Insights and Lessons from the First Tencent Social Advertising University Algorithm Competition

The first Tencent Social Advertising University Algorithm Competition focused on mobile app conversion‑rate prediction, offering real‑world advertising data for students to develop advanced algorithms.

Beijing University’s team "到底对不队" achieved fourth place overall and won the best‑presentation award.

1. Deep Business‑Logic Analysis to Identify Pain Points – The team spent extensive time understanding the complex conversion funnel (impression‑click‑download‑re‑open) to pinpoint critical features for engineering.

2. Single‑Model Innovation and Optimization – They evaluated classic high‑performing models, studied related papers, and enhanced DeepFM to maximize its effectiveness.

3. Multi‑Model Fusion for Complementary Strengths – By combining diverse models in a staged fusion process and applying result correction, they achieved significant performance gains.

4. Strong Team Collaboration – Clear role division, regular progress communication, and task‑management tools were essential, as the competition required both coding skill and strategic coordination.

5. Learning from Existing Research and Peers – The team continuously consulted papers, sought advice from experienced peers, and exchanged ideas with other contestants, treating competitors as allies.

The authors also reflect on pitfalls: insufficient big‑data processing experience caused a drop in rankings after data changes, and inadequate feature extraction for the new dataset delayed model tuning. They emphasize early preparation for data handling and feature engineering.

In the final on‑site defense, the team’s thorough summary of successes and failures earned them the best‑presentation award and valuable feedback from industry experts, highlighting the practical benefits of participating in such competitions.