张策

Developed and deployed push personalized sending timing, UV click rate, and close rate models from scratch, cumulatively increasing Daily Active Users (DAU) by +0.7% and reducing closing rates by -5%.

Designed and implemented a causal DNN model and order-preserving planning model, improving DAU by +1.8% through personalized content selection and constrained resource allocation.

Engineered a multi-treatment S-learner learner and optimized features and loss functions, resulting in significant model performance improvements for push causal DNN systems.

Refined network models by integrating DIN and sim structures and employing ple network multi-objective modeling, enhancing user interest capture and optimizing multi-objective fusion sorting formulas.

Optimized search keyword recall by increasing users' recent search channels and capturing short-term interest points, leading to improved recall effectiveness and user satisfaction.

Contributed as a core project team member to achieve the 100 million DAU target for Little Red Book.

Optimized the Hand Hundred PUSH sorting module by integrating DCN-M structure and advanced feature engineering, leading to a 3% increase in revenue.

Enhanced rerank algorithms with high-propagation short video interest signals and user search query signals, improving short video distribution and click-through rate by +10.56%.

Led new user optimization initiatives for Hand 100 PUSH, significantly improving next-day first call rate and seven-day attribution rate.

Developed and implemented new user local recall, query recall, and registered interest point recall systems, strengthening the platform's ability to capture new user interests.

Improved model perception and recommendation capabilities by integrating new user signals, sub-bucket ID features, cross-features, and weighted new user sample loss.

Contributed to completing the 0.6 billion MAU target for Hand 100.