How Generative AI is Revolutionizing Ad Recommendation Systems

Event Overview

On July 12, 2025, from 09:00 to 17:00, DataFun will host the "DataFun Summit 2025: Large Model Technology Summit" featuring a session titled "Generative AI‑Based Recommendation System".

Speaker

Ji Zhi, senior algorithm engineer at Baidu, leads information‑flow ad recall and creative direction, and has received Baidu’s highest award.

Talk Details

Title: "Generative Ad Recommendation Recall". In generative recommendation tasks, the item to be predicted is a composite (ad title, brand, multimodal information) rather than a simple token. The core challenges are representing these items and performing sequence modeling based on those representations.

The technical solution has evolved through three stages: dense representation with contrastive learning, sparse representation with sparse ID generation, and finally a sparse‑dense cascade representation combined with an integrated generation metric, each improving recall efficiency.

Outline

Core issues of generative recall tasks

Solution 1: "Dense Representation + Contrastive Learning Metric"

Solution 2: "Sparse Representation + Sparse ID Generation"

Solution 3: "Sparse‑Dense Cascade Representation + Integrated Generation Metric"

Long‑sequence challenges & optimization

Audience Takeaways

Understanding the core problems of generative recommendation recall

Insights into the thinking and exploration process

Exposure to cutting‑edge solutions

Landing Challenges & Key Solutions

Addressing ultra‑long sequence latency challenges in real‑world deployment.