A joint research work co-authored by Fuli Feng received the Best Paper Award Honorable Mention at the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2021) held online on 11 - 15 July 2021.
Fuli Feng is currently a senior research fellow at NExT Research Centre. Together with his team, comprising Yang Zhang, Xiangnan He (NExT's Alumni), Tianxin Wei, Yongdong Zhang (University of Science and Technology of China), Chonggang Song, and Guohui Ling (Tencent), received the award for their paper entitled "Causal Intervention for Leveraging Popularity Bias in Recommendation".
The research paper studies how to leverage popularity bias to improve recommendation accuracy, which is a recent problem that has received a lot of research attention. It focuses on two key aspects: how to remove the bad impact of popularity bias during training, and how to inject the desired popularity bias in the inference stage that generates top-K recommendations.
As a result, the team proposes a new training and inference paradigm for recommendation named Popularity-bias Deconfounding and Adjusting (PDA), which aims to remove the confounding popularity bias in model training and adjust the recommendation score with the desired popularity bias via causal intervention. After demonstrating this paradigm on the latent factor model and the three real-world datasets (from Kwai, Douban, and Tencent), the empirical studies validate that this model is helpful in discovering users' real interests and effective in improving recommendation performance.
References:
Research Paper: https://arxiv.org/abs/2105.06067
Source code: https://github.com/zyang1580/PDA