Live Deep Dive into Two Award‑Winning WSDM 2025 Papers on Popularity Bias in Recommendation Models and Graph‑Based Causal Inference

The article announces a live broadcast that will deeply analyze two frontier research papers selected for the WSDM 2025 conference (International Conference on Web Search and Data Mining) held in San Francisco, USA, from May 15‑19, 2025.

Paper 1: “How Do Recommendation Models Amplify Popularity Bias? An Analysis from the Spectral Perspective.” The best‑paper award work, authored by a team from Zhejiang University, University of Science and Technology of China, and Ant Group, investigates the long‑standing popularity bias where popular items are over‑recommended while long‑tail content is ignored. It uniquely employs matrix spectral analysis to expose the low‑rank approximation mechanism that implicitly magnifies bias, and introduces a lightweight regularization method (ReSN) that balances effectiveness and computational efficiency.

Paper 2: “Graph Disentangle Causal Model: Enhancing Causal Inference in Networked Observational Data.” This oral presentation proposes a novel causal inference paradigm that tackles confounding variables in networked data (e.g., social graphs, user‑behavior graphs). By tightly coupling graph neural networks with structural causal models, the Graph Disentangle Causal Model (GDCM) achieves disentanglement of graph structures and dynamic sparsification, addressing a major industry challenge.

The live session will feature the authors sharing design ideas, validation processes, and technical breakthroughs of both papers. The broadcast will be streamed simultaneously on WeChat Video Channels (Ant Tech Research Institute), Ant Tech’s Bilibili channel, and other Ant Group platforms.

Live Guide: Date & Time: May 15, 2025, 18:00‑20:00 (UTC+8) Platforms: WeChat Video, Ant Tech Bilibili channel, and Ant Group’s official streams. Audience members are encouraged to reserve a spot and follow the channels for the live event.