Engineer

• Developed a deep attention-based neural network in an efficient way to model the user’s interest on candidate ads, given the history of user clicked ads.• Implemented an Uplift model to calculate the lift of user’s revenue, given a particular treatment such as the delivery of coupons, and to give each individual a heterogenous treatment based on the predicted effect.

• Developed an end-to-end news commenting system for online social communities to improve their user experience. Partners include Netease News, Netease Music, SMZDM.com, etc.• Proposed a knowledge-graph enhanced comment generation model based on CVAE-Transformer and GNN.• Proposed a BERT-based query-response matching model whose universal sentence encoder can be applied to a real-time information retrieval system.• Proposed a personalized query-response ranking model which can output a personalized response based on a given user profile.• Contributed to developing a knowledge-grounded dialogue generation model for the WeChat official accounts custom service platform, based on a pre-trained UNILM model and a triple-gated CopyNet.• Contributed to developing a template-based sporting news generation system for 2018 World Cup

• Developed an advertising title generation system using a vanilla attention-based LSTM seq2seq model.• Developed a bidding query recommendation system using classical collaborative filtering algorithm for L1 searching and a hybrid deep structured semantic model for L2 ranking.• Proposed a hybrid LSTM-DSSM query matching model using a semi-supervised manner.• Proposed a query2vec approach using advertisers’ bidding behavior, motivated by the Skip-gram and Glove model.• Integrated a simplified bidding word management system with Duer OS, which was also shortlisted for the 17th Hackathon Dark Horse award of Baidu.