Research and Product Applications of Causal Inference for Solving Recommendation System Bias

Dai Quanyu , senior researcher at Huawei Noah's Ark Lab, holds a Ph.D. from Hong Kong Polytechnic University and focuses on recommendation systems, causal inference, and graph representation learning. He has published in top venues such as KDD, WWW, SIGIR, AAAI, TKDE, and TNNLS.

Talk Title: Research and Product Applications of Causal Inference for Solving Recommendation System Bias

The presentation explains that recommendation systems form a closed‑loop feedback mechanism that suffers from multiple biases (user selection, exposure, video length, etc.). Using the potential outcomes framework, the talk analyzes the root causes of these biases and proposes targeted causal‑based correction algorithms, including inverse propensity weighting, double robust learning, and multi‑robust learning. It also shares experiences of deploying these methods in industrial products, with research results published at conferences such as KDD, AAAI, and IJCAI.

Audience Benefits:

Understand the fundamental causes of recommendation bias from a causal inference perspective.

Learn practical bias‑mitigation techniques such as inverse propensity scoring, double robust, and multi‑robust learning.

Gain insight into real‑world product applications of causal bias‑correction algorithms.